From aea3671e634732f949e6defa571b7656e9817a90 Mon Sep 17 00:00:00 2001 From: Joseph Pollack Date: Tue, 2 Dec 2025 02:59:39 +0100 Subject: [PATCH] fix typo mkdocs yaml --- mkdocs.yml | 2 - pyproject.toml | 2 + site/404.html | 2 +- site/api/agents/index.html | 53 +- site/api/models/index.html | 2 +- site/api/orchestrators/index.html | 36 +- site/api/services/index.html | 79 +-- site/api/tools/index.html | 95 ++-- site/architecture/agents/index.html | 2 +- .../graph-orchestration/index.html | 9 - .../graph_orchestration/index.html | 22 +- site/architecture/middleware/index.html | 60 ++- site/architecture/orchestrators/index.html | 2 +- site/architecture/services/index.html | 59 +- site/architecture/tools/index.html | 37 +- .../architecture/workflow-diagrams/index.html | 47 +- site/architecture/workflows/index.html | 503 ------------------ site/configuration/CONFIGURATION/index.html | 130 ----- site/configuration/index.html | 248 ++++----- site/contributing/code-quality/index.html | 13 +- site/contributing/code-style/index.html | 26 +- site/contributing/error-handling/index.html | 18 +- .../implementation-patterns/index.html | 20 +- site/contributing/index.html | 67 ++- .../prompt-engineering/index.html | 2 +- site/contributing/testing/index.html | 50 +- site/getting-started/examples/index.html | 98 ++-- site/getting-started/installation/index.html | 48 +- .../mcp-integration/index.html | 70 +-- site/getting-started/quick-start/index.html | 50 +- site/index.html | 25 +- site/license/index.html | 2 +- site/overview/architecture/index.html | 2 +- site/overview/features/index.html | 2 +- site/overview/quick-start/index.html | 37 +- site/search/search_index.json | 2 +- site/sitemap.xml | 78 ++- site/team/index.html | 2 +- uv.lock | 52 ++ 39 files changed, 775 insertions(+), 1279 deletions(-) delete mode 100644 site/architecture/graph-orchestration/index.html delete mode 100644 site/architecture/workflows/index.html delete mode 100644 site/configuration/CONFIGURATION/index.html diff --git a/mkdocs.yml b/mkdocs.yml index 35d01f0..1d3d31b 100644 --- a/mkdocs.yml +++ b/mkdocs.yml @@ -55,8 +55,6 @@ plugins: lang: - en separator: '[\s\-,:!=\[\]()"`/]+|\.(?!\d)|&[lg]t;|&' - prebuild_index: true # Faster search initialization - indexing: full # Full-text indexing for better results - mermaid2 - codeinclude - git-revision-date-localized: diff --git a/pyproject.toml b/pyproject.toml index 3080234..6bef4f1 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -70,6 +70,7 @@ dev = [ "mkdocs-material>=9.0.0", "mkdocs-mermaid2-plugin>=1.1.0", "mkdocs-codeinclude-plugin>=0.2.0", + "mkdocs-git-revision-date-localized-plugin>=1.2.0", "mkdocs-minify-plugin>=0.8.0", "pymdown-extensions>=10.17.2", ] @@ -192,6 +193,7 @@ dev = [ "mkdocs-codeinclude-plugin>=0.2.1", "mkdocs-material>=9.7.0", "mkdocs-mermaid2-plugin>=1.2.3", + "mkdocs-git-revision-date-localized-plugin>=1.2.0", "mkdocs-minify-plugin>=0.8.0", "structlog>=25.5.0", "ty>=0.0.1a28", diff --git a/site/404.html b/site/404.html index 7e9ca86..e27584c 100644 --- a/site/404.html +++ b/site/404.html @@ -1 +1 @@ - The DETERMINATOR

404 - Not found

\ No newline at end of file + The DETERMINATOR

404 - Not found

\ No newline at end of file diff --git a/site/api/agents/index.html b/site/api/agents/index.html index 1001855..5f6acf4 100644 --- a/site/api/agents/index.html +++ b/site/api/agents/index.html @@ -1,52 +1 @@ - Agents API Reference - The DETERMINATOR
Skip to content

Agents API Reference

This page documents the API for DeepCritical agents.

KnowledgeGapAgent

Module: src.agents.knowledge_gap

Purpose: Evaluates research state and identifies knowledge gaps.

Methods

evaluate

async def evaluate(
-    self,
-    query: str,
-    background_context: str,
-    conversation_history: Conversation,
-    iteration: int,
-    time_elapsed_minutes: float,
-    max_time_minutes: float
-) -> KnowledgeGapOutput
-

Evaluates research completeness and identifies outstanding knowledge gaps.

Parameters: - query: Research query string - background_context: Background context for the query - conversation_history: Conversation history with previous iterations - iteration: Current iteration number - time_elapsed_minutes: Elapsed time in minutes - max_time_minutes: Maximum time limit in minutes

Returns: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

ToolSelectorAgent

Module: src.agents.tool_selector

Purpose: Selects appropriate tools for addressing knowledge gaps.

Methods

select_tools

async def select_tools(
-    self,
-    query: str,
-    knowledge_gaps: list[str],
-    available_tools: list[str]
-) -> AgentSelectionPlan
-

Selects tools for addressing knowledge gaps.

Parameters: - query: Research query string - knowledge_gaps: List of knowledge gaps to address - available_tools: List of available tool names

Returns: AgentSelectionPlan with list of AgentTask objects.

WriterAgent

Module: src.agents.writer

Purpose: Generates final reports from research findings.

Methods

write_report

async def write_report(
-    self,
-    query: str,
-    findings: str,
-    output_length: str = "medium",
-    output_instructions: str | None = None
-) -> str
-

Generates a markdown report from research findings.

Parameters: - query: Research query string - findings: Research findings to include in report - output_length: Desired output length ("short", "medium", "long") - output_instructions: Additional instructions for report generation

Returns: Markdown string with numbered citations.

LongWriterAgent

Module: src.agents.long_writer

Purpose: Long-form report generation with section-by-section writing.

Methods

write_next_section

async def write_next_section(
-    self,
-    query: str,
-    draft: ReportDraft,
-    section_title: str,
-    section_content: str
-) -> LongWriterOutput
-

Writes the next section of a long-form report.

Parameters: - query: Research query string - draft: Current report draft - section_title: Title of the section to write - section_content: Content/guidance for the section

Returns: LongWriterOutput with updated draft.

write_report

async def write_report(
-    self,
-    query: str,
-    report_title: str,
-    report_draft: ReportDraft
-) -> str
-

Generates final report from draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Complete report draft

Returns: Final markdown report string.

ProofreaderAgent

Module: src.agents.proofreader

Purpose: Proofreads and polishes report drafts.

Methods

proofread

async def proofread(
-    self,
-    query: str,
-    report_title: str,
-    report_draft: ReportDraft
-) -> str
-

Proofreads and polishes a report draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Report draft to proofread

Returns: Polished markdown string.

ThinkingAgent

Module: src.agents.thinking

Purpose: Generates observations from conversation history.

Methods

generate_observations

async def generate_observations(
-    self,
-    query: str,
-    background_context: str,
-    conversation_history: Conversation
-) -> str
-

Generates observations from conversation history.

Parameters: - query: Research query string - background_context: Background context - conversation_history: Conversation history

Returns: Observation string.

InputParserAgent

Module: src.agents.input_parser

Purpose: Parses and improves user queries, detects research mode.

Methods

parse_query

async def parse_query(
-    self,
-    query: str
-) -> ParsedQuery
-

Parses and improves a user query.

Parameters: - query: Original query string

Returns: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: "iterative" or "deep" - key_entities: List of key entities - research_questions: List of research questions

Factory Functions

All agents have factory functions in src.agent_factory.agents:

Parameters: - model: Optional Pydantic AI model. If None, uses get_model() from settings.

Returns: Agent instance.

See Also

\ No newline at end of file + Agents API Reference - The DETERMINATOR
Skip to content

Agents API Reference

This page documents the API for DeepCritical agents.

KnowledgeGapAgent

Module: src.agents.knowledge_gap

Purpose: Evaluates research state and identifies knowledge gaps.

Methods

evaluate

Evaluates research completeness and identifies outstanding knowledge gaps.

Parameters: - query: Research query string - background_context: Background context for the query (default: "") - conversation_history: History of actions, findings, and thoughts as string (default: "") - iteration: Current iteration number (default: 0) - time_elapsed_minutes: Elapsed time in minutes (default: 0.0) - max_time_minutes: Maximum time limit in minutes (default: 10)

Returns: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

ToolSelectorAgent

Module: src.agents.tool_selector

Purpose: Selects appropriate tools for addressing knowledge gaps.

Methods

select_tools

Selects tools for addressing a knowledge gap.

Parameters: - gap: The knowledge gap to address - query: Research query string - background_context: Optional background context (default: "") - conversation_history: History of actions, findings, and thoughts as string (default: "")

Returns: AgentSelectionPlan with list of AgentTask objects.

WriterAgent

Module: src.agents.writer

Purpose: Generates final reports from research findings.

Methods

write_report

Generates a markdown report from research findings.

Parameters: - query: Research query string - findings: Research findings to include in report - output_length: Optional description of desired output length (default: "") - output_instructions: Optional additional instructions for report generation (default: "")

Returns: Markdown string with numbered citations.

LongWriterAgent

Module: src.agents.long_writer

Purpose: Long-form report generation with section-by-section writing.

Methods

write_next_section

Writes the next section of a long-form report.

Parameters: - original_query: The original research query - report_draft: Current report draft as string (all sections written so far) - next_section_title: Title of the section to write - next_section_draft: Draft content for the next section

Returns: LongWriterOutput with formatted section and references.

write_report

Generates final report from draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Complete report draft

Returns: Final markdown report string.

ProofreaderAgent

Module: src.agents.proofreader

Purpose: Proofreads and polishes report drafts.

Methods

proofread

Proofreads and polishes a report draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Report draft to proofread

Returns: Polished markdown string.

ThinkingAgent

Module: src.agents.thinking

Purpose: Generates observations from conversation history.

Methods

generate_observations

Generates observations from conversation history.

Parameters: - query: Research query string - background_context: Optional background context (default: "") - conversation_history: History of actions, findings, and thoughts as string (default: "") - iteration: Current iteration number (default: 1)

Returns: Observation string.

InputParserAgent

Module: src.agents.input_parser

Purpose: Parses and improves user queries, detects research mode.

Methods

parse

Parses and improves a user query.

Parameters: - query: Original query string

Returns: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: "iterative" or "deep" - key_entities: List of key entities - research_questions: List of research questions

Factory Functions

All agents have factory functions in src.agent_factory.agents:

Parameters: - model: Optional Pydantic AI model. If None, uses get_model() from settings. - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars)

Returns: Agent instance.

See Also

\ No newline at end of file diff --git a/site/api/models/index.html b/site/api/models/index.html index 51613ec..863fc63 100644 --- a/site/api/models/index.html +++ b/site/api/models/index.html @@ -1 +1 @@ - Models API Reference - The DETERMINATOR
Skip to content

Models API Reference

This page documents the Pydantic models used throughout DeepCritical.

Evidence

Module: src.utils.models

Purpose: Represents evidence from search results.

Fields: - citation: Citation information (title, URL, date, authors) - content: Evidence text content - relevance_score: Relevance score (0.0-1.0) - metadata: Additional metadata dictionary

Citation

Module: src.utils.models

Purpose: Citation information for evidence.

Fields: - title: Article/trial title - url: Source URL - date: Publication date (optional) - authors: List of authors (optional)

KnowledgeGapOutput

Module: src.utils.models

Purpose: Output from knowledge gap evaluation.

Fields: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

AgentSelectionPlan

Module: src.utils.models

Purpose: Plan for tool/agent selection.

Fields: - tasks: List of agent tasks to execute

AgentTask

Module: src.utils.models

Purpose: Individual agent task.

Fields: - agent_name: Name of agent to use - query: Task query - context: Additional context dictionary

ReportDraft

Module: src.utils.models

Purpose: Draft structure for long-form reports.

Fields: - title: Report title - sections: List of report sections - references: List of citations

ReportSection

Module: src.utils.models

Purpose: Individual section in a report draft.

Fields: - title: Section title - content: Section content - order: Section order number

ParsedQuery

Module: src.utils.models

Purpose: Parsed and improved query.

Fields: - original_query: Original query string - improved_query: Refined query string - research_mode: Research mode ("iterative" or "deep") - key_entities: List of key entities - research_questions: List of research questions

Conversation

Module: src.utils.models

Purpose: Conversation history with iterations.

Fields: - iterations: List of iteration data

IterationData

Module: src.utils.models

Purpose: Data for a single iteration.

Fields: - iteration: Iteration number - observations: Generated observations - knowledge_gaps: Identified knowledge gaps - tool_calls: Tool calls made - findings: Findings from tools - thoughts: Agent thoughts

AgentEvent

Module: src.utils.models

Purpose: Event emitted during research execution.

Fields: - type: Event type (e.g., "started", "search_complete", "complete") - iteration: Iteration number (optional) - data: Event data dictionary

BudgetStatus

Module: src.utils.models

Purpose: Current budget status.

Fields: - tokens_used: Tokens used so far - tokens_limit: Token limit - time_elapsed_seconds: Elapsed time in seconds - time_limit_seconds: Time limit in seconds - iterations: Current iteration count - iterations_limit: Iteration limit

See Also

\ No newline at end of file + Models API Reference - The DETERMINATOR
Skip to content

Models API Reference

This page documents the Pydantic models used throughout DeepCritical.

Evidence

Module: src.utils.models

Purpose: Represents evidence from search results.

Fields: - citation: Citation information (title, URL, date, authors) - content: Evidence text content - relevance: Relevance score (0.0-1.0) - metadata: Additional metadata dictionary

Citation

Module: src.utils.models

Purpose: Citation information for evidence.

Fields: - source: Source name (e.g., "pubmed", "clinicaltrials", "europepmc", "web", "rag") - title: Article/trial title - url: Source URL - date: Publication date (YYYY-MM-DD or "Unknown") - authors: List of authors (optional)

KnowledgeGapOutput

Module: src.utils.models

Purpose: Output from knowledge gap evaluation.

Fields: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

AgentSelectionPlan

Module: src.utils.models

Purpose: Plan for tool/agent selection.

Fields: - tasks: List of agent tasks to execute

AgentTask

Module: src.utils.models

Purpose: Individual agent task.

Fields: - gap: The knowledge gap being addressed (optional) - agent: Name of agent to use - query: The specific query for the agent - entity_website: The website of the entity being researched, if known (optional)

ReportDraft

Module: src.utils.models

Purpose: Draft structure for long-form reports.

Fields: - sections: List of report sections

ReportSection

Module: src.utils.models

Purpose: Individual section in a report draft.

Fields: - section_title: The title of the section - section_content: The content of the section

ParsedQuery

Module: src.utils.models

Purpose: Parsed and improved query.

Fields: - original_query: Original query string - improved_query: Refined query string - research_mode: Research mode ("iterative" or "deep") - key_entities: List of key entities - research_questions: List of research questions

Conversation

Module: src.utils.models

Purpose: Conversation history with iterations.

Fields: - history: List of iteration data

IterationData

Module: src.utils.models

Purpose: Data for a single iteration.

Fields: - gap: The gap addressed in the iteration - tool_calls: The tool calls made - findings: The findings collected from tool calls - thought: The thinking done to reflect on the success of the iteration and next steps

AgentEvent

Module: src.utils.models

Purpose: Event emitted during research execution.

Fields: - type: Event type (e.g., "started", "search_complete", "complete") - iteration: Iteration number (optional) - data: Event data dictionary

BudgetStatus

Module: src.utils.models

Purpose: Current budget status.

Fields: - tokens_used: Total tokens used - tokens_limit: Token budget limit - time_elapsed_seconds: Time elapsed in seconds - time_limit_seconds: Time budget limit (default: 600.0 seconds / 10 minutes) - iterations: Number of iterations completed - iterations_limit: Maximum iterations (default: 10) - iteration_tokens: Tokens used per iteration (iteration number -> token count)

See Also

\ No newline at end of file diff --git a/site/api/orchestrators/index.html b/site/api/orchestrators/index.html index 9311993..2398678 100644 --- a/site/api/orchestrators/index.html +++ b/site/api/orchestrators/index.html @@ -1,35 +1 @@ - Orchestrators API Reference - The DETERMINATOR
Skip to content

Orchestrators API Reference

This page documents the API for DeepCritical orchestrators.

IterativeResearchFlow

Module: src.orchestrator.research_flow

Purpose: Single-loop research with search-judge-synthesize cycles.

Methods

run

async def run(
-    self,
-    query: str,
-    background_context: str = "",
-    max_iterations: int | None = None,
-    max_time_minutes: float | None = None,
-    token_budget: int | None = None
-) -> AsyncGenerator[AgentEvent, None]
-

Runs iterative research flow.

Parameters: - query: Research query string - background_context: Background context (default: "") - max_iterations: Maximum iterations (default: from settings) - max_time_minutes: Maximum time in minutes (default: from settings) - token_budget: Token budget (default: from settings)

Yields: AgentEvent objects for: - started: Research started - search_complete: Search completed - judge_complete: Evidence evaluation completed - synthesizing: Generating report - complete: Research completed - error: Error occurred

DeepResearchFlow

Module: src.orchestrator.research_flow

Purpose: Multi-section parallel research with planning and synthesis.

Methods

run

async def run(
-    self,
-    query: str,
-    background_context: str = "",
-    max_iterations_per_section: int | None = None,
-    max_time_minutes: float | None = None,
-    token_budget: int | None = None
-) -> AsyncGenerator[AgentEvent, None]
-

Runs deep research flow.

Parameters: - query: Research query string - background_context: Background context (default: "") - max_iterations_per_section: Maximum iterations per section (default: from settings) - max_time_minutes: Maximum time in minutes (default: from settings) - token_budget: Token budget (default: from settings)

Yields: AgentEvent objects for: - started: Research started - planning: Creating research plan - looping: Running parallel research loops - synthesizing: Synthesizing results - complete: Research completed - error: Error occurred

GraphOrchestrator

Module: src.orchestrator.graph_orchestrator

Purpose: Graph-based execution using Pydantic AI agents as nodes.

Methods

run

async def run(
-    self,
-    query: str,
-    research_mode: str = "auto",
-    use_graph: bool = True
-) -> AsyncGenerator[AgentEvent, None]
-

Runs graph-based research orchestration.

Parameters: - query: Research query string - research_mode: Research mode ("iterative", "deep", or "auto") - use_graph: Whether to use graph execution (default: True)

Yields: AgentEvent objects during graph execution.

Orchestrator Factory

Module: src.orchestrator_factory

Purpose: Factory for creating orchestrators.

Functions

create_orchestrator

def create_orchestrator(
-    search_handler: SearchHandlerProtocol,
-    judge_handler: JudgeHandlerProtocol,
-    config: dict[str, Any],
-    mode: str | None = None
-) -> Any
-

Creates an orchestrator instance.

Parameters: - search_handler: Search handler protocol implementation - judge_handler: Judge handler protocol implementation - config: Configuration dictionary - mode: Orchestrator mode ("simple", "advanced", "magentic", or None for auto-detect)

Returns: Orchestrator instance.

Raises: - ValueError: If requirements not met

Modes: - "simple": Legacy orchestrator - "advanced" or "magentic": Magentic orchestrator (requires OpenAI API key) - None: Auto-detect based on API key availability

MagenticOrchestrator

Module: src.orchestrator_magentic

Purpose: Multi-agent coordination using Microsoft Agent Framework.

Methods

run

async def run(
-    self,
-    query: str,
-    max_rounds: int = 15,
-    max_stalls: int = 3
-) -> AsyncGenerator[AgentEvent, None]
-

Runs Magentic orchestration.

Parameters: - query: Research query string - max_rounds: Maximum rounds (default: 15) - max_stalls: Maximum stalls before reset (default: 3)

Yields: AgentEvent objects converted from Magentic events.

Requirements: - agent-framework-core package - OpenAI API key

See Also

\ No newline at end of file + Orchestrators API Reference - The DETERMINATOR
Skip to content

Orchestrators API Reference

This page documents the API for DeepCritical orchestrators.

IterativeResearchFlow

Module: src.orchestrator.research_flow

Purpose: Single-loop research with search-judge-synthesize cycles.

Methods

run

Runs iterative research flow.

Parameters: - query: Research query string - background_context: Background context (default: "") - output_length: Optional description of desired output length (default: "") - output_instructions: Optional additional instructions for report generation (default: "")

Returns: Final report string.

Note: max_iterations, max_time_minutes, and token_budget are constructor parameters, not run() parameters.

DeepResearchFlow

Module: src.orchestrator.research_flow

Purpose: Multi-section parallel research with planning and synthesis.

Methods

run

Runs deep research flow.

Parameters: - query: Research query string

Returns: Final report string.

Note: max_iterations_per_section, max_time_minutes, and token_budget are constructor parameters, not run() parameters.

GraphOrchestrator

Module: src.orchestrator.graph_orchestrator

Purpose: Graph-based execution using Pydantic AI agents as nodes.

Methods

run

Runs graph-based research orchestration.

Parameters: - query: Research query string

Yields: AgentEvent objects during graph execution.

Note: research_mode and use_graph are constructor parameters, not run() parameters.

Orchestrator Factory

Module: src.orchestrator_factory

Purpose: Factory for creating orchestrators.

Functions

create_orchestrator

Creates an orchestrator instance.

Parameters: - search_handler: Search handler protocol implementation (optional, required for simple mode) - judge_handler: Judge handler protocol implementation (optional, required for simple mode) - config: Configuration object (optional) - mode: Orchestrator mode ("simple", "advanced", "magentic", "iterative", "deep", "auto", or None for auto-detect) - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars)

Returns: Orchestrator instance.

Raises: - ValueError: If requirements not met

Modes: - "simple": Legacy orchestrator - "advanced" or "magentic": Magentic orchestrator (requires OpenAI API key) - None: Auto-detect based on API key availability

MagenticOrchestrator

Module: src.orchestrator_magentic

Purpose: Multi-agent coordination using Microsoft Agent Framework.

Methods

run

Runs Magentic orchestration.

Parameters: - query: Research query string

Yields: AgentEvent objects converted from Magentic events.

Note: max_rounds and max_stalls are constructor parameters, not run() parameters.

Requirements: - agent-framework-core package - OpenAI API key

See Also

\ No newline at end of file diff --git a/site/api/services/index.html b/site/api/services/index.html index e86cf38..1207a98 100644 --- a/site/api/services/index.html +++ b/site/api/services/index.html @@ -1,30 +1,49 @@ - Services API Reference - The DETERMINATOR
Skip to content

Services API Reference

This page documents the API for DeepCritical services.

EmbeddingService

Module: src.services.embeddings

Purpose: Local sentence-transformers for semantic search and deduplication.

Methods

embed

async def embed(self, text: str) -> list[float]
-

Generates embedding for a text string.

Parameters: - text: Text to embed

Returns: Embedding vector as list of floats.

embed_batch

async def embed_batch(self, texts: list[str]) -> list[list[float]]
-

Generates embeddings for multiple texts.

Parameters: - texts: List of texts to embed

Returns: List of embedding vectors.

similarity

async def similarity(self, text1: str, text2: str) -> float
-

Calculates similarity between two texts.

Parameters: - text1: First text - text2: Second text

Returns: Similarity score (0.0-1.0).

find_duplicates

async def find_duplicates(
-    self,
-    texts: list[str],
-    threshold: float = 0.85
-) -> list[tuple[int, int]]
-

Finds duplicate texts based on similarity threshold.

Parameters: - texts: List of texts to check - threshold: Similarity threshold (default: 0.85)

Returns: List of (index1, index2) tuples for duplicate pairs.

Factory Function

get_embedding_service

@lru_cache(maxsize=1)
-def get_embedding_service() -> EmbeddingService
-

Returns singleton EmbeddingService instance.

LlamaIndexRAGService

Module: src.services.rag

Purpose: Retrieval-Augmented Generation using LlamaIndex.

Methods

ingest_evidence

async def ingest_evidence(self, evidence: list[Evidence]) -> None
-

Ingests evidence into RAG service.

Parameters: - evidence: List of Evidence objects to ingest

Note: Requires OpenAI API key for embeddings.

retrieve

async def retrieve(
-    self,
-    query: str,
-    top_k: int = 5
-) -> list[Document]
-

Retrieves relevant documents for a query.

Parameters: - query: Search query string - top_k: Number of top results to return (default: 5)

Returns: List of Document objects with metadata.

query

async def query(
-    self,
-    query: str,
-    top_k: int = 5
-) -> str
-

Queries RAG service and returns formatted results.

Parameters: - query: Search query string - top_k: Number of top results to return (default: 5)

Returns: Formatted query results as string.

Factory Function

get_rag_service

@lru_cache(maxsize=1)
-def get_rag_service() -> LlamaIndexRAGService | None
-

Returns singleton LlamaIndexRAGService instance, or None if OpenAI key not available.

StatisticalAnalyzer

Module: src.services.statistical_analyzer

Purpose: Secure execution of AI-generated statistical code.

Methods

analyze

async def analyze(
-    self,
-    hypothesis: str,
-    evidence: list[Evidence],
-    data_description: str | None = None
-) -> AnalysisResult
-

Analyzes a hypothesis using statistical methods.

Parameters: - hypothesis: Hypothesis to analyze - evidence: List of Evidence objects - data_description: Optional data description

Returns: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - code: Generated analysis code - output: Execution output - error: Error message if execution failed

Note: Requires Modal credentials for sandbox execution.

See Also

\ No newline at end of file + Services API Reference - The DETERMINATOR
Skip to content

Services API Reference

This page documents the API for DeepCritical services.

EmbeddingService

Module: src.services.embeddings

Purpose: Local sentence-transformers for semantic search and deduplication.

Methods

embed

Generates embedding for a text string.

Parameters: - text: Text to embed

Returns: Embedding vector as list of floats.

embed_batch

async def embed_batch(self, texts: list[str]) -> list[list[float]]
+

Generates embeddings for multiple texts.

Parameters: - texts: List of texts to embed

Returns: List of embedding vectors.

similarity

async def similarity(self, text1: str, text2: str) -> float
+

Calculates similarity between two texts.

Parameters: - text1: First text - text2: Second text

Returns: Similarity score (0.0-1.0).

find_duplicates

async def find_duplicates(
+    self,
+    texts: list[str],
+    threshold: float = 0.85
+) -> list[tuple[int, int]]
+

Finds duplicate texts based on similarity threshold.

Parameters: - texts: List of texts to check - threshold: Similarity threshold (default: 0.85)

Returns: List of (index1, index2) tuples for duplicate pairs.

add_evidence

async def add_evidence(
+    self,
+    evidence_id: str,
+    content: str,
+    metadata: dict[str, Any]
+) -> None
+

Adds evidence to vector store for semantic search.

Parameters: - evidence_id: Unique identifier for the evidence - content: Evidence text content - metadata: Additional metadata dictionary

search_similar

async def search_similar(
+    self,
+    query: str,
+    n_results: int = 5
+) -> list[dict[str, Any]]
+

Finds semantically similar evidence.

Parameters: - query: Search query string - n_results: Number of results to return (default: 5)

Returns: List of dictionaries with id, content, metadata, and distance keys.

deduplicate

async def deduplicate(
+    self,
+    new_evidence: list[Evidence],
+    threshold: float = 0.9
+) -> list[Evidence]
+

Removes semantically duplicate evidence.

Parameters: - new_evidence: List of evidence items to deduplicate - threshold: Similarity threshold (default: 0.9, where 0.9 = 90% similar is duplicate)

Returns: List of unique evidence items (not already in vector store).

Factory Function

get_embedding_service

@lru_cache(maxsize=1)
+def get_embedding_service() -> EmbeddingService
+

Returns singleton EmbeddingService instance.

LlamaIndexRAGService

Module: src.services.rag

Purpose: Retrieval-Augmented Generation using LlamaIndex.

Methods

ingest_evidence

Ingests evidence into RAG service.

Parameters: - evidence_list: List of Evidence objects to ingest

Note: Supports multiple embedding providers (OpenAI, local sentence-transformers, Hugging Face).

retrieve

def retrieve(
+    self,
+    query: str,
+    top_k: int | None = None
+) -> list[dict[str, Any]]
+

Retrieves relevant documents for a query.

Parameters: - query: Search query string - top_k: Number of top results to return (defaults to similarity_top_k from constructor)

Returns: List of dictionaries with text, score, and metadata keys.

query

def query(
+    self,
+    query_str: str,
+    top_k: int | None = None
+) -> str
+

Queries RAG service and returns synthesized response.

Parameters: - query_str: Query string - top_k: Number of results to use (defaults to similarity_top_k from constructor)

Returns: Synthesized response string.

Raises: - ConfigurationError: If no LLM API key is available for query synthesis

ingest_documents

def ingest_documents(self, documents: list[Any]) -> None
+

Ingests raw LlamaIndex Documents.

Parameters: - documents: List of LlamaIndex Document objects

clear_collection

def clear_collection(self) -> None
+

Clears all documents from the collection.

Factory Function

get_rag_service

def get_rag_service(
+    collection_name: str = "deepcritical_evidence",
+    oauth_token: str | None = None,
+    **kwargs: Any
+) -> LlamaIndexRAGService
+

Get or create a RAG service instance.

Parameters: - collection_name: Name of the ChromaDB collection (default: "deepcritical_evidence") - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars) - **kwargs: Additional arguments for LlamaIndexRAGService (e.g., use_openai_embeddings=False)

Returns: Configured LlamaIndexRAGService instance.

Note: By default, uses local embeddings (sentence-transformers) which require no API keys.

StatisticalAnalyzer

Module: src.services.statistical_analyzer

Purpose: Secure execution of AI-generated statistical code.

Methods

analyze

async def analyze(
+    self,
+    query: str,
+    evidence: list[Evidence],
+    hypothesis: dict[str, Any] | None = None
+) -> AnalysisResult
+

Analyzes a research question using statistical methods.

Parameters: - query: The research question - evidence: List of Evidence objects to analyze - hypothesis: Optional hypothesis dict with drug, target, pathway, effect, confidence keys

Returns: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - confidence: Confidence in verdict (0.0-1.0) - statistical_evidence: Summary of statistical findings - code_generated: Python code that was executed - execution_output: Output from code execution - key_takeaways: Key takeaways from analysis - limitations: List of limitations

Note: Requires Modal credentials for sandbox execution.

See Also

\ No newline at end of file diff --git a/site/api/tools/index.html b/site/api/tools/index.html index 30dd049..b7d76db 100644 --- a/site/api/tools/index.html +++ b/site/api/tools/index.html @@ -1,44 +1,51 @@ - Tools API Reference - The DETERMINATOR
Skip to content

Tools API Reference

This page documents the API for DeepCritical search tools.

SearchTool Protocol

All tools implement the SearchTool protocol:

class SearchTool(Protocol):
-    @property
-    def name(self) -> str: ...
-    
-    async def search(
-        self, 
-        query: str, 
-        max_results: int = 10
-    ) -> list[Evidence]: ...
-

PubMedTool

Module: src.tools.pubmed

Purpose: Search peer-reviewed biomedical literature from PubMed.

Properties

name

@property
-def name(self) -> str
-

Returns tool name: "pubmed"

Methods

async def search(
-    self,
-    query: str,
-    max_results: int = 10
-) -> list[Evidence]
-

Searches PubMed for articles.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with PubMed articles.

Raises: - SearchError: If search fails - RateLimitError: If rate limit is exceeded

ClinicalTrialsTool

Module: src.tools.clinicaltrials

Purpose: Search ClinicalTrials.gov for interventional studies.

Properties

name

@property
-def name(self) -> str
-

Returns tool name: "clinicaltrials"

Methods

search

async def search(
-    self,
-    query: str,
-    max_results: int = 10
-) -> list[Evidence]
-

Searches ClinicalTrials.gov for trials.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with clinical trials.

Note: Only returns interventional studies with status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION

Raises: - SearchError: If search fails

EuropePMCTool

Module: src.tools.europepmc

Purpose: Search Europe PMC for preprints and peer-reviewed articles.

Properties

name

@property
-def name(self) -> str
-

Returns tool name: "europepmc"

Methods

search

async def search(
-    self,
-    query: str,
-    max_results: int = 10
-) -> list[Evidence]
-

Searches Europe PMC for articles and preprints.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with articles/preprints.

Note: Includes both preprints (marked with [PREPRINT - Not peer-reviewed]) and peer-reviewed articles.

Raises: - SearchError: If search fails

RAGTool

Module: src.tools.rag_tool

Purpose: Semantic search within collected evidence.

Properties

name

@property
-def name(self) -> str
-

Returns tool name: "rag"

Methods

search

async def search(
-    self,
-    query: str,
-    max_results: int = 10
-) -> list[Evidence]
-

Searches collected evidence using semantic similarity.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects from collected evidence.

Note: Requires evidence to be ingested into RAG service first.

SearchHandler

Module: src.tools.search_handler

Purpose: Orchestrates parallel searches across multiple tools.

Methods

search

async def search(
-    self,
-    query: str,
-    tools: list[SearchTool] | None = None,
-    max_results_per_tool: int = 10
-) -> SearchResult
-

Searches multiple tools in parallel.

Parameters: - query: Search query string - tools: List of tools to use (default: all available tools) - max_results_per_tool: Maximum results per tool (default: 10)

Returns: SearchResult with: - evidence: Aggregated list of evidence - tool_results: Results per tool - total_count: Total number of results

Note: Uses asyncio.gather() for parallel execution. Handles tool failures gracefully.

See Also

\ No newline at end of file + Tools API Reference - The DETERMINATOR
Skip to content

Tools API Reference

This page documents the API for DeepCritical search tools.

SearchTool Protocol

All tools implement the SearchTool protocol:

class SearchTool(Protocol):
+    @property
+    def name(self) -> str: ...
+    
+    async def search(
+        self, 
+        query: str, 
+        max_results: int = 10
+    ) -> list[Evidence]: ...
+

PubMedTool

Module: src.tools.pubmed

Purpose: Search peer-reviewed biomedical literature from PubMed.

Properties

name

@property
+def name(self) -> str
+

Returns tool name: "pubmed"

Methods

async def search(
+    self,
+    query: str,
+    max_results: int = 10
+) -> list[Evidence]
+

Searches PubMed for articles.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with PubMed articles.

Raises: - SearchError: If search fails (timeout, HTTP error, XML parsing error) - RateLimitError: If rate limit is exceeded (429 status code)

Note: Uses NCBI E-utilities (ESearch → EFetch). Rate limit: 0.34s between requests. Handles single vs. multiple articles.

ClinicalTrialsTool

Module: src.tools.clinicaltrials

Purpose: Search ClinicalTrials.gov for interventional studies.

Properties

name

@property
+def name(self) -> str
+

Returns tool name: "clinicaltrials"

Methods

search

async def search(
+    self,
+    query: str,
+    max_results: int = 10
+) -> list[Evidence]
+

Searches ClinicalTrials.gov for trials.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with clinical trials.

Note: Only returns interventional studies with status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION. Uses requests library (NOT httpx - WAF blocks httpx). Runs in thread pool for async compatibility.

Raises: - SearchError: If search fails (HTTP error, request exception)

EuropePMCTool

Module: src.tools.europepmc

Purpose: Search Europe PMC for preprints and peer-reviewed articles.

Properties

name

@property
+def name(self) -> str
+

Returns tool name: "europepmc"

Methods

search

async def search(
+    self,
+    query: str,
+    max_results: int = 10
+) -> list[Evidence]
+

Searches Europe PMC for articles and preprints.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with articles/preprints.

Note: Includes both preprints (marked with [PREPRINT - Not peer-reviewed]) and peer-reviewed articles. Handles preprint markers. Builds URLs from DOI or PMID.

Raises: - SearchError: If search fails (HTTP error, connection error)

RAGTool

Module: src.tools.rag_tool

Purpose: Semantic search within collected evidence.

Initialization

def __init__(
+    self,
+    rag_service: LlamaIndexRAGService | None = None,
+    oauth_token: str | None = None
+) -> None
+

Parameters: - rag_service: Optional RAG service instance. If None, will be lazy-initialized. - oauth_token: Optional OAuth token from HuggingFace login (for RAG LLM)

Properties

name

@property
+def name(self) -> str
+

Returns tool name: "rag"

Methods

search

async def search(
+    self,
+    query: str,
+    max_results: int = 10
+) -> list[Evidence]
+

Searches collected evidence using semantic similarity.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects from collected evidence.

Raises: - ConfigurationError: If RAG service is unavailable

Note: Requires evidence to be ingested into RAG service first. Wraps LlamaIndexRAGService. Returns Evidence from RAG results.

SearchHandler

Module: src.tools.search_handler

Purpose: Orchestrates parallel searches across multiple tools.

Initialization

def __init__(
+    self,
+    tools: list[SearchTool],
+    timeout: float = 30.0,
+    include_rag: bool = False,
+    auto_ingest_to_rag: bool = True,
+    oauth_token: str | None = None
+) -> None
+

Parameters: - tools: List of search tools to use - timeout: Timeout for each search in seconds (default: 30.0) - include_rag: Whether to include RAG tool in searches (default: False) - auto_ingest_to_rag: Whether to automatically ingest results into RAG (default: True) - oauth_token: Optional OAuth token from HuggingFace login (for RAG LLM)

Methods

execute

Searches multiple tools in parallel.

Parameters: - query: Search query string - max_results_per_tool: Maximum results per tool (default: 10)

Returns: SearchResult with: - query: The search query - evidence: Aggregated list of evidence - sources_searched: List of source names searched - total_found: Total number of results - errors: List of error messages from failed tools

Raises: - SearchError: If search times out

Note: Uses asyncio.gather() for parallel execution. Handles tool failures gracefully (returns errors in SearchResult.errors). Automatically ingests evidence into RAG if enabled.

See Also

\ No newline at end of file diff --git a/site/architecture/agents/index.html b/site/architecture/agents/index.html index bd43b2d..bcc3d61 100644 --- a/site/architecture/agents/index.html +++ b/site/architecture/agents/index.html @@ -1 +1 @@ - Agents - The DETERMINATOR
Skip to content

Agents Architecture

DeepCritical uses Pydantic AI agents for all AI-powered operations. All agents follow a consistent pattern and use structured output types.

Agent Pattern

All agents use the Pydantic AI Agent class with the following structure:

  • System Prompt: Module-level constant with date injection
  • Agent Class: __init__(model: Any | None = None)
  • Main Method: Async method (e.g., async def evaluate(), async def write_report())
  • Factory Function: def create_agent_name(model: Any | None = None) -> AgentName

Model Initialization

Agents use get_model() from src/agent_factory/judges.py if no model is provided. This supports:

  • OpenAI models
  • Anthropic models
  • HuggingFace Inference API models

The model selection is based on the configured LLM_PROVIDER in settings.

Error Handling

Agents return fallback values on failure rather than raising exceptions:

  • KnowledgeGapOutput(research_complete=False, outstanding_gaps=[...])
  • Empty strings for text outputs
  • Default structured outputs

All errors are logged with context using structlog.

Input Validation

All agents validate inputs:

  • Check that queries/inputs are not empty
  • Truncate very long inputs with warnings
  • Handle None values gracefully

Output Types

Agents use structured output types from src/utils/models.py:

  • KnowledgeGapOutput: Research completeness evaluation
  • AgentSelectionPlan: Tool selection plan
  • ReportDraft: Long-form report structure
  • ParsedQuery: Query parsing and mode detection

For text output (writer agents), agents return str directly.

Agent Types

Knowledge Gap Agent

File: src/agents/knowledge_gap.py

Purpose: Evaluates research state and identifies knowledge gaps.

Output: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

Methods: - async def evaluate(query, background_context, conversation_history, iteration, time_elapsed_minutes, max_time_minutes) -> KnowledgeGapOutput

Tool Selector Agent

File: src/agents/tool_selector.py

Purpose: Selects appropriate tools for addressing knowledge gaps.

Output: AgentSelectionPlan with list of AgentTask objects.

Available Agents: - WebSearchAgent: General web search for fresh information - SiteCrawlerAgent: Research specific entities/companies - RAGAgent: Semantic search within collected evidence

Writer Agent

File: src/agents/writer.py

Purpose: Generates final reports from research findings.

Output: Markdown string with numbered citations.

Methods: - async def write_report(query, findings, output_length, output_instructions) -> str

Features: - Validates inputs - Truncates very long findings (max 50000 chars) with warning - Retry logic for transient failures (3 retries) - Citation validation before returning

Long Writer Agent

File: src/agents/long_writer.py

Purpose: Long-form report generation with section-by-section writing.

Input/Output: Uses ReportDraft models.

Methods: - async def write_next_section(query, draft, section_title, section_content) -> LongWriterOutput - async def write_report(query, report_title, report_draft) -> str

Features: - Writes sections iteratively - Aggregates references across sections - Reformats section headings and references - Deduplicates and renumbers references

Proofreader Agent

File: src/agents/proofreader.py

Purpose: Proofreads and polishes report drafts.

Input: ReportDraft Output: Polished markdown string

Methods: - async def proofread(query, report_title, report_draft) -> str

Features: - Removes duplicate content across sections - Adds executive summary if multiple sections - Preserves all references and citations - Improves flow and readability

Thinking Agent

File: src/agents/thinking.py

Purpose: Generates observations from conversation history.

Output: Observation string

Methods: - async def generate_observations(query, background_context, conversation_history) -> str

Input Parser Agent

File: src/agents/input_parser.py

Purpose: Parses and improves user queries, detects research mode.

Output: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: "iterative" or "deep" - key_entities: List of key entities - research_questions: List of research questions

Factory Functions

All agents have factory functions in src/agent_factory/agents.py:

Factory functions: - Use get_model() if no model provided - Raise ConfigurationError if creation fails - Log agent creation

See Also

\ No newline at end of file + Agents - The DETERMINATOR
Skip to content

Agents Architecture

DeepCritical uses Pydantic AI agents for all AI-powered operations. All agents follow a consistent pattern and use structured output types.

Agent Pattern

Pydantic AI Agents

Pydantic AI agents use the Agent class with the following structure:

  • System Prompt: Module-level constant with date injection
  • Agent Class: __init__(model: Any | None = None)
  • Main Method: Async method (e.g., async def evaluate(), async def write_report())
  • Factory Function: def create_agent_name(model: Any | None = None, oauth_token: str | None = None) -> AgentName

Note: Factory functions accept an optional oauth_token parameter for HuggingFace authentication, which takes priority over environment variables.

Model Initialization

Agents use get_model() from src/agent_factory/judges.py if no model is provided. This supports:

  • OpenAI models
  • Anthropic models
  • HuggingFace Inference API models

The model selection is based on the configured LLM_PROVIDER in settings.

Error Handling

Agents return fallback values on failure rather than raising exceptions:

  • KnowledgeGapOutput(research_complete=False, outstanding_gaps=[...])
  • Empty strings for text outputs
  • Default structured outputs

All errors are logged with context using structlog.

Input Validation

All agents validate inputs:

  • Check that queries/inputs are not empty
  • Truncate very long inputs with warnings
  • Handle None values gracefully

Output Types

Agents use structured output types from src/utils/models.py:

  • KnowledgeGapOutput: Research completeness evaluation
  • AgentSelectionPlan: Tool selection plan
  • ReportDraft: Long-form report structure
  • ParsedQuery: Query parsing and mode detection

For text output (writer agents), agents return str directly.

Agent Types

Knowledge Gap Agent

File: src/agents/knowledge_gap.py

Purpose: Evaluates research state and identifies knowledge gaps.

Output: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

Methods: - async def evaluate(query, background_context, conversation_history, iteration, time_elapsed_minutes, max_time_minutes) -> KnowledgeGapOutput

Tool Selector Agent

File: src/agents/tool_selector.py

Purpose: Selects appropriate tools for addressing knowledge gaps.

Output: AgentSelectionPlan with list of AgentTask objects.

Available Agents: - WebSearchAgent: General web search for fresh information - SiteCrawlerAgent: Research specific entities/companies - RAGAgent: Semantic search within collected evidence

Writer Agent

File: src/agents/writer.py

Purpose: Generates final reports from research findings.

Output: Markdown string with numbered citations.

Methods: - async def write_report(query, findings, output_length, output_instructions) -> str

Features: - Validates inputs - Truncates very long findings (max 50000 chars) with warning - Retry logic for transient failures (3 retries) - Citation validation before returning

Long Writer Agent

File: src/agents/long_writer.py

Purpose: Long-form report generation with section-by-section writing.

Input/Output: Uses ReportDraft models.

Methods: - async def write_next_section(query, draft, section_title, section_content) -> LongWriterOutput - async def write_report(query, report_title, report_draft) -> str

Features: - Writes sections iteratively - Aggregates references across sections - Reformats section headings and references - Deduplicates and renumbers references

Proofreader Agent

File: src/agents/proofreader.py

Purpose: Proofreads and polishes report drafts.

Input: ReportDraft Output: Polished markdown string

Methods: - async def proofread(query, report_title, report_draft) -> str

Features: - Removes duplicate content across sections - Adds executive summary if multiple sections - Preserves all references and citations - Improves flow and readability

Thinking Agent

File: src/agents/thinking.py

Purpose: Generates observations from conversation history.

Output: Observation string

Methods: - async def generate_observations(query, background_context, conversation_history) -> str

Input Parser Agent

File: src/agents/input_parser.py

Purpose: Parses and improves user queries, detects research mode.

Output: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: "iterative" or "deep" - key_entities: List of key entities - research_questions: List of research questions

Magentic Agents

The following agents use the BaseAgent pattern from agent-framework and are used exclusively with MagenticOrchestrator:

Hypothesis Agent

File: src/agents/hypothesis_agent.py

Purpose: Generates mechanistic hypotheses based on evidence.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Uses internal Pydantic AI Agent with HypothesisAssessment output type - Accesses shared evidence_store for evidence - Uses embedding service for diverse evidence selection (MMR algorithm) - Stores hypotheses in shared context

Search Agent

File: src/agents/search_agent.py

Purpose: Wraps SearchHandler as an agent for Magentic orchestrator.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Executes searches via SearchHandlerProtocol - Deduplicates evidence using embedding service - Searches for semantically related evidence - Updates shared evidence store

Analysis Agent

File: src/agents/analysis_agent.py

Purpose: Performs statistical analysis using Modal sandbox.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Wraps StatisticalAnalyzer service - Analyzes evidence and hypotheses - Returns verdict (SUPPORTED/REFUTED/INCONCLUSIVE) - Stores analysis results in shared context

Report Agent (Magentic)

File: src/agents/report_agent.py

Purpose: Generates structured scientific reports from evidence and hypotheses.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Uses internal Pydantic AI Agent with ResearchReport output type - Accesses shared evidence store and hypotheses - Validates citations before returning - Formats report as markdown

Judge Agent

File: src/agents/judge_agent.py

Purpose: Evaluates evidence quality and determines if sufficient for synthesis.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse - async def run_stream(messages, thread, **kwargs) -> AsyncIterable[AgentRunResponseUpdate]

Features: - Wraps JudgeHandlerProtocol - Accesses shared evidence store - Returns JudgeAssessment with sufficient flag, confidence, and recommendation

Agent Patterns

DeepCritical uses two distinct agent patterns:

1. Pydantic AI Agents (Traditional Pattern)

These agents use the Pydantic AI Agent class directly and are used in iterative and deep research flows:

  • Pattern: Agent(model, output_type, system_prompt)
  • Initialization: __init__(model: Any | None = None)
  • Methods: Agent-specific async methods (e.g., async def evaluate(), async def write_report())
  • Examples: KnowledgeGapAgent, ToolSelectorAgent, WriterAgent, LongWriterAgent, ProofreaderAgent, ThinkingAgent, InputParserAgent

2. Magentic Agents (Agent-Framework Pattern)

These agents use the BaseAgent class from agent-framework and are used in Magentic orchestrator:

  • Pattern: BaseAgent from agent-framework with async def run() method
  • Initialization: __init__(evidence_store, embedding_service, ...)
  • Methods: async def run(messages, thread, **kwargs) -> AgentRunResponse
  • Examples: HypothesisAgent, SearchAgent, AnalysisAgent, ReportAgent, JudgeAgent

Note: Magentic agents are used exclusively with the MagenticOrchestrator and follow the agent-framework protocol for multi-agent coordination.

Factory Functions

All agents have factory functions in src/agent_factory/agents.py:

Factory functions: - Use get_model() if no model provided - Accept oauth_token parameter for HuggingFace authentication - Raise ConfigurationError if creation fails - Log agent creation

See Also

\ No newline at end of file diff --git a/site/architecture/graph-orchestration/index.html b/site/architecture/graph-orchestration/index.html deleted file mode 100644 index a454685..0000000 --- a/site/architecture/graph-orchestration/index.html +++ /dev/null @@ -1,9 +0,0 @@ - Graph Orchestration Architecture - The DETERMINATOR
Skip to content

Graph Orchestration Architecture

Overview

Phase 4 implements a graph-based orchestration system for research workflows using Pydantic AI agents as nodes. This enables better parallel execution, conditional routing, and state management compared to simple agent chains.

Graph Structure

Nodes

Graph nodes represent different stages in the research workflow:

  1. Agent Nodes: Execute Pydantic AI agents
  2. Input: Prompt/query
  3. Output: Structured or unstructured response

  4. Examples: KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent

  5. State Nodes: Update or read workflow state

  6. Input: Current state
  7. Output: Updated state

  8. Examples: Update evidence, update conversation history

  9. Decision Nodes: Make routing decisions based on conditions

  10. Input: Current state/results
  11. Output: Next node ID

  12. Examples: Continue research vs. complete research

  13. Parallel Nodes: Execute multiple nodes concurrently

  14. Input: List of node IDs
  15. Output: Aggregated results
  16. Examples: Parallel iterative research loops

Edges

Edges define transitions between nodes:

  1. Sequential Edges: Always traversed (no condition)
  2. From: Source node
  3. To: Target node

  4. Condition: None (always True)

  5. Conditional Edges: Traversed based on condition

  6. From: Source node
  7. To: Target node
  8. Condition: Callable that returns bool

  9. Example: If research complete → go to writer, else → continue loop

  10. Parallel Edges: Used for parallel execution branches

  11. From: Parallel node
  12. To: Multiple target nodes
  13. Execution: All targets run concurrently

Graph Patterns

Iterative Research Graph

[Input] → [Thinking] → [Knowledge Gap] → [Decision: Complete?]
-                                              ↓ No          ↓ Yes
-                                    [Tool Selector]    [Writer]
-                                              ↓
-                                    [Execute Tools] → [Loop Back]
-

Deep Research Graph

[Input] → [Planner] → [Parallel Iterative Loops] → [Synthesizer]
-                           ↓         ↓         ↓
-                        [Loop1]  [Loop2]  [Loop3]
-

State Management

State is managed via WorkflowState using ContextVar for thread-safe isolation:

  • Evidence: Collected evidence from searches
  • Conversation: Iteration history (gaps, tool calls, findings, thoughts)
  • Embedding Service: For semantic search

State transitions occur at state nodes, which update the global workflow state.

Execution Flow

  1. Graph Construction: Build graph from nodes and edges
  2. Graph Validation: Ensure graph is valid (no cycles, all nodes reachable)
  3. Graph Execution: Traverse graph from entry node
  4. Node Execution: Execute each node based on type
  5. Edge Evaluation: Determine next node(s) based on edges
  6. Parallel Execution: Use asyncio.gather() for parallel nodes
  7. State Updates: Update state at state nodes
  8. Event Streaming: Yield events during execution for UI

Conditional Routing

Decision nodes evaluate conditions and return next node IDs:

  • Knowledge Gap Decision: If research_complete → writer, else → tool selector
  • Budget Decision: If budget exceeded → exit, else → continue
  • Iteration Decision: If max iterations → exit, else → continue

Parallel Execution

Parallel nodes execute multiple nodes concurrently:

  • Each parallel branch runs independently
  • Results are aggregated after all branches complete
  • State is synchronized after parallel execution
  • Errors in one branch don't stop other branches

Budget Enforcement

Budget constraints are enforced at decision nodes:

  • Token Budget: Track LLM token usage
  • Time Budget: Track elapsed time
  • Iteration Budget: Track iteration count

If any budget is exceeded, execution routes to exit node.

Error Handling

Errors are handled at multiple levels:

  1. Node Level: Catch errors in individual node execution
  2. Graph Level: Handle errors during graph traversal
  3. State Level: Rollback state changes on error

Errors are logged and yield error events for UI.

Backward Compatibility

Graph execution is optional via feature flag:

  • USE_GRAPH_EXECUTION=true: Use graph-based execution
  • USE_GRAPH_EXECUTION=false: Use agent chain execution (existing)

This allows gradual migration and fallback if needed.

\ No newline at end of file diff --git a/site/architecture/graph_orchestration/index.html b/site/architecture/graph_orchestration/index.html index b090d72..e54981e 100644 --- a/site/architecture/graph_orchestration/index.html +++ b/site/architecture/graph_orchestration/index.html @@ -1,12 +1,12 @@ - Graph Orchestration - The DETERMINATOR
Skip to content

Graph Orchestration Architecture

Graph Patterns

Iterative Research Graph

[Input] → [Thinking] → [Knowledge Gap] → [Decision: Complete?]
-                                              ↓ No          ↓ Yes
-                                    [Tool Selector]    [Writer]
-                                              ↓
-                                    [Execute Tools] → [Loop Back]
-

Deep Research Graph

[Input] → [Planner] → [Parallel Iterative Loops] → [Synthesizer]
-                           ↓         ↓         ↓
-                        [Loop1]  [Loop2]  [Loop3]
-

Deep Research


+ Graph Orchestration - The DETERMINATOR      
 
 
  
 
  
 
 
 
 
  
 
 
 
 
 
Graph Orchestration Architecture
 
Overview
 
DeepCritical implements a graph-based orchestration system for research workflows using Pydantic AI agents as nodes. This enables better parallel execution, conditional routing, and state management compared to simple agent chains.
 
Graph Patterns
 
Iterative Research Graph
 
The iterative research graph follows this pattern:
 
[Input] → [Thinking] → [Knowledge Gap] → [Decision: Complete?]
+                                              ↓ No          ↓ Yes
+                                    [Tool Selector]    [Writer]
+                                              ↓
+                                    [Execute Tools] → [Loop Back]
+
 
Node IDs: thinking → knowledge_gap → continue_decision → tool_selector/writer → execute_tools → (loop back to thinking)
 
Special Node Handling: - execute_tools: State node that uses search_handler to execute searches and add evidence to workflow state - continue_decision: Decision node that routes based on research_complete flag from KnowledgeGapOutput
 
Deep Research Graph
 
The deep research graph follows this pattern:
 
[Input] → [Planner] → [Store Plan] → [Parallel Loops] → [Collect Drafts] → [Synthesizer]
+                                        ↓         ↓         ↓
+                                     [Loop1]  [Loop2]  [Loop3]
+
 
Node IDs: planner → store_plan → parallel_loops → collect_drafts → synthesizer
 
Special Node Handling: - planner: Agent node that creates ReportPlan with report outline - store_plan: State node that stores ReportPlan in context for parallel loops - parallel_loops: Parallel node that executes IterativeResearchFlow instances for each section - collect_drafts: State node that collects section drafts from parallel loops - synthesizer: Agent node that calls LongWriterAgent.write_report() directly with ReportDraft
 
Deep Research
 

 sequenceDiagram
     actor User
     participant GraphOrchestrator
@@ -41,7 +41,7 @@
     end
     
     GraphOrchestrator->>User: AsyncGenerator[AgentEvent]
-
 
Iterative Research
 
sequenceDiagram
+
 
Iterative Research
 
sequenceDiagram
     participant IterativeFlow
     participant ThinkingAgent
     participant KnowledgeGapAgent
@@ -72,4 +72,4 @@
             IterativeFlow->>JudgeHandler: assess_evidence()
             JudgeHandler-->>IterativeFlow: should_continue
         end
-    end
 
Graph Structure
 
Nodes
 
Graph nodes represent different stages in the research workflow:
 
     
  1. Agent Nodes: Execute Pydantic AI agents
    2.  
  2. Input: Prompt/query
    3.  
  3. Output: Structured or unstructured response
    4.  
  4.  
    Examples: KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent
     
    5.  
  5.  
    State Nodes: Update or read workflow state
     
    6.  
  6. Input: Current state
    7.  
  7. Output: Updated state
    8.  
  8.  
    Examples: Update evidence, update conversation history
     
    9.  
  9.  
    Decision Nodes: Make routing decisions based on conditions
     
    10.  
  10. Input: Current state/results
    11.  
  11. Output: Next node ID
    12.  
  12.  
    Examples: Continue research vs. complete research
     
    13.  
  13.  
    Parallel Nodes: Execute multiple nodes concurrently
     
    14.  
  14. Input: List of node IDs
    15.  
  15. Output: Aggregated results
    16.  
  16. Examples: Parallel iterative research loops
    17.  
 
Edges
 
Edges define transitions between nodes:
 
     
  1. Sequential Edges: Always traversed (no condition)
    2.  
  2. From: Source node
    3.  
  3. To: Target node
    4.  
  4.  
    Condition: None (always True)
     
    5.  
  5.  
    Conditional Edges: Traversed based on condition
     
    6.  
  6. From: Source node
    7.  
  7. To: Target node
    8.  
  8. Condition: Callable that returns bool
    9.  
  9.  
    Example: If research complete → go to writer, else → continue loop
     
    10.  
  10.  
    Parallel Edges: Used for parallel execution branches
     
    11.  
  11. From: Parallel node
    12.  
  12. To: Multiple target nodes
    13.  
  13. Execution: All targets run concurrently
    14.  
 
State Management
 
State is managed via WorkflowState using ContextVar for thread-safe isolation:
 
     
  • Evidence: Collected evidence from searches
    •  
  • Conversation: Iteration history (gaps, tool calls, findings, thoughts)
    •  
  • Embedding Service: For semantic search
    •  
 
State transitions occur at state nodes, which update the global workflow state.
 
Execution Flow
 
     
  1. Graph Construction: Build graph from nodes and edges
    2.  
  2. Graph Validation: Ensure graph is valid (no cycles, all nodes reachable)
    3.  
  3. Graph Execution: Traverse graph from entry node
    4.  
  4. Node Execution: Execute each node based on type
    5.  
  5. Edge Evaluation: Determine next node(s) based on edges
    6.  
  6. Parallel Execution: Use asyncio.gather() for parallel nodes
    7.  
  7. State Updates: Update state at state nodes
    8.  
  8. Event Streaming: Yield events during execution for UI
    9.  
 
Conditional Routing
 
Decision nodes evaluate conditions and return next node IDs:
 
     
  • Knowledge Gap Decision: If research_complete → writer, else → tool selector
    •  
  • Budget Decision: If budget exceeded → exit, else → continue
    •  
  • Iteration Decision: If max iterations → exit, else → continue
    •  
 
Parallel Execution
 
Parallel nodes execute multiple nodes concurrently:
 
     
  • Each parallel branch runs independently
    •  
  • Results are aggregated after all branches complete
    •  
  • State is synchronized after parallel execution
    •  
  • Errors in one branch don't stop other branches
    •  
 
Budget Enforcement
 
Budget constraints are enforced at decision nodes:
 
     
  • Token Budget: Track LLM token usage
    •  
  • Time Budget: Track elapsed time
    •  
  • Iteration Budget: Track iteration count
    •  
 
If any budget is exceeded, execution routes to exit node.
 
Error Handling
 
Errors are handled at multiple levels:
 
     
  1. Node Level: Catch errors in individual node execution
    2.  
  2. Graph Level: Handle errors during graph traversal
    3.  
  3. State Level: Rollback state changes on error
    4.  
 
Errors are logged and yield error events for UI.
 
Backward Compatibility
 
Graph execution is optional via feature flag:
 
     
  • USE_GRAPH_EXECUTION=true: Use graph-based execution
    •  
  • USE_GRAPH_EXECUTION=false: Use agent chain execution (existing)
    •  
 
This allows gradual migration and fallback if needed.
 
See Also
  
 
  
  
  
 
 
 
    
\ No newline at end of file
+    end

Graph Structure

Nodes

Graph nodes represent different stages in the research workflow:

  1. Agent Nodes: Execute Pydantic AI agents
  2. Input: Prompt/query
  3. Output: Structured or unstructured response

  4. Examples: KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent

  5. State Nodes: Update or read workflow state

  6. Input: Current state
  7. Output: Updated state

  8. Examples: Update evidence, update conversation history

  9. Decision Nodes: Make routing decisions based on conditions

  10. Input: Current state/results
  11. Output: Next node ID

  12. Examples: Continue research vs. complete research

  13. Parallel Nodes: Execute multiple nodes concurrently

  14. Input: List of node IDs
  15. Output: Aggregated results
  16. Examples: Parallel iterative research loops

Edges

Edges define transitions between nodes:

  1. Sequential Edges: Always traversed (no condition)
  2. From: Source node
  3. To: Target node

  4. Condition: None (always True)

  5. Conditional Edges: Traversed based on condition

  6. From: Source node
  7. To: Target node
  8. Condition: Callable that returns bool

  9. Example: If research complete → go to writer, else → continue loop

  10. Parallel Edges: Used for parallel execution branches

  11. From: Parallel node
  12. To: Multiple target nodes
  13. Execution: All targets run concurrently

State Management

State is managed via WorkflowState using ContextVar for thread-safe isolation:

  • Evidence: Collected evidence from searches
  • Conversation: Iteration history (gaps, tool calls, findings, thoughts)
  • Embedding Service: For semantic search

State transitions occur at state nodes, which update the global workflow state.

Execution Flow

  1. Graph Construction: Build graph from nodes and edges using create_iterative_graph() or create_deep_graph()
  2. Graph Validation: Ensure graph is valid (no cycles, all nodes reachable) via ResearchGraph.validate_structure()
  3. Graph Execution: Traverse graph from entry node using GraphOrchestrator._execute_graph()
  4. Node Execution: Execute each node based on type:
  5. Agent Nodes: Call agent.run() with transformed input
  6. State Nodes: Update workflow state via state_updater function
  7. Decision Nodes: Evaluate decision_function to get next node ID
  8. Parallel Nodes: Execute all parallel nodes concurrently via asyncio.gather()
  9. Edge Evaluation: Determine next node(s) based on edges and conditions
  10. Parallel Execution: Use asyncio.gather() for parallel nodes
  11. State Updates: Update state at state nodes via GraphExecutionContext.update_state()
  12. Event Streaming: Yield AgentEvent objects during execution for UI

GraphExecutionContext

The GraphExecutionContext class manages execution state during graph traversal:

  • State: Current WorkflowState instance
  • Budget Tracker: BudgetTracker instance for budget enforcement
  • Node Results: Dictionary storing results from each node execution
  • Visited Nodes: Set of node IDs that have been executed
  • Current Node: ID of the node currently being executed

Methods: - set_node_result(node_id, result): Store result from node execution - get_node_result(node_id): Retrieve stored result - has_visited(node_id): Check if node was visited - mark_visited(node_id): Mark node as visited - update_state(updater, data): Update workflow state

Conditional Routing

Decision nodes evaluate conditions and return next node IDs:

  • Knowledge Gap Decision: If research_complete → writer, else → tool selector
  • Budget Decision: If budget exceeded → exit, else → continue
  • Iteration Decision: If max iterations → exit, else → continue

Parallel Execution

Parallel nodes execute multiple nodes concurrently:

  • Each parallel branch runs independently
  • Results are aggregated after all branches complete
  • State is synchronized after parallel execution
  • Errors in one branch don't stop other branches

Budget Enforcement

Budget constraints are enforced at decision nodes:

  • Token Budget: Track LLM token usage
  • Time Budget: Track elapsed time
  • Iteration Budget: Track iteration count

If any budget is exceeded, execution routes to exit node.

Error Handling

Errors are handled at multiple levels:

  1. Node Level: Catch errors in individual node execution
  2. Graph Level: Handle errors during graph traversal
  3. State Level: Rollback state changes on error

Errors are logged and yield error events for UI.

Backward Compatibility

Graph execution is optional via feature flag:

  • USE_GRAPH_EXECUTION=true: Use graph-based execution
  • USE_GRAPH_EXECUTION=false: Use agent chain execution (existing)

This allows gradual migration and fallback if needed.

See Also

\ No newline at end of file diff --git a/site/architecture/middleware/index.html b/site/architecture/middleware/index.html index 69cae34..928003f 100644 --- a/site/architecture/middleware/index.html +++ b/site/architecture/middleware/index.html @@ -1,20 +1,40 @@ - Middleware - The DETERMINATOR
Skip to content

Middleware Architecture

DeepCritical uses middleware for state management, budget tracking, and workflow coordination.

State Management

WorkflowState

File: src/middleware/state_machine.py

Purpose: Thread-safe state management for research workflows

Implementation: Uses ContextVar for thread-safe isolation

State Components: - evidence: list[Evidence]: Collected evidence from searches - conversation: Conversation: Iteration history (gaps, tool calls, findings, thoughts) - embedding_service: Any: Embedding service for semantic search

Methods: - add_evidence(evidence: Evidence): Adds evidence with URL-based deduplication - async search_related(query: str, top_k: int = 5) -> list[Evidence]: Semantic search

Initialization:

Access:

Workflow Manager

File: src/middleware/workflow_manager.py

Purpose: Coordinates parallel research loops

Methods: - add_loop(loop: ResearchLoop): Add a research loop to manage - async run_loops_parallel() -> list[ResearchLoop]: Run all loops in parallel - update_loop_status(loop_id: str, status: str): Update loop status - sync_loop_evidence_to_state(): Synchronize evidence from loops to global state

Features: - Uses asyncio.gather() for parallel execution - Handles errors per loop (doesn't fail all if one fails) - Tracks loop status: pending, running, completed, failed, cancelled - Evidence deduplication across parallel loops

Usage: 

from src.middleware.workflow_manager import WorkflowManager
-
-manager = WorkflowManager()
-manager.add_loop(loop1)
-manager.add_loop(loop2)
-completed_loops = await manager.run_loops_parallel()
-

Budget Tracker

File: src/middleware/budget_tracker.py

Purpose: Tracks and enforces resource limits

Budget Components: - Tokens: LLM token usage - Time: Elapsed time in seconds - Iterations: Number of iterations

Methods: - create_budget(token_limit, time_limit_seconds, iterations_limit) -> BudgetStatus - add_tokens(tokens: int): Add token usage - start_timer(): Start time tracking - update_timer(): Update elapsed time - increment_iteration(): Increment iteration count - check_budget() -> BudgetStatus: Check current budget status - can_continue() -> bool: Check if research can continue

Token Estimation: - estimate_tokens(text: str) -> int: ~4 chars per token - estimate_llm_call_tokens(prompt: str, response: str) -> int: Estimate LLM call tokens

Usage: 

from src.middleware.budget_tracker import BudgetTracker
-
-tracker = BudgetTracker()
-budget = tracker.create_budget(
-    token_limit=100000,
-    time_limit_seconds=600,
-    iterations_limit=10
-)
-tracker.start_timer()
-# ... research operations ...
-if not tracker.can_continue():
-    # Budget exceeded, stop research
-    pass
-

Models

All middleware models are defined in src/utils/models.py:

  • IterationData: Data for a single iteration
  • Conversation: Conversation history with iterations
  • ResearchLoop: Research loop state and configuration
  • BudgetStatus: Current budget status

Thread Safety

All middleware components use ContextVar for thread-safe isolation:

  • Each request/thread has its own workflow state
  • No global mutable state
  • Safe for concurrent requests

See Also

\ No newline at end of file + Middleware - The DETERMINATOR
Skip to content

Middleware Architecture

DeepCritical uses middleware for state management, budget tracking, and workflow coordination.

State Management

WorkflowState

File: src/middleware/state_machine.py

Purpose: Thread-safe state management for research workflows

Implementation: Uses ContextVar for thread-safe isolation

State Components: - evidence: list[Evidence]: Collected evidence from searches - conversation: Conversation: Iteration history (gaps, tool calls, findings, thoughts) - embedding_service: Any: Embedding service for semantic search

Methods: - add_evidence(new_evidence: list[Evidence]) -> int: Adds evidence with URL-based deduplication. Returns the number of new items added (excluding duplicates). - async search_related(query: str, n_results: int = 5) -> list[Evidence]: Semantic search for related evidence using embedding service

Initialization:

Access:

Workflow Manager

File: src/middleware/workflow_manager.py

Purpose: Coordinates parallel research loops

Methods: - async add_loop(loop_id: str, query: str) -> ResearchLoop: Add a new research loop to manage - async run_loops_parallel(loop_configs: list[dict], loop_func: Callable, judge_handler: Any | None = None, budget_tracker: Any | None = None) -> list[Any]: Run multiple research loops in parallel. Takes configuration dicts and a loop function. - async update_loop_status(loop_id: str, status: LoopStatus, error: str | None = None): Update loop status - async sync_loop_evidence_to_state(loop_id: str): Synchronize evidence from a specific loop to global state

Features: - Uses asyncio.gather() for parallel execution - Handles errors per loop (doesn't fail all if one fails) - Tracks loop status: pending, running, completed, failed, cancelled - Evidence deduplication across parallel loops

Usage: 

from src.middleware.workflow_manager import WorkflowManager
+
+manager = WorkflowManager()
+await manager.add_loop("loop1", "Research query 1")
+await manager.add_loop("loop2", "Research query 2")
+
+async def run_research(config: dict) -> str:
+    loop_id = config["loop_id"]
+    query = config["query"]
+    # ... research logic ...
+    return "report"
+
+results = await manager.run_loops_parallel(
+    loop_configs=[
+        {"loop_id": "loop1", "query": "Research query 1"},
+        {"loop_id": "loop2", "query": "Research query 2"},
+    ],
+    loop_func=run_research,
+)
+

Budget Tracker

File: src/middleware/budget_tracker.py

Purpose: Tracks and enforces resource limits

Budget Components: - Tokens: LLM token usage - Time: Elapsed time in seconds - Iterations: Number of iterations

Methods: - create_budget(loop_id: str, tokens_limit: int = 100000, time_limit_seconds: float = 600.0, iterations_limit: int = 10) -> BudgetStatus: Create a budget for a specific loop - add_tokens(loop_id: str, tokens: int): Add token usage to a loop's budget - start_timer(loop_id: str): Start time tracking for a loop - update_timer(loop_id: str): Update elapsed time for a loop - increment_iteration(loop_id: str): Increment iteration count for a loop - check_budget(loop_id: str) -> tuple[bool, str]: Check if a loop's budget has been exceeded. Returns (exceeded: bool, reason: str) - can_continue(loop_id: str) -> bool: Check if a loop can continue based on budget

Token Estimation: - estimate_tokens(text: str) -> int: ~4 chars per token - estimate_llm_call_tokens(prompt: str, response: str) -> int: Estimate LLM call tokens

Usage: 

from src.middleware.budget_tracker import BudgetTracker
+
+tracker = BudgetTracker()
+budget = tracker.create_budget(
+    loop_id="research_loop",
+    tokens_limit=100000,
+    time_limit_seconds=600,
+    iterations_limit=10
+)
+tracker.start_timer("research_loop")
+# ... research operations ...
+tracker.add_tokens("research_loop", 5000)
+tracker.update_timer("research_loop")
+exceeded, reason = tracker.check_budget("research_loop")
+if exceeded:
+    # Budget exceeded, stop research
+    pass
+if not tracker.can_continue("research_loop"):
+    # Budget exceeded, stop research
+    pass
+

Models

All middleware models are defined in src/utils/models.py:

  • IterationData: Data for a single iteration
  • Conversation: Conversation history with iterations
  • ResearchLoop: Research loop state and configuration
  • BudgetStatus: Current budget status

Thread Safety

All middleware components use ContextVar for thread-safe isolation:

  • Each request/thread has its own workflow state
  • No global mutable state
  • Safe for concurrent requests

See Also

\ No newline at end of file diff --git a/site/architecture/orchestrators/index.html b/site/architecture/orchestrators/index.html index 2270d50..c02ded5 100644 --- a/site/architecture/orchestrators/index.html +++ b/site/architecture/orchestrators/index.html @@ -1 +1 @@ - Orchestrators - The DETERMINATOR
Skip to content

Orchestrators Architecture

DeepCritical supports multiple orchestration patterns for research workflows.

Research Flows

IterativeResearchFlow

File: src/orchestrator/research_flow.py

Pattern: Generate observations → Evaluate gaps → Select tools → Execute → Judge → Continue/Complete

Agents Used: - KnowledgeGapAgent: Evaluates research completeness - ToolSelectorAgent: Selects tools for addressing gaps - ThinkingAgent: Generates observations - WriterAgent: Creates final report - JudgeHandler: Assesses evidence sufficiency

Features: - Tracks iterations, time, budget - Supports graph execution (use_graph=True) and agent chains (use_graph=False) - Iterates until research complete or constraints met

Usage:

DeepResearchFlow

File: src/orchestrator/research_flow.py

Pattern: Planner → Parallel iterative loops per section → Synthesizer

Agents Used: - PlannerAgent: Breaks query into report sections - IterativeResearchFlow: Per-section research (parallel) - LongWriterAgent or ProofreaderAgent: Final synthesis

Features: - Uses WorkflowManager for parallel execution - Budget tracking per section and globally - State synchronization across parallel loops - Supports graph execution and agent chains

Usage:

Graph Orchestrator

File: src/orchestrator/graph_orchestrator.py

Purpose: Graph-based execution using Pydantic AI agents as nodes

Features: - Uses Pydantic AI Graphs (when available) or agent chains (fallback) - Routes based on research mode (iterative/deep/auto) - Streams AgentEvent objects for UI

Node Types: - Agent Nodes: Execute Pydantic AI agents - State Nodes: Update or read workflow state - Decision Nodes: Make routing decisions - Parallel Nodes: Execute multiple nodes concurrently

Edge Types: - Sequential Edges: Always traversed - Conditional Edges: Traversed based on condition - Parallel Edges: Used for parallel execution branches

Orchestrator Factory

File: src/orchestrator_factory.py

Purpose: Factory for creating orchestrators

Modes: - Simple: Legacy orchestrator (backward compatible) - Advanced: Magentic orchestrator (requires OpenAI API key) - Auto-detect: Chooses based on API key availability

Usage:

Magentic Orchestrator

File: src/orchestrator_magentic.py

Purpose: Multi-agent coordination using Microsoft Agent Framework

Features: - Uses agent-framework-core - ChatAgent pattern with internal LLMs per agent - MagenticBuilder with participants: searcher, hypothesizer, judge, reporter - Manager orchestrates agents via OpenAIChatClient - Requires OpenAI API key (function calling support) - Event-driven: converts Magentic events to AgentEvent for UI streaming

Requirements: - agent-framework-core package - OpenAI API key

Hierarchical Orchestrator

File: src/orchestrator_hierarchical.py

Purpose: Hierarchical orchestrator using middleware and sub-teams

Features: - Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge - Adapts Magentic ChatAgent to SubIterationTeam protocol - Event-driven via asyncio.Queue for coordination - Supports sub-iteration patterns for complex research tasks

Legacy Simple Mode

File: src/legacy_orchestrator.py

Purpose: Linear search-judge-synthesize loop

Features: - Uses SearchHandlerProtocol and JudgeHandlerProtocol - Generator-based design yielding AgentEvent objects - Backward compatibility for simple use cases

State Initialization

All orchestrators must initialize workflow state:

Event Streaming

All orchestrators yield AgentEvent objects:

Event Types: - started: Research started - search_complete: Search completed - judge_complete: Evidence evaluation completed - hypothesizing: Generating hypotheses - synthesizing: Synthesizing results - complete: Research completed - error: Error occurred

Event Structure:

See Also

\ No newline at end of file + Orchestrators - The DETERMINATOR
Skip to content

Orchestrators Architecture

DeepCritical supports multiple orchestration patterns for research workflows.

Research Flows

IterativeResearchFlow

File: src/orchestrator/research_flow.py

Pattern: Generate observations → Evaluate gaps → Select tools → Execute → Judge → Continue/Complete

Agents Used: - KnowledgeGapAgent: Evaluates research completeness - ToolSelectorAgent: Selects tools for addressing gaps - ThinkingAgent: Generates observations - WriterAgent: Creates final report - JudgeHandler: Assesses evidence sufficiency

Features: - Tracks iterations, time, budget - Supports graph execution (use_graph=True) and agent chains (use_graph=False) - Iterates until research complete or constraints met

Usage:

DeepResearchFlow

File: src/orchestrator/research_flow.py

Pattern: Planner → Parallel iterative loops per section → Synthesizer

Agents Used: - PlannerAgent: Breaks query into report sections - IterativeResearchFlow: Per-section research (parallel) - LongWriterAgent or ProofreaderAgent: Final synthesis

Features: - Uses WorkflowManager for parallel execution - Budget tracking per section and globally - State synchronization across parallel loops - Supports graph execution and agent chains

Usage:

Graph Orchestrator

File: src/orchestrator/graph_orchestrator.py

Purpose: Graph-based execution using Pydantic AI agents as nodes

Features: - Uses graph execution (use_graph=True) or agent chains (use_graph=False) as fallback - Routes based on research mode (iterative/deep/auto) - Streams AgentEvent objects for UI - Uses GraphExecutionContext to manage execution state

Node Types: - Agent Nodes: Execute Pydantic AI agents - State Nodes: Update or read workflow state - Decision Nodes: Make routing decisions - Parallel Nodes: Execute multiple nodes concurrently

Edge Types: - Sequential Edges: Always traversed - Conditional Edges: Traversed based on condition - Parallel Edges: Used for parallel execution branches

Special Node Handling:

The GraphOrchestrator has special handling for certain nodes:

  • execute_tools node: State node that uses search_handler to execute searches and add evidence to workflow state
  • parallel_loops node: Parallel node that executes IterativeResearchFlow instances for each section in deep research mode
  • synthesizer node: Agent node that calls LongWriterAgent.write_report() directly with ReportDraft instead of using agent.run()
  • writer node: Agent node that calls WriterAgent.write_report() directly with findings instead of using agent.run()

GraphExecutionContext:

The orchestrator uses GraphExecutionContext to manage execution state: - Tracks current node, visited nodes, and node results - Manages workflow state and budget tracker - Provides methods to store and retrieve node execution results

Orchestrator Factory

File: src/orchestrator_factory.py

Purpose: Factory for creating orchestrators

Modes: - Simple: Legacy orchestrator (backward compatible) - Advanced: Magentic orchestrator (requires OpenAI API key) - Auto-detect: Chooses based on API key availability

Usage:

Magentic Orchestrator

File: src/orchestrator_magentic.py

Purpose: Multi-agent coordination using Microsoft Agent Framework

Features: - Uses agent-framework-core - ChatAgent pattern with internal LLMs per agent - MagenticBuilder with participants: - searcher: SearchAgent (wraps SearchHandler) - hypothesizer: HypothesisAgent (generates hypotheses) - judge: JudgeAgent (evaluates evidence) - reporter: ReportAgent (generates final report) - Manager orchestrates agents via chat client (OpenAI or HuggingFace) - Event-driven: converts Magentic events to AgentEvent for UI streaming via _process_event() method - Supports max rounds, stall detection, and reset handling

Event Processing:

The orchestrator processes Magentic events and converts them to AgentEvent: - MagenticOrchestratorMessageEvent → AgentEvent with type based on message content - MagenticAgentMessageEvent → AgentEvent with type based on agent name - MagenticAgentDeltaEvent → AgentEvent for streaming updates - MagenticFinalResultEvent → AgentEvent with type "complete"

Requirements: - agent-framework-core package - OpenAI API key or HuggingFace authentication

Hierarchical Orchestrator

File: src/orchestrator_hierarchical.py

Purpose: Hierarchical orchestrator using middleware and sub-teams

Features: - Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge - Adapts Magentic ChatAgent to SubIterationTeam protocol - Event-driven via asyncio.Queue for coordination - Supports sub-iteration patterns for complex research tasks

Legacy Simple Mode

File: src/legacy_orchestrator.py

Purpose: Linear search-judge-synthesize loop

Features: - Uses SearchHandlerProtocol and JudgeHandlerProtocol - Generator-based design yielding AgentEvent objects - Backward compatibility for simple use cases

State Initialization

All orchestrators must initialize workflow state:

Event Streaming

All orchestrators yield AgentEvent objects:

Event Types: - started: Research started - searching: Search in progress - search_complete: Search completed - judging: Evidence evaluation in progress - judge_complete: Evidence evaluation completed - looping: Iteration in progress - hypothesizing: Generating hypotheses - analyzing: Statistical analysis in progress - analysis_complete: Statistical analysis completed - synthesizing: Synthesizing results - complete: Research completed - error: Error occurred - streaming: Streaming update (delta events)

Event Structure:

See Also

\ No newline at end of file diff --git a/site/architecture/services/index.html b/site/architecture/services/index.html index 44fe943..31b4851 100644 --- a/site/architecture/services/index.html +++ b/site/architecture/services/index.html @@ -1,29 +1,30 @@ - Services - The DETERMINATOR
Skip to content

Services Architecture

DeepCritical provides several services for embeddings, RAG, and statistical analysis.

Embedding Service

File: src/services/embeddings.py

Purpose: Local sentence-transformers for semantic search and deduplication

Features: - No API Key Required: Uses local sentence-transformers models - Async-Safe: All operations use run_in_executor() to avoid blocking - ChromaDB Storage: Vector storage for embeddings - Deduplication: 0.85 similarity threshold (85% similarity = duplicate)

Model: Configurable via settings.local_embedding_model (default: all-MiniLM-L6-v2)

Methods: - async def embed(text: str) -> list[float]: Generate embeddings - async def embed_batch(texts: list[str]) -> list[list[float]]: Batch embedding - async def similarity(text1: str, text2: str) -> float: Calculate similarity - async def find_duplicates(texts: list[str], threshold: float = 0.85) -> list[tuple[int, int]]: Find duplicates

Usage: 

from src.services.embeddings import get_embedding_service
-
-service = get_embedding_service()
-embedding = await service.embed("text to embed")
-

LlamaIndex RAG Service

File: src/services/rag.py

Purpose: Retrieval-Augmented Generation using LlamaIndex

Features: - OpenAI Embeddings: Requires OPENAI_API_KEY - ChromaDB Storage: Vector database for document storage - Metadata Preservation: Preserves source, title, URL, date, authors - Lazy Initialization: Graceful fallback if OpenAI key not available

Methods: - async def ingest_evidence(evidence: list[Evidence]) -> None: Ingest evidence into RAG - async def retrieve(query: str, top_k: int = 5) -> list[Document]: Retrieve relevant documents - async def query(query: str, top_k: int = 5) -> str: Query with RAG

Usage: 

from src.services.rag import get_rag_service
-
-service = get_rag_service()
-if service:
-    documents = await service.retrieve("query", top_k=5)
-

Statistical Analyzer

File: src/services/statistical_analyzer.py

Purpose: Secure execution of AI-generated statistical code

Features: - Modal Sandbox: Secure, isolated execution environment - Code Generation: Generates Python code via LLM - Library Pinning: Version-pinned libraries in SANDBOX_LIBRARIES - Network Isolation: block_network=True by default

Libraries Available: - pandas, numpy, scipy - matplotlib, scikit-learn - statsmodels

Output: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - code: Generated analysis code - output: Execution output - error: Error message if execution failed

Usage: 

from src.services.statistical_analyzer import StatisticalAnalyzer
-
-analyzer = StatisticalAnalyzer()
-result = await analyzer.analyze(
-    hypothesis="Metformin reduces cancer risk",
-    evidence=evidence_list
-)
-

Singleton Pattern

All services use the singleton pattern with @lru_cache(maxsize=1):

@lru_cache(maxsize=1)
-def get_embedding_service() -> EmbeddingService:
-    return EmbeddingService()
-

This ensures: - Single instance per process - Lazy initialization - No dependencies required at import time

Service Availability

Services check availability before use:

from src.utils.config import settings
-
-if settings.modal_available:
-    # Use Modal sandbox
-    pass
-
-if settings.has_openai_key:
-    # Use OpenAI embeddings for RAG
-    pass
-

See Also

\ No newline at end of file + Services - The DETERMINATOR
Skip to content

Services Architecture

DeepCritical provides several services for embeddings, RAG, and statistical analysis.

Embedding Service

File: src/services/embeddings.py

Purpose: Local sentence-transformers for semantic search and deduplication

Features: - No API Key Required: Uses local sentence-transformers models - Async-Safe: All operations use run_in_executor() to avoid blocking the event loop - ChromaDB Storage: In-memory vector storage for embeddings - Deduplication: 0.9 similarity threshold by default (90% similarity = duplicate, configurable)

Model: Configurable via settings.local_embedding_model (default: all-MiniLM-L6-v2)

Methods: - async def embed(text: str) -> list[float]: Generate embeddings (async-safe via run_in_executor()) - async def embed_batch(texts: list[str]) -> list[list[float]]: Batch embedding (more efficient) - async def add_evidence(evidence_id: str, content: str, metadata: dict[str, Any]) -> None: Add evidence to vector store - async def search_similar(query: str, n_results: int = 5) -> list[dict[str, Any]]: Find semantically similar evidence - async def deduplicate(new_evidence: list[Evidence], threshold: float = 0.9) -> list[Evidence]: Remove semantically duplicate evidence

Usage: 

from src.services.embeddings import get_embedding_service
+
+service = get_embedding_service()
+embedding = await service.embed("text to embed")
+

LlamaIndex RAG Service

File: src/services/llamaindex_rag.py

Purpose: Retrieval-Augmented Generation using LlamaIndex

Features: - Multiple Embedding Providers: OpenAI embeddings (requires OPENAI_API_KEY) or local sentence-transformers (no API key) - Multiple LLM Providers: HuggingFace LLM (preferred) or OpenAI LLM (fallback) for query synthesis - ChromaDB Storage: Vector database for document storage (supports in-memory mode) - Metadata Preservation: Preserves source, title, URL, date, authors - Lazy Initialization: Graceful fallback if dependencies not available

Initialization Parameters: - use_openai_embeddings: bool | None: Force OpenAI embeddings (None = auto-detect) - use_in_memory: bool: Use in-memory ChromaDB client (useful for tests) - oauth_token: str | None: Optional OAuth token from HuggingFace login (takes priority over env vars)

Methods: - async def ingest_evidence(evidence: list[Evidence]) -> None: Ingest evidence into RAG - async def retrieve(query: str, top_k: int = 5) -> list[Document]: Retrieve relevant documents - async def query(query: str, top_k: int = 5) -> str: Query with RAG

Usage: 

from src.services.llamaindex_rag import get_rag_service
+
+service = get_rag_service(
+    use_openai_embeddings=False,  # Use local embeddings
+    use_in_memory=True,  # Use in-memory ChromaDB
+    oauth_token=token  # Optional HuggingFace token
+)
+if service:
+    documents = await service.retrieve("query", top_k=5)
+

Statistical Analyzer

File: src/services/statistical_analyzer.py

Purpose: Secure execution of AI-generated statistical code

Features: - Modal Sandbox: Secure, isolated execution environment - Code Generation: Generates Python code via LLM - Library Pinning: Version-pinned libraries in SANDBOX_LIBRARIES - Network Isolation: block_network=True by default

Libraries Available: - pandas, numpy, scipy - matplotlib, scikit-learn - statsmodels

Output: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - code: Generated analysis code - output: Execution output - error: Error message if execution failed

Usage: 

from src.services.statistical_analyzer import StatisticalAnalyzer
+
+analyzer = StatisticalAnalyzer()
+result = await analyzer.analyze(
+    hypothesis="Metformin reduces cancer risk",
+    evidence=evidence_list
+)
+

Singleton Pattern

Services use singleton patterns for lazy initialization:

EmbeddingService: Uses a global variable pattern:

LlamaIndexRAGService: Direct instantiation (no caching):

This ensures: - Single instance per process - Lazy initialization - No dependencies required at import time

Service Availability

Services check availability before use:

from src.utils.config import settings
+
+if settings.modal_available:
+    # Use Modal sandbox
+    pass
+
+if settings.has_openai_key:
+    # Use OpenAI embeddings for RAG
+    pass
+

See Also

\ No newline at end of file diff --git a/site/architecture/tools/index.html b/site/architecture/tools/index.html index 89dded2..e535c5c 100644 --- a/site/architecture/tools/index.html +++ b/site/architecture/tools/index.html @@ -1,18 +1,19 @@ - Tools - The DETERMINATOR
Skip to content

Tools Architecture

DeepCritical implements a protocol-based search tool system for retrieving evidence from multiple sources.

SearchTool Protocol

All tools implement the SearchTool protocol from src/tools/base.py:

Rate Limiting

All tools use the @retry decorator from tenacity:

@retry(
-    stop=stop_after_attempt(3), 
-    wait=wait_exponential(...)
-)
-async def search(self, query: str, max_results: int = 10) -> list[Evidence]:
-    # Implementation
-

Tools with API rate limits implement _rate_limit() method and use shared rate limiters from src/tools/rate_limiter.py.

Error Handling

Tools raise custom exceptions:

  • SearchError: General search failures
  • RateLimitError: Rate limit exceeded

Tools handle HTTP errors (429, 500, timeout) and return empty lists on non-critical errors (with warning logs).

Query Preprocessing

Tools use preprocess_query() from src/tools/query_utils.py to:

  • Remove noise from queries
  • Expand synonyms
  • Normalize query format

Evidence Conversion

All tools convert API responses to Evidence objects with:

  • Citation: Title, URL, date, authors
  • content: Evidence text
  • relevance_score: 0.0-1.0 relevance score
  • metadata: Additional metadata

Missing fields are handled gracefully with defaults.

Tool Implementations

PubMed Tool

File: src/tools/pubmed.py

API: NCBI E-utilities (ESearch → EFetch)

Rate Limiting: - 0.34s between requests (3 req/sec without API key) - 0.1s between requests (10 req/sec with NCBI API key)

Features: - XML parsing with xmltodict - Handles single vs. multiple articles - Query preprocessing - Evidence conversion with metadata extraction

ClinicalTrials Tool

File: src/tools/clinicaltrials.py

API: ClinicalTrials.gov API v2

Important: Uses requests library (NOT httpx) because WAF blocks httpx TLS fingerprint.

Execution: Runs in thread pool: await asyncio.to_thread(requests.get, ...)

Filtering: - Only interventional studies - Status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION

Features: - Parses nested JSON structure - Extracts trial metadata - Evidence conversion

Europe PMC Tool

File: src/tools/europepmc.py

API: Europe PMC REST API

Features: - Handles preprint markers: [PREPRINT - Not peer-reviewed] - Builds URLs from DOI or PMID - Checks pubTypeList for preprint detection - Includes both preprints and peer-reviewed articles

RAG Tool

File: src/tools/rag_tool.py

Purpose: Semantic search within collected evidence

Implementation: Wraps LlamaIndexRAGService

Features: - Returns Evidence from RAG results - Handles evidence ingestion - Semantic similarity search - Metadata preservation

Search Handler

File: src/tools/search_handler.py

Purpose: Orchestrates parallel searches across multiple tools

Features: - Uses asyncio.gather() with return_exceptions=True - Aggregates results into SearchResult - Handles tool failures gracefully - Deduplicates results by URL

Tool Registration

Tools are registered in the search handler:

from src.tools.pubmed import PubMedTool
-from src.tools.clinicaltrials import ClinicalTrialsTool
-from src.tools.europepmc import EuropePMCTool
-
-search_handler = SearchHandler(
-    tools=[
-        PubMedTool(),
-        ClinicalTrialsTool(),
-        EuropePMCTool(),
-    ]
-)
-

See Also

\ No newline at end of file + Tools - The DETERMINATOR
Skip to content

Tools Architecture

DeepCritical implements a protocol-based search tool system for retrieving evidence from multiple sources.

SearchTool Protocol

All tools implement the SearchTool protocol from src/tools/base.py:

Rate Limiting

All tools use the @retry decorator from tenacity:

Tools with API rate limits implement _rate_limit() method and use shared rate limiters from src/tools/rate_limiter.py.

Error Handling

Tools raise custom exceptions:

  • SearchError: General search failures
  • RateLimitError: Rate limit exceeded

Tools handle HTTP errors (429, 500, timeout) and return empty lists on non-critical errors (with warning logs).

Query Preprocessing

Tools use preprocess_query() from src/tools/query_utils.py to:

  • Remove noise from queries
  • Expand synonyms
  • Normalize query format

Evidence Conversion

All tools convert API responses to Evidence objects with:

  • Citation: Title, URL, date, authors
  • content: Evidence text
  • relevance_score: 0.0-1.0 relevance score
  • metadata: Additional metadata

Missing fields are handled gracefully with defaults.

Tool Implementations

PubMed Tool

File: src/tools/pubmed.py

API: NCBI E-utilities (ESearch → EFetch)

Rate Limiting: - 0.34s between requests (3 req/sec without API key) - 0.1s between requests (10 req/sec with NCBI API key)

Features: - XML parsing with xmltodict - Handles single vs. multiple articles - Query preprocessing - Evidence conversion with metadata extraction

ClinicalTrials Tool

File: src/tools/clinicaltrials.py

API: ClinicalTrials.gov API v2

Important: Uses requests library (NOT httpx) because WAF blocks httpx TLS fingerprint.

Execution: Runs in thread pool: await asyncio.to_thread(requests.get, ...)

Filtering: - Only interventional studies - Status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION

Features: - Parses nested JSON structure - Extracts trial metadata - Evidence conversion

Europe PMC Tool

File: src/tools/europepmc.py

API: Europe PMC REST API

Features: - Handles preprint markers: [PREPRINT - Not peer-reviewed] - Builds URLs from DOI or PMID - Checks pubTypeList for preprint detection - Includes both preprints and peer-reviewed articles

RAG Tool

File: src/tools/rag_tool.py

Purpose: Semantic search within collected evidence

Implementation: Wraps LlamaIndexRAGService

Features: - Returns Evidence from RAG results - Handles evidence ingestion - Semantic similarity search - Metadata preservation

Search Handler

File: src/tools/search_handler.py

Purpose: Orchestrates parallel searches across multiple tools

Initialization Parameters: - tools: list[SearchTool]: List of search tools to use - timeout: float = 30.0: Timeout for each search in seconds - include_rag: bool = False: Whether to include RAG tool in searches - auto_ingest_to_rag: bool = True: Whether to automatically ingest results into RAG - oauth_token: str | None = None: Optional OAuth token from HuggingFace login (for RAG LLM)

Methods: - async def execute(query: str, max_results_per_tool: int = 10) -> SearchResult: Execute search across all tools in parallel

Features: - Uses asyncio.gather() with return_exceptions=True for parallel execution - Aggregates results into SearchResult with evidence and metadata - Handles tool failures gracefully (continues with other tools) - Deduplicates results by URL - Automatically ingests results into RAG if auto_ingest_to_rag=True - Can add RAG tool dynamically via add_rag_tool() method

Tool Registration

Tools are registered in the search handler:

from src.tools.pubmed import PubMedTool
+from src.tools.clinicaltrials import ClinicalTrialsTool
+from src.tools.europepmc import EuropePMCTool
+from src.tools.search_handler import SearchHandler
+
+search_handler = SearchHandler(
+    tools=[
+        PubMedTool(),
+        ClinicalTrialsTool(),
+        EuropePMCTool(),
+    ],
+    include_rag=True,  # Include RAG tool for semantic search
+    auto_ingest_to_rag=True,  # Automatically ingest results into RAG
+    oauth_token=token  # Optional HuggingFace token for RAG LLM
+)
+
+# Execute search
+result = await search_handler.execute("query", max_results_per_tool=10)
+

See Also

\ No newline at end of file diff --git a/site/architecture/workflow-diagrams/index.html b/site/architecture/workflow-diagrams/index.html index 17f6d1e..e9729db 100644 --- a/site/architecture/workflow-diagrams/index.html +++ b/site/architecture/workflow-diagrams/index.html @@ -1,4 +1,4 @@ - Workflow Diagrams - The DETERMINATOR
Skip to content

DeepCritical Workflow - Simplified Magentic Architecture

Architecture Pattern: Microsoft Magentic Orchestration Design Philosophy: Simple, dynamic, manager-driven coordination Key Innovation: Intelligent manager replaces rigid sequential phases

1. High-Level Magentic Workflow

flowchart TD
+ Workflow Diagrams - The DETERMINATOR      
 
 
  
 
  
 
 
 
 
  
 
 
 
 
 
DeepCritical Workflow - Simplified Magentic Architecture
 
 
Architecture Pattern: Microsoft Magentic Orchestration Design Philosophy: Simple, dynamic, manager-driven coordination Key Innovation: Intelligent manager replaces rigid sequential phases
 
 
 
1. High-Level Magentic Workflow
 
flowchart TD
     Start([User Query]) --> Manager[Magentic Manager<br/>Plan • Select • Assess • Adapt]
 
     Manager -->|Plans| Task1[Task Decomposition]
@@ -31,7 +31,7 @@
     style ReportAgent fill:#fff4e6
     style Decision fill:#ffd6d6
     style Synthesis fill:#d4edda
-    style Output fill:#e1f5e1
 
2. Magentic Manager: The 6-Phase Cycle
 
flowchart LR
+    style Output fill:#e1f5e1
 
2. Magentic Manager: The 6-Phase Cycle
 
flowchart LR
     P1[1. Planning<br/>Analyze task<br/>Create strategy] --> P2[2. Agent Selection<br/>Pick best agent<br/>for subtask]
     P2 --> P3[3. Execution<br/>Run selected<br/>agent with tools]
     P3 --> P4[4. Assessment<br/>Evaluate quality<br/>Check progress]
@@ -47,7 +47,7 @@
     style P4 fill:#ffd6d6
     style P5 fill:#fff3cd
     style P6 fill:#d4edda
-    style Done fill:#e1f5e1
 
3. Simplified Agent Architecture
 
graph TB
+    style Done fill:#e1f5e1
 
3. Simplified Agent Architecture
 
graph TB
     subgraph "Orchestration Layer"
         Manager[Magentic Manager<br/>• Plans workflow<br/>• Selects agents<br/>• Assesses quality<br/>• Adapts strategy]
         SharedContext[(Shared Context<br/>• Hypotheses<br/>• Search Results<br/>• Analysis<br/>• Progress)]
@@ -93,7 +93,7 @@
     style WebSearch fill:#e6f3ff
     style CodeExec fill:#e6f3ff
     style RAG fill:#e6f3ff
-    style Viz fill:#e6f3ff
 
4. Dynamic Workflow Example
 
sequenceDiagram
+    style Viz fill:#e6f3ff
 
4. Dynamic Workflow Example
 
sequenceDiagram
     participant User
     participant Manager
     participant HypAgent
@@ -129,7 +129,7 @@
     ReportAgent-->>Manager: Returns formatted report
     Note over Manager: SYNTHESIZE: Combine all results
 
-    Manager->>User: Final Research Report
 
5. Manager Decision Logic
 
flowchart TD
+    Manager->>User: Final Research Report
 
5. Manager Decision Logic
 
flowchart TD
     Start([Manager Receives Task]) --> Plan[Create Initial Plan]
 
     Plan --> Select[Select Agent for Next Subtask]
@@ -164,7 +164,7 @@
     style Q3 fill:#ffe6e6
     style Q4 fill:#ffe6e6
     style Synth fill:#d4edda
-    style Done fill:#e1f5e1
 
6. Hypothesis Agent Workflow
 
flowchart LR
+    style Done fill:#e1f5e1
 
6. Hypothesis Agent Workflow
 
flowchart LR
     Input[Research Query] --> Domain[Identify Domain<br/>& Key Concepts]
     Domain --> Context[Retrieve Background<br/>Knowledge]
     Context --> Generate[Generate 3-5<br/>Initial Hypotheses]
@@ -176,7 +176,7 @@
 
     style Input fill:#e1f5e1
     style Output fill:#fff4e6
-    style Struct fill:#e6f3ff
 
7. Search Agent Workflow
 
flowchart TD
+    style Struct fill:#e6f3ff
 
7. Search Agent Workflow
 
flowchart TD
     Input[Hypotheses] --> Strategy[Formulate Search<br/>Strategy per Hypothesis]
 
     Strategy --> Multi[Multi-Source Search]
@@ -199,7 +199,7 @@
     style Input fill:#fff4e6
     style Multi fill:#ffe6e6
     style Vector fill:#ffe6f0
-    style Output fill:#e6f3ff
 
8. Analysis Agent Workflow
 
flowchart TD
+    style Output fill:#e6f3ff
 
8. Analysis Agent Workflow
 
flowchart TD
     Input1[Hypotheses] --> Extract
     Input2[Search Results] --> Extract[Extract Evidence<br/>per Hypothesis]
 
@@ -224,7 +224,7 @@
     style Input1 fill:#fff4e6
     style Input2 fill:#e6f3ff
     style Execute fill:#ffe6e6
-    style Output fill:#e6ffe6
 
9. Report Agent Workflow
 
flowchart TD
+    style Output fill:#e6ffe6
 
9. Report Agent Workflow
 
flowchart TD
     Input1[Query] --> Assemble
     Input2[Hypotheses] --> Assemble
     Input3[Search Results] --> Assemble
@@ -264,7 +264,7 @@
     style Input2 fill:#fff4e6
     style Input3 fill:#e6f3ff
     style Input4 fill:#e6ffe6
-    style Output fill:#d4edda
 
10. Data Flow & Event Streaming
 
flowchart TD
+    style Output fill:#d4edda
 
10. Data Flow & Event Streaming
 
flowchart TD
     User[👤 User] -->|Research Query| UI[Gradio UI]
     UI -->|Submit| Manager[Magentic Manager]
 
@@ -303,7 +303,7 @@
     style Context fill:#ffe6f0
     style VectorDB fill:#ffe6f0
     style WebSearch fill:#f0f0f0
-    style CodeExec fill:#f0f0f0
 
11. MCP Tool Architecture
 
graph TB
+    style CodeExec fill:#f0f0f0
 
11. MCP Tool Architecture
 
graph TB
     subgraph "Agent Layer"
         Manager[Magentic Manager]
         HypAgent[Hypothesis Agent]
@@ -351,7 +351,7 @@
     style Server1 fill:#e6f3ff
     style Server2 fill:#e6f3ff
     style Server3 fill:#e6f3ff
-    style Server4 fill:#e6f3ff
 
12. Progress Tracking & Stall Detection
 
stateDiagram-v2
+    style Server4 fill:#e6f3ff
 
12. Progress Tracking & Stall Detection
 
stateDiagram-v2
     [*] --> Initialization: User Query
 
     Initialization --> Planning: Manager starts
@@ -391,7 +391,7 @@
         Stall = no new progress
         after agent execution
         Triggers plan reset
-    end note
 
13. Gradio UI Integration
 
graph TD
+    end note
 
13. Gradio UI Integration
 
graph TD
     App[Gradio App<br/>DeepCritical Research Agent]
 
     App --> Input[Input Section]
@@ -424,7 +424,7 @@
     style Input fill:#fff4e6
     style Status fill:#e6f3ff
     style Output fill:#e6ffe6
-    style Workflow fill:#ffe6e6
 
14. Complete System Context
 
graph LR
+    style Workflow fill:#ffe6e6
 
14. Complete System Context
 
graph LR
     User[👤 Researcher<br/>Asks research questions] -->|Submits query| DC[DeepCritical<br/>Magentic Workflow]
 
     DC -->|Literature search| PubMed[PubMed API<br/>Medical papers]
@@ -453,7 +453,7 @@
     style Claude fill:#ffd6d6
     style Modal fill:#f0f0f0
     style Chroma fill:#ffe6f0
-    style HF fill:#d4edda
 
15. Workflow Timeline (Simplified)
 
gantt
+    style HF fill:#d4edda
 
15. Workflow Timeline (Simplified)
 
gantt
     title DeepCritical Magentic Workflow - Typical Execution
     dateFormat mm:ss
     axisFormat %M:%S
@@ -485,19 +485,4 @@
     Formatting             :r3, after r2, 10s
 
     section Manager Synthesis
-    Final synthesis        :f1, after r3, 10s
 
 
Key Differences from Original Design
 
  
 Aspect Original (Judge-in-Loop) New (Magentic) 
   
 Control Flow Fixed sequential phases Dynamic agent selection 
 
 Quality Control Separate Judge Agent Manager assessment built-in 
 
 Retry Logic Phase-level with feedback Agent-level with adaptation 
 
 Flexibility Rigid 4-phase pipeline Adaptive workflow 
 
 Complexity 5 agents (including Judge) 4 agents (no Judge) 
 
 Progress Tracking Manual state management Built-in round/stall detection 
 
 Agent Coordination Sequential handoff Manager-driven dynamic selection 
 
 Error Recovery Retry same phase Try different agent or replan 
  
 
 
Simplified Design Principles
 
     
  1. Manager is Intelligent: LLM-powered manager handles planning, selection, and quality assessment
    2.  
  2. No Separate Judge: Manager's assessment phase replaces dedicated Judge Agent
    3.  
  3. Dynamic Workflow: Agents can be called multiple times in any order based on need
    4.  
  4. Built-in Safety: max_round_count (15) and max_stall_count (3) prevent infinite loops
    5.  
  5. Event-Driven UI: Real-time streaming updates to Gradio interface
    6.  
  6. MCP-Powered Tools: All external capabilities via Model Context Protocol
    7.  
  7. Shared Context: Centralized state accessible to all agents
    8.  
  8. Progress Awareness: Manager tracks what's been done and what's needed
    9.  
 
 
Legend
 
     
  • 🔴 Red/Pink: Manager, orchestration, decision-making
    •  
  • 🟡 Yellow/Orange: Specialist agents, processing
    •  
  • 🔵 Blue: Data, tools, MCP services
    •  
  • 🟣 Purple/Pink: Storage, databases, state
    •  
  • 🟢 Green: User interactions, final outputs
    •  
  • ⚪ Gray: External services, APIs
    •  
 
 
Implementation Highlights
 
Simple 4-Agent Setup: 
workflow = (
-    MagenticBuilder()
-    .participants(
-        hypothesis=HypothesisAgent(tools=[background_tool]),
-        search=SearchAgent(tools=[web_search, rag_tool]),
-        analysis=AnalysisAgent(tools=[code_execution]),
-        report=ReportAgent(tools=[code_execution, visualization])
-    )
-    .with_standard_manager(
-        chat_client=AnthropicClient(model="claude-sonnet-4"),
-        max_round_count=15,    # Prevent infinite loops
-        max_stall_count=3      # Detect stuck workflows
-    )
-    .build()
-)
-
 
Manager handles quality assessment in its instructions: - Checks hypothesis quality (testable, novel, clear) - Validates search results (relevant, authoritative, recent) - Assesses analysis soundness (methodology, evidence, conclusions) - Ensures report completeness (all sections, proper citations)
 
No separate Judge Agent needed - manager does it all!
 
 
Document Version: 2.0 (Magentic Simplified) Last Updated: 2025-11-24 Architecture: Microsoft Magentic Orchestration Pattern Agents: 4 (Hypothesis, Search, Analysis, Report) + 1 Manager License: MIT
 
See Also
  
 
  
  
  
 
 
 
    
\ No newline at end of file
+    Final synthesis        :f1, after r3, 10s

Key Differences from Original Design

Aspect Original (Judge-in-Loop) New (Magentic)
Control Flow Fixed sequential phases Dynamic agent selection
Quality Control Separate Judge Agent Manager assessment built-in
Retry Logic Phase-level with feedback Agent-level with adaptation
Flexibility Rigid 4-phase pipeline Adaptive workflow
Complexity 5 agents (including Judge) 4 agents (no Judge)
Progress Tracking Manual state management Built-in round/stall detection
Agent Coordination Sequential handoff Manager-driven dynamic selection
Error Recovery Retry same phase Try different agent or replan

Simplified Design Principles

  1. Manager is Intelligent: LLM-powered manager handles planning, selection, and quality assessment
  2. No Separate Judge: Manager's assessment phase replaces dedicated Judge Agent
  3. Dynamic Workflow: Agents can be called multiple times in any order based on need
  4. Built-in Safety: max_round_count (15) and max_stall_count (3) prevent infinite loops
  5. Event-Driven UI: Real-time streaming updates to Gradio interface
  6. MCP-Powered Tools: All external capabilities via Model Context Protocol
  7. Shared Context: Centralized state accessible to all agents
  8. Progress Awareness: Manager tracks what's been done and what's needed

Legend

  • 🔴 Red/Pink: Manager, orchestration, decision-making
  • 🟡 Yellow/Orange: Specialist agents, processing
  • 🔵 Blue: Data, tools, MCP services
  • 🟣 Purple/Pink: Storage, databases, state
  • 🟢 Green: User interactions, final outputs
  • ⚪ Gray: External services, APIs

Implementation Highlights

Simple 4-Agent Setup:

Manager handles quality assessment in its instructions: - Checks hypothesis quality (testable, novel, clear) - Validates search results (relevant, authoritative, recent) - Assesses analysis soundness (methodology, evidence, conclusions) - Ensures report completeness (all sections, proper citations)

No separate Judge Agent needed - manager does it all!

Document Version: 2.0 (Magentic Simplified) Last Updated: 2025-11-24 Architecture: Microsoft Magentic Orchestration Pattern Agents: 4 (Hypothesis, Search, Analysis, Report) + 1 Manager License: MIT

See Also

\ No newline at end of file diff --git a/site/architecture/workflows/index.html b/site/architecture/workflows/index.html deleted file mode 100644 index 6f7acb9..0000000 --- a/site/architecture/workflows/index.html +++ /dev/null @@ -1,503 +0,0 @@ - DeepCritical Workflow - Simplified Magentic Architecture - The DETERMINATOR
Skip to content

DeepCritical Workflow - Simplified Magentic Architecture

Architecture Pattern: Microsoft Magentic Orchestration Design Philosophy: Simple, dynamic, manager-driven coordination Key Innovation: Intelligent manager replaces rigid sequential phases

1. High-Level Magentic Workflow

flowchart TD
-    Start([User Query]) --> Manager[Magentic Manager<br/>Plan • Select • Assess • Adapt]
-
-    Manager -->|Plans| Task1[Task Decomposition]
-    Task1 --> Manager
-
-    Manager -->|Selects & Executes| HypAgent[Hypothesis Agent]
-    Manager -->|Selects & Executes| SearchAgent[Search Agent]
-    Manager -->|Selects & Executes| AnalysisAgent[Analysis Agent]
-    Manager -->|Selects & Executes| ReportAgent[Report Agent]
-
-    HypAgent -->|Results| Manager
-    SearchAgent -->|Results| Manager
-    AnalysisAgent -->|Results| Manager
-    ReportAgent -->|Results| Manager
-
-    Manager -->|Assesses Quality| Decision{Good Enough?}
-    Decision -->|No - Refine| Manager
-    Decision -->|No - Different Agent| Manager
-    Decision -->|No - Stalled| Replan[Reset Plan]
-    Replan --> Manager
-
-    Decision -->|Yes| Synthesis[Synthesize Final Result]
-    Synthesis --> Output([Research Report])
-
-    style Start fill:#e1f5e1
-    style Manager fill:#ffe6e6
-    style HypAgent fill:#fff4e6
-    style SearchAgent fill:#fff4e6
-    style AnalysisAgent fill:#fff4e6
-    style ReportAgent fill:#fff4e6
-    style Decision fill:#ffd6d6
-    style Synthesis fill:#d4edda
-    style Output fill:#e1f5e1

2. Magentic Manager: The 6-Phase Cycle

flowchart LR
-    P1[1. Planning<br/>Analyze task<br/>Create strategy] --> P2[2. Agent Selection<br/>Pick best agent<br/>for subtask]
-    P2 --> P3[3. Execution<br/>Run selected<br/>agent with tools]
-    P3 --> P4[4. Assessment<br/>Evaluate quality<br/>Check progress]
-    P4 --> Decision{Quality OK?<br/>Progress made?}
-    Decision -->|Yes| P6[6. Synthesis<br/>Combine results<br/>Generate report]
-    Decision -->|No| P5[5. Iteration<br/>Adjust plan<br/>Try again]
-    P5 --> P2
-    P6 --> Done([Complete])
-
-    style P1 fill:#fff4e6
-    style P2 fill:#ffe6e6
-    style P3 fill:#e6f3ff
-    style P4 fill:#ffd6d6
-    style P5 fill:#fff3cd
-    style P6 fill:#d4edda
-    style Done fill:#e1f5e1

3. Simplified Agent Architecture

graph TB
-    subgraph "Orchestration Layer"
-        Manager[Magentic Manager<br/>• Plans workflow<br/>• Selects agents<br/>• Assesses quality<br/>• Adapts strategy]
-        SharedContext[(Shared Context<br/>• Hypotheses<br/>• Search Results<br/>• Analysis<br/>• Progress)]
-        Manager <--> SharedContext
-    end
-
-    subgraph "Specialist Agents"
-        HypAgent[Hypothesis Agent<br/>• Domain understanding<br/>• Hypothesis generation<br/>• Testability refinement]
-        SearchAgent[Search Agent<br/>• Multi-source search<br/>• RAG retrieval<br/>• Result ranking]
-        AnalysisAgent[Analysis Agent<br/>• Evidence extraction<br/>• Statistical analysis<br/>• Code execution]
-        ReportAgent[Report Agent<br/>• Report assembly<br/>• Visualization<br/>• Citation formatting]
-    end
-
-    subgraph "MCP Tools"
-        WebSearch[Web Search<br/>PubMed • arXiv • bioRxiv]
-        CodeExec[Code Execution<br/>Sandboxed Python]
-        RAG[RAG Retrieval<br/>Vector DB • Embeddings]
-        Viz[Visualization<br/>Charts • Graphs]
-    end
-
-    Manager -->|Selects & Directs| HypAgent
-    Manager -->|Selects & Directs| SearchAgent
-    Manager -->|Selects & Directs| AnalysisAgent
-    Manager -->|Selects & Directs| ReportAgent
-
-    HypAgent --> SharedContext
-    SearchAgent --> SharedContext
-    AnalysisAgent --> SharedContext
-    ReportAgent --> SharedContext
-
-    SearchAgent --> WebSearch
-    SearchAgent --> RAG
-    AnalysisAgent --> CodeExec
-    ReportAgent --> CodeExec
-    ReportAgent --> Viz
-
-    style Manager fill:#ffe6e6
-    style SharedContext fill:#ffe6f0
-    style HypAgent fill:#fff4e6
-    style SearchAgent fill:#fff4e6
-    style AnalysisAgent fill:#fff4e6
-    style ReportAgent fill:#fff4e6
-    style WebSearch fill:#e6f3ff
-    style CodeExec fill:#e6f3ff
-    style RAG fill:#e6f3ff
-    style Viz fill:#e6f3ff

4. Dynamic Workflow Example

sequenceDiagram
-    participant User
-    participant Manager
-    participant HypAgent
-    participant SearchAgent
-    participant AnalysisAgent
-    participant ReportAgent
-
-    User->>Manager: "Research protein folding in Alzheimer's"
-
-    Note over Manager: PLAN: Generate hypotheses → Search → Analyze → Report
-
-    Manager->>HypAgent: Generate 3 hypotheses
-    HypAgent-->>Manager: Returns 3 hypotheses
-    Note over Manager: ASSESS: Good quality, proceed
-
-    Manager->>SearchAgent: Search literature for hypothesis 1
-    SearchAgent-->>Manager: Returns 15 papers
-    Note over Manager: ASSESS: Good results, continue
-
-    Manager->>SearchAgent: Search for hypothesis 2
-    SearchAgent-->>Manager: Only 2 papers found
-    Note over Manager: ASSESS: Insufficient, refine search
-
-    Manager->>SearchAgent: Refined query for hypothesis 2
-    SearchAgent-->>Manager: Returns 12 papers
-    Note over Manager: ASSESS: Better, proceed
-
-    Manager->>AnalysisAgent: Analyze evidence for all hypotheses
-    AnalysisAgent-->>Manager: Returns analysis with code
-    Note over Manager: ASSESS: Complete, generate report
-
-    Manager->>ReportAgent: Create comprehensive report
-    ReportAgent-->>Manager: Returns formatted report
-    Note over Manager: SYNTHESIZE: Combine all results
-
-    Manager->>User: Final Research Report

5. Manager Decision Logic

flowchart TD
-    Start([Manager Receives Task]) --> Plan[Create Initial Plan]
-
-    Plan --> Select[Select Agent for Next Subtask]
-    Select --> Execute[Execute Agent]
-    Execute --> Collect[Collect Results]
-
-    Collect --> Assess[Assess Quality & Progress]
-
-    Assess --> Q1{Quality Sufficient?}
-    Q1 -->|No| Q2{Same Agent Can Fix?}
-    Q2 -->|Yes| Feedback[Provide Specific Feedback]
-    Feedback --> Execute
-    Q2 -->|No| Different[Try Different Agent]
-    Different --> Select
-
-    Q1 -->|Yes| Q3{Task Complete?}
-    Q3 -->|No| Q4{Making Progress?}
-    Q4 -->|Yes| Select
-    Q4 -->|No - Stalled| Replan[Reset Plan & Approach]
-    Replan --> Plan
-
-    Q3 -->|Yes| Synth[Synthesize Final Result]
-    Synth --> Done([Return Report])
-
-    style Start fill:#e1f5e1
-    style Plan fill:#fff4e6
-    style Select fill:#ffe6e6
-    style Execute fill:#e6f3ff
-    style Assess fill:#ffd6d6
-    style Q1 fill:#ffe6e6
-    style Q2 fill:#ffe6e6
-    style Q3 fill:#ffe6e6
-    style Q4 fill:#ffe6e6
-    style Synth fill:#d4edda
-    style Done fill:#e1f5e1

6. Hypothesis Agent Workflow

flowchart LR
-    Input[Research Query] --> Domain[Identify Domain<br/>& Key Concepts]
-    Domain --> Context[Retrieve Background<br/>Knowledge]
-    Context --> Generate[Generate 3-5<br/>Initial Hypotheses]
-    Generate --> Refine[Refine for<br/>Testability]
-    Refine --> Rank[Rank by<br/>Quality Score]
-    Rank --> Output[Return Top<br/>Hypotheses]
-
-    Output --> Struct[Hypothesis Structure:<br/>• Statement<br/>• Rationale<br/>• Testability Score<br/>• Data Requirements<br/>• Expected Outcomes]
-
-    style Input fill:#e1f5e1
-    style Output fill:#fff4e6
-    style Struct fill:#e6f3ff

7. Search Agent Workflow

flowchart TD
-    Input[Hypotheses] --> Strategy[Formulate Search<br/>Strategy per Hypothesis]
-
-    Strategy --> Multi[Multi-Source Search]
-
-    Multi --> PubMed[PubMed Search<br/>via MCP]
-    Multi --> ArXiv[arXiv Search<br/>via MCP]
-    Multi --> BioRxiv[bioRxiv Search<br/>via MCP]
-
-    PubMed --> Aggregate[Aggregate Results]
-    ArXiv --> Aggregate
-    BioRxiv --> Aggregate
-
-    Aggregate --> Filter[Filter & Rank<br/>by Relevance]
-    Filter --> Dedup[Deduplicate<br/>Cross-Reference]
-    Dedup --> Embed[Embed Documents<br/>via MCP]
-    Embed --> Vector[(Vector DB)]
-    Vector --> RAGRetrieval[RAG Retrieval<br/>Top-K per Hypothesis]
-    RAGRetrieval --> Output[Return Contextualized<br/>Search Results]
-
-    style Input fill:#fff4e6
-    style Multi fill:#ffe6e6
-    style Vector fill:#ffe6f0
-    style Output fill:#e6f3ff

8. Analysis Agent Workflow

flowchart TD
-    Input1[Hypotheses] --> Extract
-    Input2[Search Results] --> Extract[Extract Evidence<br/>per Hypothesis]
-
-    Extract --> Methods[Determine Analysis<br/>Methods Needed]
-
-    Methods --> Branch{Requires<br/>Computation?}
-    Branch -->|Yes| GenCode[Generate Python<br/>Analysis Code]
-    Branch -->|No| Qual[Qualitative<br/>Synthesis]
-
-    GenCode --> Execute[Execute Code<br/>via MCP Sandbox]
-    Execute --> Interpret1[Interpret<br/>Results]
-    Qual --> Interpret2[Interpret<br/>Findings]
-
-    Interpret1 --> Synthesize[Synthesize Evidence<br/>Across Sources]
-    Interpret2 --> Synthesize
-
-    Synthesize --> Verdict[Determine Verdict<br/>per Hypothesis]
-    Verdict --> Support[• Supported<br/>• Refuted<br/>• Inconclusive]
-    Support --> Gaps[Identify Knowledge<br/>Gaps & Limitations]
-    Gaps --> Output[Return Analysis<br/>Report]
-
-    style Input1 fill:#fff4e6
-    style Input2 fill:#e6f3ff
-    style Execute fill:#ffe6e6
-    style Output fill:#e6ffe6

9. Report Agent Workflow

flowchart TD
-    Input1[Query] --> Assemble
-    Input2[Hypotheses] --> Assemble
-    Input3[Search Results] --> Assemble
-    Input4[Analysis] --> Assemble[Assemble Report<br/>Sections]
-
-    Assemble --> Exec[Executive Summary]
-    Assemble --> Intro[Introduction]
-    Assemble --> Methods[Methods]
-    Assemble --> Results[Results per<br/>Hypothesis]
-    Assemble --> Discussion[Discussion]
-    Assemble --> Future[Future Directions]
-    Assemble --> Refs[References]
-
-    Results --> VizCheck{Needs<br/>Visualization?}
-    VizCheck -->|Yes| GenViz[Generate Viz Code]
-    GenViz --> ExecViz[Execute via MCP<br/>Create Charts]
-    ExecViz --> Combine
-    VizCheck -->|No| Combine[Combine All<br/>Sections]
-
-    Exec --> Combine
-    Intro --> Combine
-    Methods --> Combine
-    Discussion --> Combine
-    Future --> Combine
-    Refs --> Combine
-
-    Combine --> Format[Format Output]
-    Format --> MD[Markdown]
-    Format --> PDF[PDF]
-    Format --> JSON[JSON]
-
-    MD --> Output[Return Final<br/>Report]
-    PDF --> Output
-    JSON --> Output
-
-    style Input1 fill:#e1f5e1
-    style Input2 fill:#fff4e6
-    style Input3 fill:#e6f3ff
-    style Input4 fill:#e6ffe6
-    style Output fill:#d4edda

10. Data Flow & Event Streaming

flowchart TD
-    User[👤 User] -->|Research Query| UI[Gradio UI]
-    UI -->|Submit| Manager[Magentic Manager]
-
-    Manager -->|Event: Planning| UI
-    Manager -->|Select Agent| HypAgent[Hypothesis Agent]
-    HypAgent -->|Event: Delta/Message| UI
-    HypAgent -->|Hypotheses| Context[(Shared Context)]
-
-    Context -->|Retrieved by| Manager
-    Manager -->|Select Agent| SearchAgent[Search Agent]
-    SearchAgent -->|MCP Request| WebSearch[Web Search Tool]
-    WebSearch -->|Results| SearchAgent
-    SearchAgent -->|Event: Delta/Message| UI
-    SearchAgent -->|Documents| Context
-    SearchAgent -->|Embeddings| VectorDB[(Vector DB)]
-
-    Context -->|Retrieved by| Manager
-    Manager -->|Select Agent| AnalysisAgent[Analysis Agent]
-    AnalysisAgent -->|MCP Request| CodeExec[Code Execution Tool]
-    CodeExec -->|Results| AnalysisAgent
-    AnalysisAgent -->|Event: Delta/Message| UI
-    AnalysisAgent -->|Analysis| Context
-
-    Context -->|Retrieved by| Manager
-    Manager -->|Select Agent| ReportAgent[Report Agent]
-    ReportAgent -->|MCP Request| CodeExec
-    ReportAgent -->|Event: Delta/Message| UI
-    ReportAgent -->|Report| Context
-
-    Manager -->|Event: Final Result| UI
-    UI -->|Display| User
-
-    style User fill:#e1f5e1
-    style UI fill:#e6f3ff
-    style Manager fill:#ffe6e6
-    style Context fill:#ffe6f0
-    style VectorDB fill:#ffe6f0
-    style WebSearch fill:#f0f0f0
-    style CodeExec fill:#f0f0f0

11. MCP Tool Architecture

graph TB
-    subgraph "Agent Layer"
-        Manager[Magentic Manager]
-        HypAgent[Hypothesis Agent]
-        SearchAgent[Search Agent]
-        AnalysisAgent[Analysis Agent]
-        ReportAgent[Report Agent]
-    end
-
-    subgraph "MCP Protocol Layer"
-        Registry[MCP Tool Registry<br/>• Discovers tools<br/>• Routes requests<br/>• Manages connections]
-    end
-
-    subgraph "MCP Servers"
-        Server1[Web Search Server<br/>localhost:8001<br/>• PubMed<br/>• arXiv<br/>• bioRxiv]
-        Server2[Code Execution Server<br/>localhost:8002<br/>• Sandboxed Python<br/>• Package management]
-        Server3[RAG Server<br/>localhost:8003<br/>• Vector embeddings<br/>• Similarity search]
-        Server4[Visualization Server<br/>localhost:8004<br/>• Chart generation<br/>• Plot rendering]
-    end
-
-    subgraph "External Services"
-        PubMed[PubMed API]
-        ArXiv[arXiv API]
-        BioRxiv[bioRxiv API]
-        Modal[Modal Sandbox]
-        ChromaDB[(ChromaDB)]
-    end
-
-    SearchAgent -->|Request| Registry
-    AnalysisAgent -->|Request| Registry
-    ReportAgent -->|Request| Registry
-
-    Registry --> Server1
-    Registry --> Server2
-    Registry --> Server3
-    Registry --> Server4
-
-    Server1 --> PubMed
-    Server1 --> ArXiv
-    Server1 --> BioRxiv
-    Server2 --> Modal
-    Server3 --> ChromaDB
-
-    style Manager fill:#ffe6e6
-    style Registry fill:#fff4e6
-    style Server1 fill:#e6f3ff
-    style Server2 fill:#e6f3ff
-    style Server3 fill:#e6f3ff
-    style Server4 fill:#e6f3ff

12. Progress Tracking & Stall Detection

stateDiagram-v2
-    [*] --> Initialization: User Query
-
-    Initialization --> Planning: Manager starts
-
-    Planning --> AgentExecution: Select agent
-
-    AgentExecution --> Assessment: Collect results
-
-    Assessment --> QualityCheck: Evaluate output
-
-    QualityCheck --> AgentExecution: Poor quality<br/>(retry < max_rounds)
-    QualityCheck --> Planning: Poor quality<br/>(try different agent)
-    QualityCheck --> NextAgent: Good quality<br/>(task incomplete)
-    QualityCheck --> Synthesis: Good quality<br/>(task complete)
-
-    NextAgent --> AgentExecution: Select next agent
-
-    state StallDetection <<choice>>
-    Assessment --> StallDetection: Check progress
-    StallDetection --> Planning: No progress<br/>(stall count < max)
-    StallDetection --> ErrorRecovery: No progress<br/>(max stalls reached)
-
-    ErrorRecovery --> PartialReport: Generate partial results
-    PartialReport --> [*]
-
-    Synthesis --> FinalReport: Combine all outputs
-    FinalReport --> [*]
-
-    note right of QualityCheck
-        Manager assesses:
-        • Output completeness
-        • Quality metrics
-        • Progress made
-    end note
-
-    note right of StallDetection
-        Stall = no new progress
-        after agent execution
-        Triggers plan reset
-    end note

13. Gradio UI Integration

graph TD
-    App[Gradio App<br/>DeepCritical Research Agent]
-
-    App --> Input[Input Section]
-    App --> Status[Status Section]
-    App --> Output[Output Section]
-
-    Input --> Query[Research Question<br/>Text Area]
-    Input --> Controls[Controls]
-    Controls --> MaxHyp[Max Hypotheses: 1-10]
-    Controls --> MaxRounds[Max Rounds: 5-20]
-    Controls --> Submit[Start Research Button]
-
-    Status --> Log[Real-time Event Log<br/>• Manager planning<br/>• Agent selection<br/>• Execution updates<br/>• Quality assessment]
-    Status --> Progress[Progress Tracker<br/>• Current agent<br/>• Round count<br/>• Stall count]
-
-    Output --> Tabs[Tabbed Results]
-    Tabs --> Tab1[Hypotheses Tab<br/>Generated hypotheses with scores]
-    Tabs --> Tab2[Search Results Tab<br/>Papers & sources found]
-    Tabs --> Tab3[Analysis Tab<br/>Evidence & verdicts]
-    Tabs --> Tab4[Report Tab<br/>Final research report]
-    Tab4 --> Download[Download Report<br/>MD / PDF / JSON]
-
-    Submit -.->|Triggers| Workflow[Magentic Workflow]
-    Workflow -.->|MagenticOrchestratorMessageEvent| Log
-    Workflow -.->|MagenticAgentDeltaEvent| Log
-    Workflow -.->|MagenticAgentMessageEvent| Log
-    Workflow -.->|MagenticFinalResultEvent| Tab4
-
-    style App fill:#e1f5e1
-    style Input fill:#fff4e6
-    style Status fill:#e6f3ff
-    style Output fill:#e6ffe6
-    style Workflow fill:#ffe6e6

14. Complete System Context

graph LR
-    User[👤 Researcher<br/>Asks research questions] -->|Submits query| DC[DeepCritical<br/>Magentic Workflow]
-
-    DC -->|Literature search| PubMed[PubMed API<br/>Medical papers]
-    DC -->|Preprint search| ArXiv[arXiv API<br/>Scientific preprints]
-    DC -->|Biology search| BioRxiv[bioRxiv API<br/>Biology preprints]
-    DC -->|Agent reasoning| Claude[Claude API<br/>Sonnet 4 / Opus]
-    DC -->|Code execution| Modal[Modal Sandbox<br/>Safe Python env]
-    DC -->|Vector storage| Chroma[ChromaDB<br/>Embeddings & RAG]
-
-    DC -->|Deployed on| HF[HuggingFace Spaces<br/>Gradio 6.0]
-
-    PubMed -->|Results| DC
-    ArXiv -->|Results| DC
-    BioRxiv -->|Results| DC
-    Claude -->|Responses| DC
-    Modal -->|Output| DC
-    Chroma -->|Context| DC
-
-    DC -->|Research report| User
-
-    style User fill:#e1f5e1
-    style DC fill:#ffe6e6
-    style PubMed fill:#e6f3ff
-    style ArXiv fill:#e6f3ff
-    style BioRxiv fill:#e6f3ff
-    style Claude fill:#ffd6d6
-    style Modal fill:#f0f0f0
-    style Chroma fill:#ffe6f0
-    style HF fill:#d4edda

15. Workflow Timeline (Simplified)

gantt
-    title DeepCritical Magentic Workflow - Typical Execution
-    dateFormat mm:ss
-    axisFormat %M:%S
-
-    section Manager Planning
-    Initial planning         :p1, 00:00, 10s
-
-    section Hypothesis Agent
-    Generate hypotheses      :h1, after p1, 30s
-    Manager assessment       :h2, after h1, 5s
-
-    section Search Agent
-    Search hypothesis 1      :s1, after h2, 20s
-    Search hypothesis 2      :s2, after s1, 20s
-    Search hypothesis 3      :s3, after s2, 20s
-    RAG processing          :s4, after s3, 15s
-    Manager assessment      :s5, after s4, 5s
-
-    section Analysis Agent
-    Evidence extraction     :a1, after s5, 15s
-    Code generation        :a2, after a1, 20s
-    Code execution         :a3, after a2, 25s
-    Synthesis              :a4, after a3, 20s
-    Manager assessment     :a5, after a4, 5s
-
-    section Report Agent
-    Report assembly        :r1, after a5, 30s
-    Visualization          :r2, after r1, 15s
-    Formatting             :r3, after r2, 10s
-
-    section Manager Synthesis
-    Final synthesis        :f1, after r3, 10s

Key Differences from Original Design

Aspect Original (Judge-in-Loop) New (Magentic)
Control Flow Fixed sequential phases Dynamic agent selection
Quality Control Separate Judge Agent Manager assessment built-in
Retry Logic Phase-level with feedback Agent-level with adaptation
Flexibility Rigid 4-phase pipeline Adaptive workflow
Complexity 5 agents (including Judge) 4 agents (no Judge)
Progress Tracking Manual state management Built-in round/stall detection
Agent Coordination Sequential handoff Manager-driven dynamic selection
Error Recovery Retry same phase Try different agent or replan

Simplified Design Principles

  1. Manager is Intelligent: LLM-powered manager handles planning, selection, and quality assessment
  2. No Separate Judge: Manager's assessment phase replaces dedicated Judge Agent
  3. Dynamic Workflow: Agents can be called multiple times in any order based on need
  4. Built-in Safety: max_round_count (15) and max_stall_count (3) prevent infinite loops
  5. Event-Driven UI: Real-time streaming updates to Gradio interface
  6. MCP-Powered Tools: All external capabilities via Model Context Protocol
  7. Shared Context: Centralized state accessible to all agents
  8. Progress Awareness: Manager tracks what's been done and what's needed

Legend

  • 🔴 Red/Pink: Manager, orchestration, decision-making
  • 🟡 Yellow/Orange: Specialist agents, processing
  • 🔵 Blue: Data, tools, MCP services
  • 🟣 Purple/Pink: Storage, databases, state
  • 🟢 Green: User interactions, final outputs
  • ⚪ Gray: External services, APIs

Implementation Highlights

Simple 4-Agent Setup: 

workflow = (
-    MagenticBuilder()
-    .participants(
-        hypothesis=HypothesisAgent(tools=[background_tool]),
-        search=SearchAgent(tools=[web_search, rag_tool]),
-        analysis=AnalysisAgent(tools=[code_execution]),
-        report=ReportAgent(tools=[code_execution, visualization])
-    )
-    .with_standard_manager(
-        chat_client=AnthropicClient(model="claude-sonnet-4"),
-        max_round_count=15,    # Prevent infinite loops
-        max_stall_count=3      # Detect stuck workflows
-    )
-    .build()
-)
-

Manager handles quality assessment in its instructions: - Checks hypothesis quality (testable, novel, clear) - Validates search results (relevant, authoritative, recent) - Assesses analysis soundness (methodology, evidence, conclusions) - Ensures report completeness (all sections, proper citations)

No separate Judge Agent needed - manager does it all!

Document Version: 2.0 (Magentic Simplified) Last Updated: 2025-11-24 Architecture: Microsoft Magentic Orchestration Pattern Agents: 4 (Hypothesis, Search, Analysis, Report) + 1 Manager License: MIT

\ No newline at end of file diff --git a/site/configuration/CONFIGURATION/index.html b/site/configuration/CONFIGURATION/index.html deleted file mode 100644 index 502999c..0000000 --- a/site/configuration/CONFIGURATION/index.html +++ /dev/null @@ -1,130 +0,0 @@ - Configuration Guide - The DETERMINATOR
Skip to content

Configuration Guide

Overview

DeepCritical uses Pydantic Settings for centralized configuration management. All settings are defined in the Settings class in src/utils/config.py and can be configured via environment variables or a .env file.

The configuration system provides:

  • Type Safety: Strongly-typed fields with Pydantic validation
  • Environment File Support: Automatically loads from .env file (if present)
  • Case-Insensitive: Environment variables are case-insensitive
  • Singleton Pattern: Global settings instance for easy access throughout the codebase
  • Validation: Automatic validation on load with helpful error messages

Quick Start

  1. Create a .env file in the project root
  2. Set at least one LLM API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, or HF_TOKEN)
  3. Optionally configure other services as needed
  4. The application will automatically load and validate your configuration

Configuration System Architecture

Settings Class

The Settings class extends BaseSettings from pydantic_settings and defines all application configuration:

Singleton Instance

A global settings instance is available for import:

Usage Pattern

Access configuration throughout the codebase:

from src.utils.config import settings
-
-# Check if API keys are available
-if settings.has_openai_key:
-    # Use OpenAI
-    pass
-
-# Access configuration values
-max_iterations = settings.max_iterations
-web_search_provider = settings.web_search_provider
-

Required Configuration

LLM Provider

You must configure at least one LLM provider. The system supports:

  • OpenAI: Requires OPENAI_API_KEY
  • Anthropic: Requires ANTHROPIC_API_KEY
  • HuggingFace: Optional HF_TOKEN or HUGGINGFACE_API_KEY (can work without key for public models)

OpenAI Configuration

LLM_PROVIDER=openai
-OPENAI_API_KEY=your_openai_api_key_here
-OPENAI_MODEL=gpt-5.1
-

The default model is defined in the Settings class:

Anthropic Configuration

LLM_PROVIDER=anthropic
-ANTHROPIC_API_KEY=your_anthropic_api_key_here
-ANTHROPIC_MODEL=claude-sonnet-4-5-20250929
-

The default model is defined in the Settings class:

HuggingFace Configuration

HuggingFace can work without an API key for public models, but an API key provides higher rate limits:

# Option 1: Using HF_TOKEN (preferred)
-HF_TOKEN=your_huggingface_token_here
-
-# Option 2: Using HUGGINGFACE_API_KEY (alternative)
-HUGGINGFACE_API_KEY=your_huggingface_api_key_here
-
-# Default model
-HUGGINGFACE_MODEL=meta-llama/Llama-3.1-8B-Instruct
-

The HuggingFace token can be set via either environment variable:

Optional Configuration

Embedding Configuration

DeepCritical supports multiple embedding providers for semantic search and RAG:

# Embedding Provider: "openai", "local", or "huggingface"
-EMBEDDING_PROVIDER=local
-
-# OpenAI Embedding Model (used by LlamaIndex RAG)
-OPENAI_EMBEDDING_MODEL=text-embedding-3-small
-
-# Local Embedding Model (sentence-transformers, used by EmbeddingService)
-LOCAL_EMBEDDING_MODEL=all-MiniLM-L6-v2
-
-# HuggingFace Embedding Model
-HUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2
-

The embedding provider configuration:

Note: OpenAI embeddings require OPENAI_API_KEY. The local provider (default) uses sentence-transformers and requires no API key.

Web Search Configuration

DeepCritical supports multiple web search providers:

# Web Search Provider: "serper", "searchxng", "brave", "tavily", or "duckduckgo"
-# Default: "duckduckgo" (no API key required)
-WEB_SEARCH_PROVIDER=duckduckgo
-
-# Serper API Key (for Google search via Serper)
-SERPER_API_KEY=your_serper_api_key_here
-
-# SearchXNG Host URL (for self-hosted search)
-SEARCHXNG_HOST=http://localhost:8080
-
-# Brave Search API Key
-BRAVE_API_KEY=your_brave_api_key_here
-
-# Tavily API Key
-TAVILY_API_KEY=your_tavily_api_key_here
-

The web search provider configuration:

Note: DuckDuckGo is the default and requires no API key, making it ideal for development and testing.

PubMed Configuration

PubMed search supports optional NCBI API key for higher rate limits:

# NCBI API Key (optional, for higher rate limits: 10 req/sec vs 3 req/sec)
-NCBI_API_KEY=your_ncbi_api_key_here
-

The PubMed tool uses this configuration:

Agent Configuration

Control agent behavior and research loop execution:

# Maximum iterations per research loop (1-50, default: 10)
-MAX_ITERATIONS=10
-
-# Search timeout in seconds
-SEARCH_TIMEOUT=30
-
-# Use graph-based execution for research flows
-USE_GRAPH_EXECUTION=false
-

The agent configuration fields:

Budget & Rate Limiting Configuration

Control resource limits for research loops:

# Default token budget per research loop (1000-1000000, default: 100000)
-DEFAULT_TOKEN_LIMIT=100000
-
-# Default time limit per research loop in minutes (1-120, default: 10)
-DEFAULT_TIME_LIMIT_MINUTES=10
-
-# Default iterations limit per research loop (1-50, default: 10)
-DEFAULT_ITERATIONS_LIMIT=10
-

The budget configuration with validation:

RAG Service Configuration

Configure the Retrieval-Augmented Generation service:

# ChromaDB collection name for RAG
-RAG_COLLECTION_NAME=deepcritical_evidence
-
-# Number of top results to retrieve from RAG (1-50, default: 5)
-RAG_SIMILARITY_TOP_K=5
-
-# Automatically ingest evidence into RAG
-RAG_AUTO_INGEST=true
-

The RAG configuration:

ChromaDB Configuration

Configure the vector database for embeddings and RAG:

# ChromaDB storage path
-CHROMA_DB_PATH=./chroma_db
-
-# Whether to persist ChromaDB to disk
-CHROMA_DB_PERSIST=true
-
-# ChromaDB server host (for remote ChromaDB, optional)
-CHROMA_DB_HOST=localhost
-
-# ChromaDB server port (for remote ChromaDB, optional)
-CHROMA_DB_PORT=8000
-

The ChromaDB configuration:

External Services

Modal is used for secure sandbox execution of statistical analysis:

# Modal Token ID (for Modal sandbox execution)
-MODAL_TOKEN_ID=your_modal_token_id_here
-
-# Modal Token Secret
-MODAL_TOKEN_SECRET=your_modal_token_secret_here
-

The Modal configuration:

Logging Configuration

Configure structured logging:

# Log Level: "DEBUG", "INFO", "WARNING", or "ERROR"
-LOG_LEVEL=INFO
-

The logging configuration:

Logging is configured via the configure_logging() function:

Configuration Properties

The Settings class provides helpful properties for checking configuration state:

API Key Availability

Check which API keys are available:

Usage:

from src.utils.config import settings
-
-# Check API key availability
-if settings.has_openai_key:
-    # Use OpenAI
-    pass
-
-if settings.has_anthropic_key:
-    # Use Anthropic
-    pass
-
-if settings.has_huggingface_key:
-    # Use HuggingFace
-    pass
-
-if settings.has_any_llm_key:
-    # At least one LLM is available
-    pass
-

Service Availability

Check if external services are configured:

Usage:

from src.utils.config import settings
-
-# Check service availability
-if settings.modal_available:
-    # Use Modal sandbox
-    pass
-
-if settings.web_search_available:
-    # Web search is configured
-    pass
-

API Key Retrieval

Get the API key for the configured provider:

For OpenAI-specific operations (e.g., Magentic mode):

Configuration Usage in Codebase

The configuration system is used throughout the codebase:

LLM Factory

The LLM factory uses settings to create appropriate models:

Embedding Service

The embedding service uses local embedding model configuration:

Orchestrator Factory

The orchestrator factory uses settings to determine mode:

Environment Variables Reference

Required (at least one LLM)

  • OPENAI_API_KEY - OpenAI API key (required for OpenAI provider)
  • ANTHROPIC_API_KEY - Anthropic API key (required for Anthropic provider)
  • HF_TOKEN or HUGGINGFACE_API_KEY - HuggingFace API token (optional, can work without for public models)

LLM Configuration Variables

  • LLM_PROVIDER - Provider to use: "openai", "anthropic", or "huggingface" (default: "huggingface")
  • OPENAI_MODEL - OpenAI model name (default: "gpt-5.1")
  • ANTHROPIC_MODEL - Anthropic model name (default: "claude-sonnet-4-5-20250929")
  • HUGGINGFACE_MODEL - HuggingFace model ID (default: "meta-llama/Llama-3.1-8B-Instruct")

Embedding Configuration Variables

  • EMBEDDING_PROVIDER - Provider: "openai", "local", or "huggingface" (default: "local")
  • OPENAI_EMBEDDING_MODEL - OpenAI embedding model (default: "text-embedding-3-small")
  • LOCAL_EMBEDDING_MODEL - Local sentence-transformers model (default: "all-MiniLM-L6-v2")
  • HUGGINGFACE_EMBEDDING_MODEL - HuggingFace embedding model (default: "sentence-transformers/all-MiniLM-L6-v2")

Web Search Configuration Variables

  • WEB_SEARCH_PROVIDER - Provider: "serper", "searchxng", "brave", "tavily", or "duckduckgo" (default: "duckduckgo")
  • SERPER_API_KEY - Serper API key (required for Serper provider)
  • SEARCHXNG_HOST - SearchXNG host URL (required for SearchXNG provider)
  • BRAVE_API_KEY - Brave Search API key (required for Brave provider)
  • TAVILY_API_KEY - Tavily API key (required for Tavily provider)

PubMed Configuration Variables

  • NCBI_API_KEY - NCBI API key (optional, increases rate limit from 3 to 10 req/sec)

Agent Configuration Variables

  • MAX_ITERATIONS - Maximum iterations per research loop (1-50, default: 10)
  • SEARCH_TIMEOUT - Search timeout in seconds (default: 30)
  • USE_GRAPH_EXECUTION - Use graph-based execution (default: false)

Budget Configuration Variables

  • DEFAULT_TOKEN_LIMIT - Default token budget per research loop (1000-1000000, default: 100000)
  • DEFAULT_TIME_LIMIT_MINUTES - Default time limit in minutes (1-120, default: 10)
  • DEFAULT_ITERATIONS_LIMIT - Default iterations limit (1-50, default: 10)

RAG Configuration Variables

  • RAG_COLLECTION_NAME - ChromaDB collection name (default: "deepcritical_evidence")
  • RAG_SIMILARITY_TOP_K - Number of top results to retrieve (1-50, default: 5)
  • RAG_AUTO_INGEST - Automatically ingest evidence into RAG (default: true)

ChromaDB Configuration Variables

  • CHROMA_DB_PATH - ChromaDB storage path (default: "./chroma_db")
  • CHROMA_DB_PERSIST - Whether to persist ChromaDB to disk (default: true)
  • CHROMA_DB_HOST - ChromaDB server host (optional, for remote ChromaDB)
  • CHROMA_DB_PORT - ChromaDB server port (optional, for remote ChromaDB)

External Services Variables

  • MODAL_TOKEN_ID - Modal token ID (optional, for Modal sandbox execution)
  • MODAL_TOKEN_SECRET - Modal token secret (optional, for Modal sandbox execution)

Logging Configuration Variables

  • LOG_LEVEL - Log level: "DEBUG", "INFO", "WARNING", or "ERROR" (default: "INFO")

Validation

Settings are validated on load using Pydantic validation:

  • Type Checking: All fields are strongly typed
  • Range Validation: Numeric fields have min/max constraints (e.g., ge=1, le=50 for max_iterations)
  • Literal Validation: Enum fields only accept specific values (e.g., Literal["openai", "anthropic", "huggingface"])
  • Required Fields: API keys are checked when accessed via get_api_key() or get_openai_api_key()

Validation Examples

The max_iterations field has range validation:

The llm_provider field has literal validation:

Error Handling

Configuration errors raise ConfigurationError from src/utils/exceptions.py:

Error Handling Example

from src.utils.config import settings
-from src.utils.exceptions import ConfigurationError
-
-try:
-    api_key = settings.get_api_key()
-except ConfigurationError as e:
-    print(f"Configuration error: {e}")
-

Common Configuration Errors

  1. Missing API Key: When get_api_key() is called but the required API key is not set
  2. Invalid Provider: When llm_provider is set to an unsupported value
  3. Out of Range: When numeric values exceed their min/max constraints
  4. Invalid Literal: When enum fields receive unsupported values

Configuration Best Practices

  1. Use .env File: Store sensitive keys in .env file (add to .gitignore)
  2. Check Availability: Use properties like has_openai_key before accessing API keys
  3. Handle Errors: Always catch ConfigurationError when calling get_api_key()
  4. Validate Early: Configuration is validated on import, so errors surface immediately
  5. Use Defaults: Leverage sensible defaults for optional configuration

Future Enhancements

The following configurations are planned for future phases:

  1. Additional LLM Providers: DeepSeek, OpenRouter, Gemini, Perplexity, Azure OpenAI, Local models
  2. Model Selection: Reasoning/main/fast model configuration
  3. Service Integration: Additional service integrations and configurations
\ No newline at end of file diff --git a/site/configuration/index.html b/site/configuration/index.html index 58bf165..bbcb61c 100644 --- a/site/configuration/index.html +++ b/site/configuration/index.html @@ -1,124 +1,124 @@ - Configuration Guide - The DETERMINATOR
Skip to content

Configuration Guide

Overview

DeepCritical uses Pydantic Settings for centralized configuration management. All settings are defined in the Settings class in src/utils/config.py and can be configured via environment variables or a .env file.

The configuration system provides:

  • Type Safety: Strongly-typed fields with Pydantic validation
  • Environment File Support: Automatically loads from .env file (if present)
  • Case-Insensitive: Environment variables are case-insensitive
  • Singleton Pattern: Global settings instance for easy access throughout the codebase
  • Validation: Automatic validation on load with helpful error messages

Quick Start

  1. Create a .env file in the project root
  2. Set at least one LLM API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, or HF_TOKEN)
  3. Optionally configure other services as needed
  4. The application will automatically load and validate your configuration

Configuration System Architecture

Settings Class

The [Settings][settings-class] class extends BaseSettings from pydantic_settings and defines all application configuration:

View source

Singleton Instance

A global settings instance is available for import:

View source

Usage Pattern

Access configuration throughout the codebase:

from src.utils.config import settings
-
-# Check if API keys are available
-if settings.has_openai_key:
-    # Use OpenAI
-    pass
-
-# Access configuration values
-max_iterations = settings.max_iterations
-web_search_provider = settings.web_search_provider
-

Required Configuration

LLM Provider

You must configure at least one LLM provider. The system supports:

  • OpenAI: Requires OPENAI_API_KEY
  • Anthropic: Requires ANTHROPIC_API_KEY
  • HuggingFace: Optional HF_TOKEN or HUGGINGFACE_API_KEY (can work without key for public models)

OpenAI Configuration

LLM_PROVIDER=openai
-OPENAI_API_KEY=your_openai_api_key_here
-OPENAI_MODEL=gpt-5.1
-

The default model is defined in the Settings class:

Anthropic Configuration

LLM_PROVIDER=anthropic
-ANTHROPIC_API_KEY=your_anthropic_api_key_here
-ANTHROPIC_MODEL=claude-sonnet-4-5-20250929
-

The default model is defined in the Settings class:

HuggingFace Configuration

HuggingFace can work without an API key for public models, but an API key provides higher rate limits:

# Option 1: Using HF_TOKEN (preferred)
-HF_TOKEN=your_huggingface_token_here
-
-# Option 2: Using HUGGINGFACE_API_KEY (alternative)
-HUGGINGFACE_API_KEY=your_huggingface_api_key_here
-
-# Default model
-HUGGINGFACE_MODEL=meta-llama/Llama-3.1-8B-Instruct
-

The HuggingFace token can be set via either environment variable:

Optional Configuration

Embedding Configuration

DeepCritical supports multiple embedding providers for semantic search and RAG:

# Embedding Provider: "openai", "local", or "huggingface"
-EMBEDDING_PROVIDER=local
-
-# OpenAI Embedding Model (used by LlamaIndex RAG)
-OPENAI_EMBEDDING_MODEL=text-embedding-3-small
-
-# Local Embedding Model (sentence-transformers, used by EmbeddingService)
-LOCAL_EMBEDDING_MODEL=all-MiniLM-L6-v2
-
-# HuggingFace Embedding Model
-HUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2
-

The embedding provider configuration:

Note: OpenAI embeddings require OPENAI_API_KEY. The local provider (default) uses sentence-transformers and requires no API key.

Web Search Configuration

DeepCritical supports multiple web search providers:

# Web Search Provider: "serper", "searchxng", "brave", "tavily", or "duckduckgo"
-# Default: "duckduckgo" (no API key required)
-WEB_SEARCH_PROVIDER=duckduckgo
-
-# Serper API Key (for Google search via Serper)
-SERPER_API_KEY=your_serper_api_key_here
-
-# SearchXNG Host URL (for self-hosted search)
-SEARCHXNG_HOST=http://localhost:8080
-
-# Brave Search API Key
-BRAVE_API_KEY=your_brave_api_key_here
-
-# Tavily API Key
-TAVILY_API_KEY=your_tavily_api_key_here
-

The web search provider configuration:

Note: DuckDuckGo is the default and requires no API key, making it ideal for development and testing.

PubMed Configuration

PubMed search supports optional NCBI API key for higher rate limits:

# NCBI API Key (optional, for higher rate limits: 10 req/sec vs 3 req/sec)
-NCBI_API_KEY=your_ncbi_api_key_here
-

The PubMed tool uses this configuration:

Agent Configuration

Control agent behavior and research loop execution:

# Maximum iterations per research loop (1-50, default: 10)
-MAX_ITERATIONS=10
-
-# Search timeout in seconds
-SEARCH_TIMEOUT=30
-
-# Use graph-based execution for research flows
-USE_GRAPH_EXECUTION=false
-

The agent configuration fields:

Budget & Rate Limiting Configuration

Control resource limits for research loops:

# Default token budget per research loop (1000-1000000, default: 100000)
-DEFAULT_TOKEN_LIMIT=100000
-
-# Default time limit per research loop in minutes (1-120, default: 10)
-DEFAULT_TIME_LIMIT_MINUTES=10
-
-# Default iterations limit per research loop (1-50, default: 10)
-DEFAULT_ITERATIONS_LIMIT=10
-

The budget configuration with validation:

RAG Service Configuration

Configure the Retrieval-Augmented Generation service:

# ChromaDB collection name for RAG
-RAG_COLLECTION_NAME=deepcritical_evidence
-
-# Number of top results to retrieve from RAG (1-50, default: 5)
-RAG_SIMILARITY_TOP_K=5
-
-# Automatically ingest evidence into RAG
-RAG_AUTO_INGEST=true
-

The RAG configuration:

ChromaDB Configuration

Configure the vector database for embeddings and RAG:

# ChromaDB storage path
-CHROMA_DB_PATH=./chroma_db
-
-# Whether to persist ChromaDB to disk
-CHROMA_DB_PERSIST=true
-
-# ChromaDB server host (for remote ChromaDB, optional)
-CHROMA_DB_HOST=localhost
-
-# ChromaDB server port (for remote ChromaDB, optional)
-CHROMA_DB_PORT=8000
-

The ChromaDB configuration:

External Services

Modal is used for secure sandbox execution of statistical analysis:

# Modal Token ID (for Modal sandbox execution)
-MODAL_TOKEN_ID=your_modal_token_id_here
-
-# Modal Token Secret
-MODAL_TOKEN_SECRET=your_modal_token_secret_here
-

The Modal configuration:

Logging Configuration

Configure structured logging:

# Log Level: "DEBUG", "INFO", "WARNING", or "ERROR"
-LOG_LEVEL=INFO
-

The logging configuration:

Logging is configured via the configure_logging() function:

Configuration Properties

The Settings class provides helpful properties for checking configuration state:

API Key Availability

Check which API keys are available:

Usage:

from src.utils.config import settings
-
-# Check API key availability
-if settings.has_openai_key:
-    # Use OpenAI
-    pass
-
-if settings.has_anthropic_key:
-    # Use Anthropic
-    pass
-
-if settings.has_huggingface_key:
-    # Use HuggingFace
-    pass
-
-if settings.has_any_llm_key:
-    # At least one LLM is available
-    pass
-

Service Availability

Check if external services are configured:

Usage:

from src.utils.config import settings
-
-# Check service availability
-if settings.modal_available:
-    # Use Modal sandbox
-    pass
-
-if settings.web_search_available:
-    # Web search is configured
-    pass
-

API Key Retrieval

Get the API key for the configured provider:

For OpenAI-specific operations (e.g., Magentic mode):

Configuration Usage in Codebase

The configuration system is used throughout the codebase:

LLM Factory

The LLM factory uses settings to create appropriate models:

Embedding Service

The embedding service uses local embedding model configuration:

Orchestrator Factory

The orchestrator factory uses settings to determine mode:

Environment Variables Reference

Required (at least one LLM)

  • OPENAI_API_KEY - OpenAI API key (required for OpenAI provider)
  • ANTHROPIC_API_KEY - Anthropic API key (required for Anthropic provider)
  • HF_TOKEN or HUGGINGFACE_API_KEY - HuggingFace API token (optional, can work without for public models)

LLM Configuration Variables

  • LLM_PROVIDER - Provider to use: "openai", "anthropic", or "huggingface" (default: "huggingface")
  • OPENAI_MODEL - OpenAI model name (default: "gpt-5.1")
  • ANTHROPIC_MODEL - Anthropic model name (default: "claude-sonnet-4-5-20250929")
  • HUGGINGFACE_MODEL - HuggingFace model ID (default: "meta-llama/Llama-3.1-8B-Instruct")

Embedding Configuration Variables

  • EMBEDDING_PROVIDER - Provider: "openai", "local", or "huggingface" (default: "local")
  • OPENAI_EMBEDDING_MODEL - OpenAI embedding model (default: "text-embedding-3-small")
  • LOCAL_EMBEDDING_MODEL - Local sentence-transformers model (default: "all-MiniLM-L6-v2")
  • HUGGINGFACE_EMBEDDING_MODEL - HuggingFace embedding model (default: "sentence-transformers/all-MiniLM-L6-v2")

Web Search Configuration Variables

  • WEB_SEARCH_PROVIDER - Provider: "serper", "searchxng", "brave", "tavily", or "duckduckgo" (default: "duckduckgo")
  • SERPER_API_KEY - Serper API key (required for Serper provider)
  • SEARCHXNG_HOST - SearchXNG host URL (required for SearchXNG provider)
  • BRAVE_API_KEY - Brave Search API key (required for Brave provider)
  • TAVILY_API_KEY - Tavily API key (required for Tavily provider)

PubMed Configuration Variables

  • NCBI_API_KEY - NCBI API key (optional, increases rate limit from 3 to 10 req/sec)

Agent Configuration Variables

  • MAX_ITERATIONS - Maximum iterations per research loop (1-50, default: 10)
  • SEARCH_TIMEOUT - Search timeout in seconds (default: 30)
  • USE_GRAPH_EXECUTION - Use graph-based execution (default: false)

Budget Configuration Variables

  • DEFAULT_TOKEN_LIMIT - Default token budget per research loop (1000-1000000, default: 100000)
  • DEFAULT_TIME_LIMIT_MINUTES - Default time limit in minutes (1-120, default: 10)
  • DEFAULT_ITERATIONS_LIMIT - Default iterations limit (1-50, default: 10)

RAG Configuration Variables

  • RAG_COLLECTION_NAME - ChromaDB collection name (default: "deepcritical_evidence")
  • RAG_SIMILARITY_TOP_K - Number of top results to retrieve (1-50, default: 5)
  • RAG_AUTO_INGEST - Automatically ingest evidence into RAG (default: true)

ChromaDB Configuration Variables

  • CHROMA_DB_PATH - ChromaDB storage path (default: "./chroma_db")
  • CHROMA_DB_PERSIST - Whether to persist ChromaDB to disk (default: true)
  • CHROMA_DB_HOST - ChromaDB server host (optional, for remote ChromaDB)
  • CHROMA_DB_PORT - ChromaDB server port (optional, for remote ChromaDB)

External Services Variables

  • MODAL_TOKEN_ID - Modal token ID (optional, for Modal sandbox execution)
  • MODAL_TOKEN_SECRET - Modal token secret (optional, for Modal sandbox execution)

Logging Configuration Variables

  • LOG_LEVEL - Log level: "DEBUG", "INFO", "WARNING", or "ERROR" (default: "INFO")

Validation

Settings are validated on load using Pydantic validation:

  • Type Checking: All fields are strongly typed
  • Range Validation: Numeric fields have min/max constraints (e.g., ge=1, le=50 for max_iterations)
  • Literal Validation: Enum fields only accept specific values (e.g., Literal["openai", "anthropic", "huggingface"])
  • Required Fields: API keys are checked when accessed via get_api_key() or get_openai_api_key()

Validation Examples

The max_iterations field has range validation:

The llm_provider field has literal validation:

Error Handling

Configuration errors raise ConfigurationError from src/utils/exceptions.py:

```22:25:src/utils/exceptions.py class ConfigurationError(DeepCriticalError): """Raised when configuration is invalid."""

pass
-

```

Error Handling Example

python from src.utils.config import settings from src.utils.exceptions import ConfigurationError try: api_key = settings.get_api_key() except ConfigurationError as e: print(f"Configuration error: {e}")

Common Configuration Errors

  1. Missing API Key: When get_api_key() is called but the required API key is not set
  2. Invalid Provider: When llm_provider is set to an unsupported value
  3. Out of Range: When numeric values exceed their min/max constraints
  4. Invalid Literal: When enum fields receive unsupported values

Configuration Best Practices

  1. Use .env File: Store sensitive keys in .env file (add to .gitignore)
  2. Check Availability: Use properties like has_openai_key before accessing API keys
  3. Handle Errors: Always catch ConfigurationError when calling get_api_key()
  4. Validate Early: Configuration is validated on import, so errors surface immediately
  5. Use Defaults: Leverage sensible defaults for optional configuration

Future Enhancements

The following configurations are planned for future phases:

  1. Additional LLM Providers: DeepSeek, OpenRouter, Gemini, Perplexity, Azure OpenAI, Local models
  2. Model Selection: Reasoning/main/fast model configuration
  3. Service Integration: Additional service integrations and configurations
\ No newline at end of file + Configuration Guide - The DETERMINATOR
Skip to content

Configuration Guide

Overview

DeepCritical uses Pydantic Settings for centralized configuration management. All settings are defined in the Settings class in src/utils/config.py and can be configured via environment variables or a .env file.

The configuration system provides:

  • Type Safety: Strongly-typed fields with Pydantic validation
  • Environment File Support: Automatically loads from .env file (if present)
  • Case-Insensitive: Environment variables are case-insensitive
  • Singleton Pattern: Global settings instance for easy access throughout the codebase
  • Validation: Automatic validation on load with helpful error messages

Quick Start

  1. Create a .env file in the project root
  2. Set at least one LLM API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, or HF_TOKEN)
  3. Optionally configure other services as needed
  4. The application will automatically load and validate your configuration

Configuration System Architecture

Settings Class

The [Settings][settings-class] class extends BaseSettings from pydantic_settings and defines all application configuration:

View source

Singleton Instance

A global settings instance is available for import:

View source

Usage Pattern

Access configuration throughout the codebase:

from src.utils.config import settings
+
+# Check if API keys are available
+if settings.has_openai_key:
+    # Use OpenAI
+    pass
+
+# Access configuration values
+max_iterations = settings.max_iterations
+web_search_provider = settings.web_search_provider
+

Required Configuration

LLM Provider

You must configure at least one LLM provider. The system supports:

  • OpenAI: Requires OPENAI_API_KEY
  • Anthropic: Requires ANTHROPIC_API_KEY
  • HuggingFace: Optional HF_TOKEN or HUGGINGFACE_API_KEY (can work without key for public models)

OpenAI Configuration

LLM_PROVIDER=openai
+OPENAI_API_KEY=your_openai_api_key_here
+OPENAI_MODEL=gpt-5.1
+

The default model is defined in the Settings class:

Anthropic Configuration

LLM_PROVIDER=anthropic
+ANTHROPIC_API_KEY=your_anthropic_api_key_here
+ANTHROPIC_MODEL=claude-sonnet-4-5-20250929
+

The default model is defined in the Settings class:

HuggingFace Configuration

HuggingFace can work without an API key for public models, but an API key provides higher rate limits:

# Option 1: Using HF_TOKEN (preferred)
+HF_TOKEN=your_huggingface_token_here
+
+# Option 2: Using HUGGINGFACE_API_KEY (alternative)
+HUGGINGFACE_API_KEY=your_huggingface_api_key_here
+
+# Default model
+HUGGINGFACE_MODEL=meta-llama/Llama-3.1-8B-Instruct
+

The HuggingFace token can be set via either environment variable:

Optional Configuration

Embedding Configuration

DeepCritical supports multiple embedding providers for semantic search and RAG:

# Embedding Provider: "openai", "local", or "huggingface"
+EMBEDDING_PROVIDER=local
+
+# OpenAI Embedding Model (used by LlamaIndex RAG)
+OPENAI_EMBEDDING_MODEL=text-embedding-3-small
+
+# Local Embedding Model (sentence-transformers, used by EmbeddingService)
+LOCAL_EMBEDDING_MODEL=all-MiniLM-L6-v2
+
+# HuggingFace Embedding Model
+HUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2
+

The embedding provider configuration:

Note: OpenAI embeddings require OPENAI_API_KEY. The local provider (default) uses sentence-transformers and requires no API key.

Web Search Configuration

DeepCritical supports multiple web search providers:

# Web Search Provider: "serper", "searchxng", "brave", "tavily", or "duckduckgo"
+# Default: "duckduckgo" (no API key required)
+WEB_SEARCH_PROVIDER=duckduckgo
+
+# Serper API Key (for Google search via Serper)
+SERPER_API_KEY=your_serper_api_key_here
+
+# SearchXNG Host URL (for self-hosted search)
+SEARCHXNG_HOST=http://localhost:8080
+
+# Brave Search API Key
+BRAVE_API_KEY=your_brave_api_key_here
+
+# Tavily API Key
+TAVILY_API_KEY=your_tavily_api_key_here
+

The web search provider configuration:

Note: DuckDuckGo is the default and requires no API key, making it ideal for development and testing.

PubMed Configuration

PubMed search supports optional NCBI API key for higher rate limits:

# NCBI API Key (optional, for higher rate limits: 10 req/sec vs 3 req/sec)
+NCBI_API_KEY=your_ncbi_api_key_here
+

The PubMed tool uses this configuration:

Agent Configuration

Control agent behavior and research loop execution:

# Maximum iterations per research loop (1-50, default: 10)
+MAX_ITERATIONS=10
+
+# Search timeout in seconds
+SEARCH_TIMEOUT=30
+
+# Use graph-based execution for research flows
+USE_GRAPH_EXECUTION=false
+

The agent configuration fields:

Budget & Rate Limiting Configuration

Control resource limits for research loops:

# Default token budget per research loop (1000-1000000, default: 100000)
+DEFAULT_TOKEN_LIMIT=100000
+
+# Default time limit per research loop in minutes (1-120, default: 10)
+DEFAULT_TIME_LIMIT_MINUTES=10
+
+# Default iterations limit per research loop (1-50, default: 10)
+DEFAULT_ITERATIONS_LIMIT=10
+

The budget configuration with validation:

RAG Service Configuration

Configure the Retrieval-Augmented Generation service:

# ChromaDB collection name for RAG
+RAG_COLLECTION_NAME=deepcritical_evidence
+
+# Number of top results to retrieve from RAG (1-50, default: 5)
+RAG_SIMILARITY_TOP_K=5
+
+# Automatically ingest evidence into RAG
+RAG_AUTO_INGEST=true
+

The RAG configuration:

ChromaDB Configuration

Configure the vector database for embeddings and RAG:

# ChromaDB storage path
+CHROMA_DB_PATH=./chroma_db
+
+# Whether to persist ChromaDB to disk
+CHROMA_DB_PERSIST=true
+
+# ChromaDB server host (for remote ChromaDB, optional)
+CHROMA_DB_HOST=localhost
+
+# ChromaDB server port (for remote ChromaDB, optional)
+CHROMA_DB_PORT=8000
+

The ChromaDB configuration:

External Services

Modal is used for secure sandbox execution of statistical analysis:

# Modal Token ID (for Modal sandbox execution)
+MODAL_TOKEN_ID=your_modal_token_id_here
+
+# Modal Token Secret
+MODAL_TOKEN_SECRET=your_modal_token_secret_here
+

The Modal configuration:

Logging Configuration

Configure structured logging:

# Log Level: "DEBUG", "INFO", "WARNING", or "ERROR"
+LOG_LEVEL=INFO
+

The logging configuration:

Logging is configured via the configure_logging() function:

Configuration Properties

The Settings class provides helpful properties for checking configuration state:

API Key Availability

Check which API keys are available:

Usage:

from src.utils.config import settings
+
+# Check API key availability
+if settings.has_openai_key:
+    # Use OpenAI
+    pass
+
+if settings.has_anthropic_key:
+    # Use Anthropic
+    pass
+
+if settings.has_huggingface_key:
+    # Use HuggingFace
+    pass
+
+if settings.has_any_llm_key:
+    # At least one LLM is available
+    pass
+

Service Availability

Check if external services are configured:

Usage:

from src.utils.config import settings
+
+# Check service availability
+if settings.modal_available:
+    # Use Modal sandbox
+    pass
+
+if settings.web_search_available:
+    # Web search is configured
+    pass
+

API Key Retrieval

Get the API key for the configured provider:

For OpenAI-specific operations (e.g., Magentic mode):

Configuration Usage in Codebase

The configuration system is used throughout the codebase:

LLM Factory

The LLM factory uses settings to create appropriate models:

Embedding Service

The embedding service uses local embedding model configuration:

Orchestrator Factory

The orchestrator factory uses settings to determine mode:

Environment Variables Reference

Required (at least one LLM)

  • OPENAI_API_KEY - OpenAI API key (required for OpenAI provider)
  • ANTHROPIC_API_KEY - Anthropic API key (required for Anthropic provider)
  • HF_TOKEN or HUGGINGFACE_API_KEY - HuggingFace API token (optional, can work without for public models)

LLM Configuration Variables

  • LLM_PROVIDER - Provider to use: "openai", "anthropic", or "huggingface" (default: "huggingface")
  • OPENAI_MODEL - OpenAI model name (default: "gpt-5.1")
  • ANTHROPIC_MODEL - Anthropic model name (default: "claude-sonnet-4-5-20250929")
  • HUGGINGFACE_MODEL - HuggingFace model ID (default: "meta-llama/Llama-3.1-8B-Instruct")

Embedding Configuration Variables

  • EMBEDDING_PROVIDER - Provider: "openai", "local", or "huggingface" (default: "local")
  • OPENAI_EMBEDDING_MODEL - OpenAI embedding model (default: "text-embedding-3-small")
  • LOCAL_EMBEDDING_MODEL - Local sentence-transformers model (default: "all-MiniLM-L6-v2")
  • HUGGINGFACE_EMBEDDING_MODEL - HuggingFace embedding model (default: "sentence-transformers/all-MiniLM-L6-v2")

Web Search Configuration Variables

  • WEB_SEARCH_PROVIDER - Provider: "serper", "searchxng", "brave", "tavily", or "duckduckgo" (default: "duckduckgo")
  • SERPER_API_KEY - Serper API key (required for Serper provider)
  • SEARCHXNG_HOST - SearchXNG host URL (required for SearchXNG provider)
  • BRAVE_API_KEY - Brave Search API key (required for Brave provider)
  • TAVILY_API_KEY - Tavily API key (required for Tavily provider)

PubMed Configuration Variables

  • NCBI_API_KEY - NCBI API key (optional, increases rate limit from 3 to 10 req/sec)

Agent Configuration Variables

  • MAX_ITERATIONS - Maximum iterations per research loop (1-50, default: 10)
  • SEARCH_TIMEOUT - Search timeout in seconds (default: 30)
  • USE_GRAPH_EXECUTION - Use graph-based execution (default: false)

Budget Configuration Variables

  • DEFAULT_TOKEN_LIMIT - Default token budget per research loop (1000-1000000, default: 100000)
  • DEFAULT_TIME_LIMIT_MINUTES - Default time limit in minutes (1-120, default: 10)
  • DEFAULT_ITERATIONS_LIMIT - Default iterations limit (1-50, default: 10)

RAG Configuration Variables

  • RAG_COLLECTION_NAME - ChromaDB collection name (default: "deepcritical_evidence")
  • RAG_SIMILARITY_TOP_K - Number of top results to retrieve (1-50, default: 5)
  • RAG_AUTO_INGEST - Automatically ingest evidence into RAG (default: true)

ChromaDB Configuration Variables

  • CHROMA_DB_PATH - ChromaDB storage path (default: "./chroma_db")
  • CHROMA_DB_PERSIST - Whether to persist ChromaDB to disk (default: true)
  • CHROMA_DB_HOST - ChromaDB server host (optional, for remote ChromaDB)
  • CHROMA_DB_PORT - ChromaDB server port (optional, for remote ChromaDB)

External Services Variables

  • MODAL_TOKEN_ID - Modal token ID (optional, for Modal sandbox execution)
  • MODAL_TOKEN_SECRET - Modal token secret (optional, for Modal sandbox execution)

Logging Configuration Variables

  • LOG_LEVEL - Log level: "DEBUG", "INFO", "WARNING", or "ERROR" (default: "INFO")

Validation

Settings are validated on load using Pydantic validation:

  • Type Checking: All fields are strongly typed
  • Range Validation: Numeric fields have min/max constraints (e.g., ge=1, le=50 for max_iterations)
  • Literal Validation: Enum fields only accept specific values (e.g., Literal["openai", "anthropic", "huggingface"])
  • Required Fields: API keys are checked when accessed via get_api_key() or get_openai_api_key()

Validation Examples

The max_iterations field has range validation:

The llm_provider field has literal validation:

Error Handling

Configuration errors raise ConfigurationError from src/utils/exceptions.py:

```22:25:src/utils/exceptions.py class ConfigurationError(DeepCriticalError): """Raised when configuration is invalid."""

pass
+

```

Error Handling Example

python from src.utils.config import settings from src.utils.exceptions import ConfigurationError try: api_key = settings.get_api_key() except ConfigurationError as e: print(f"Configuration error: {e}")

Common Configuration Errors

  1. Missing API Key: When get_api_key() is called but the required API key is not set
  2. Invalid Provider: When llm_provider is set to an unsupported value
  3. Out of Range: When numeric values exceed their min/max constraints
  4. Invalid Literal: When enum fields receive unsupported values

Configuration Best Practices

  1. Use .env File: Store sensitive keys in .env file (add to .gitignore)
  2. Check Availability: Use properties like has_openai_key before accessing API keys
  3. Handle Errors: Always catch ConfigurationError when calling get_api_key()
  4. Validate Early: Configuration is validated on import, so errors surface immediately
  5. Use Defaults: Leverage sensible defaults for optional configuration

Future Enhancements

The following configurations are planned for future phases:

  1. Additional LLM Providers: DeepSeek, OpenRouter, Gemini, Perplexity, Azure OpenAI, Local models
  2. Model Selection: Reasoning/main/fast model configuration
  3. Service Integration: Additional service integrations and configurations
\ No newline at end of file diff --git a/site/contributing/code-quality/index.html b/site/contributing/code-quality/index.html index 29f5c7f..a7d5d11 100644 --- a/site/contributing/code-quality/index.html +++ b/site/contributing/code-quality/index.html @@ -1 +1,12 @@ - Code Quality & Documentation - The DETERMINATOR
Skip to content

Code Quality & Documentation

This document outlines code quality standards and documentation requirements.

Linting

  • Ruff with 100-char line length
  • Ignore rules documented in pyproject.toml:
  • PLR0913: Too many arguments (agents need many params)
  • PLR0912: Too many branches (complex orchestrator logic)
  • PLR0911: Too many return statements (complex agent logic)
  • PLR2004: Magic values (statistical constants)
  • PLW0603: Global statement (singleton pattern)
  • PLC0415: Lazy imports for optional dependencies

Type Checking

  • mypy --strict compliance
  • ignore_missing_imports = true (for optional dependencies)
  • Exclude: reference_repos/, examples/
  • All functions must have complete type annotations

Pre-commit

  • Run make check before committing
  • Must pass: lint + typecheck + test-cov
  • Pre-commit hooks installed via make install

Documentation

Docstrings

  • Google-style docstrings for all public functions
  • Include Args, Returns, Raises sections
  • Use type hints in docstrings only if needed for clarity

Example:

Code Comments

  • Explain WHY, not WHAT
  • Document non-obvious patterns (e.g., why requests not httpx for ClinicalTrials)
  • Mark critical sections: # CRITICAL: ...
  • Document rate limiting rationale
  • Explain async patterns when non-obvious

See Also

\ No newline at end of file + Code Quality & Documentation - The DETERMINATOR
Skip to content

Code Quality & Documentation

This document outlines code quality standards and documentation requirements for The DETERMINATOR.

Linting

  • Ruff with 100-char line length
  • Ignore rules documented in pyproject.toml:
  • PLR0913: Too many arguments (agents need many params)
  • PLR0912: Too many branches (complex orchestrator logic)
  • PLR0911: Too many return statements (complex agent logic)
  • PLR2004: Magic values (statistical constants)
  • PLW0603: Global statement (singleton pattern)
  • PLC0415: Lazy imports for optional dependencies
  • E402: Module level import not at top (needed for pytest.importorskip)
  • E501: Line too long (ignore line length violations)
  • RUF100: Unused noqa (version differences between local/CI)

Type Checking

  • mypy --strict compliance
  • ignore_missing_imports = true (for optional dependencies)
  • Exclude: reference_repos/, examples/
  • All functions must have complete type annotations

Pre-commit

Pre-commit hooks run automatically on commit to ensure code quality. Configuration is in .pre-commit-config.yaml.

Installation

# Install dependencies (includes pre-commit package)
+uv sync --all-extras
+
+# Set up git hooks (must be run separately)
+uv run pre-commit install
+

Note: uv sync --all-extras installs the pre-commit package, but you must run uv run pre-commit install separately to set up the git hooks.

Pre-commit Hooks

The following hooks run automatically on commit:

  1. ruff: Lints code and fixes issues automatically
  2. Runs on: src/ (excludes tests/, reference_repos/)

  3. Auto-fixes: Yes

  4. ruff-format: Formats code with ruff

  5. Runs on: src/ (excludes tests/, reference_repos/)

  6. Auto-fixes: Yes

  7. mypy: Type checking

  8. Runs on: src/ (excludes folder/)

  9. Additional dependencies: pydantic, pydantic-settings, tenacity, pydantic-ai

  10. pytest-unit: Runs unit tests (excludes OpenAI and embedding_provider tests)

  11. Runs: tests/unit/ with -m "not openai and not embedding_provider"

  12. Always runs: Yes (not just on changed files)

  13. pytest-local-embeddings: Runs local embedding tests

  14. Runs: tests/ with -m "local_embeddings"
  15. Always runs: Yes

Manual Pre-commit Run

To run pre-commit hooks manually (without committing):

uv run pre-commit run --all-files
+

Troubleshooting

  • Hooks failing: Fix the issues shown in the output, then commit again
  • Skipping hooks: Use git commit --no-verify (not recommended)
  • Hook not running: Ensure hooks are installed with uv run pre-commit install
  • Type errors: Check that all dependencies are installed with uv sync --all-extras

Documentation

Building Documentation

Documentation is built using MkDocs. Source files are in docs/, and the configuration is in mkdocs.yml.

# Build documentation
+uv run mkdocs build
+
+# Serve documentation locally (http://127.0.0.1:8000)
+uv run mkdocs serve
+

The documentation site is published at: https://deepcritical.github.io/GradioDemo/

Docstrings

  • Google-style docstrings for all public functions
  • Include Args, Returns, Raises sections
  • Use type hints in docstrings only if needed for clarity

Example:

Code Comments

  • Explain WHY, not WHAT
  • Document non-obvious patterns (e.g., why requests not httpx for ClinicalTrials)
  • Mark critical sections: # CRITICAL: ...
  • Document rate limiting rationale
  • Explain async patterns when non-obvious

See Also

\ No newline at end of file diff --git a/site/contributing/code-style/index.html b/site/contributing/code-style/index.html index 6520bd2..8b09f8f 100644 --- a/site/contributing/code-style/index.html +++ b/site/contributing/code-style/index.html @@ -1,3 +1,23 @@ - Code Style & Conventions - The DETERMINATOR
Skip to content

Code Style & Conventions

This document outlines the code style and conventions for DeepCritical.

Type Safety

  • ALWAYS use type hints for all function parameters and return types
  • Use mypy --strict compliance (no Any unless absolutely necessary)
  • Use TYPE_CHECKING imports for circular dependencies:

Pydantic Models

  • All data exchange uses Pydantic models (src/utils/models.py)
  • Models are frozen (model_config = {"frozen": True}) for immutability
  • Use Field() with descriptions for all model fields
  • Validate with ge=, le=, min_length=, max_length= constraints

Async Patterns

  • ALL I/O operations must be async (async def, await)
  • Use asyncio.gather() for parallel operations
  • CPU-bound work (embeddings, parsing) must use run_in_executor():
loop = asyncio.get_running_loop()
-result = await loop.run_in_executor(None, cpu_bound_function, args)
-
  • Never block the event loop with synchronous I/O

Common Pitfalls

  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Global mutable state: Use ContextVar or pass via parameters
  4. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)

See Also

\ No newline at end of file + Code Style & Conventions - The DETERMINATOR
Skip to content

Code Style & Conventions

This document outlines the code style and conventions for The DETERMINATOR.

Package Manager

This project uses uv as the package manager. All commands should be prefixed with uv run to ensure they run in the correct environment.

Installation

# Install uv if you haven't already (recommended: standalone installer)
+# Unix/macOS/Linux:
+curl -LsSf https://astral.sh/uv/install.sh | sh
+
+# Windows (PowerShell):
+powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
+
+# Alternative: pipx install uv
+# Or: pip install uv
+
+# Sync all dependencies including dev extras
+uv sync --all-extras
+

Running Commands

All development commands should use uv run prefix:

# Instead of: pytest tests/
+uv run pytest tests/
+
+# Instead of: ruff check src
+uv run ruff check src
+
+# Instead of: mypy src
+uv run mypy src
+

This ensures commands run in the correct virtual environment managed by uv.

Type Safety

  • ALWAYS use type hints for all function parameters and return types
  • Use mypy --strict compliance (no Any unless absolutely necessary)
  • Use TYPE_CHECKING imports for circular dependencies:

Pydantic Models

  • All data exchange uses Pydantic models (src/utils/models.py)
  • Models are frozen (model_config = {"frozen": True}) for immutability
  • Use Field() with descriptions for all model fields
  • Validate with ge=, le=, min_length=, max_length= constraints

Async Patterns

  • ALL I/O operations must be async (async def, await)
  • Use asyncio.gather() for parallel operations
  • CPU-bound work (embeddings, parsing) must use run_in_executor():
loop = asyncio.get_running_loop()
+result = await loop.run_in_executor(None, cpu_bound_function, args)
+
  • Never block the event loop with synchronous I/O

Common Pitfalls

  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Global mutable state: Use ContextVar or pass via parameters
  4. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)

See Also

\ No newline at end of file diff --git a/site/contributing/error-handling/index.html b/site/contributing/error-handling/index.html index 4ee75b2..b5e359e 100644 --- a/site/contributing/error-handling/index.html +++ b/site/contributing/error-handling/index.html @@ -1,9 +1,9 @@ - Error Handling & Logging - The DETERMINATOR
Skip to content

Error Handling & Logging

This document outlines error handling and logging conventions for DeepCritical.

Exception Hierarchy

Use custom exception hierarchy (src/utils/exceptions.py):

Error Handling Rules

  • Always chain exceptions: raise SearchError(...) from e
  • Log errors with context using structlog:
logger.error("Operation failed", error=str(e), context=value)
-
  • Never silently swallow exceptions
  • Provide actionable error messages

Logging

  • Use structlog for all logging (NOT print or logging)
  • Import: import structlog; logger = structlog.get_logger()
  • Log with structured data: logger.info("event", key=value)
  • Use appropriate levels: DEBUG, INFO, WARNING, ERROR

Logging Examples

logger.info("Starting search", query=query, tools=[t.name for t in tools])
-logger.warning("Search tool failed", tool=tool.name, error=str(result))
-logger.error("Assessment failed", error=str(e))
-

Error Chaining

Always preserve exception context:

try:
-    result = await api_call()
-except httpx.HTTPError as e:
-    raise SearchError(f"API call failed: {e}") from e
-

See Also

\ No newline at end of file + Error Handling & Logging - The DETERMINATOR
Skip to content

Error Handling & Logging

This document outlines error handling and logging conventions for The DETERMINATOR.

Exception Hierarchy

Use custom exception hierarchy (src/utils/exceptions.py):

Error Handling Rules

  • Always chain exceptions: raise SearchError(...) from e
  • Log errors with context using structlog:
logger.error("Operation failed", error=str(e), context=value)
+
  • Never silently swallow exceptions
  • Provide actionable error messages

Logging

  • Use structlog for all logging (NOT print or logging)
  • Import: import structlog; logger = structlog.get_logger()
  • Log with structured data: logger.info("event", key=value)
  • Use appropriate levels: DEBUG, INFO, WARNING, ERROR

Logging Examples

logger.info("Starting search", query=query, tools=[t.name for t in tools])
+logger.warning("Search tool failed", tool=tool.name, error=str(result))
+logger.error("Assessment failed", error=str(e))
+

Error Chaining

Always preserve exception context:

try:
+    result = await api_call()
+except httpx.HTTPError as e:
+    raise SearchError(f"API call failed: {e}") from e
+

See Also

\ No newline at end of file diff --git a/site/contributing/implementation-patterns/index.html b/site/contributing/implementation-patterns/index.html index 92110c8..eb36b7a 100644 --- a/site/contributing/implementation-patterns/index.html +++ b/site/contributing/implementation-patterns/index.html @@ -1,10 +1,10 @@ - Implementation Patterns - The DETERMINATOR
Skip to content

Implementation Patterns

This document outlines common implementation patterns used in DeepCritical.

Search Tools

All tools implement SearchTool protocol (src/tools/base.py):

  • Must have name property
  • Must implement async def search(query, max_results) -> list[Evidence]
  • Use @retry decorator from tenacity for resilience
  • Rate limiting: Implement _rate_limit() for APIs with limits (e.g., PubMed)
  • Error handling: Raise SearchError or RateLimitError on failures

Example pattern:

class MySearchTool:
-    @property
-    def name(self) -> str:
-        return "mytool"
-    
-    @retry(stop=stop_after_attempt(3), wait=wait_exponential(...))
-    async def search(self, query: str, max_results: int = 10) -> list[Evidence]:
-        # Implementation
-        return evidence_list
-

Judge Handlers

  • Implement JudgeHandlerProtocol (async def assess(question, evidence) -> JudgeAssessment)
  • Use pydantic-ai Agent with output_type=JudgeAssessment
  • System prompts in src/prompts/judge.py
  • Support fallback handlers: MockJudgeHandler, HFInferenceJudgeHandler
  • Always return valid JudgeAssessment (never raise exceptions)

Agent Factory Pattern

  • Use factory functions for creating agents (src/agent_factory/)
  • Lazy initialization for optional dependencies (e.g., embeddings, Modal)
  • Check requirements before initialization:

State Management

  • Magentic Mode: Use ContextVar for thread-safe state (src/agents/state.py)
  • Simple Mode: Pass state via function parameters
  • Never use global mutable state (except singletons via @lru_cache)

Singleton Pattern

Use @lru_cache(maxsize=1) for singletons:

  • Lazy initialization to avoid requiring dependencies at import time

See Also

\ No newline at end of file + Implementation Patterns - The DETERMINATOR
Skip to content

Implementation Patterns

This document outlines common implementation patterns used in The DETERMINATOR.

Search Tools

All tools implement SearchTool protocol (src/tools/base.py):

  • Must have name property
  • Must implement async def search(query, max_results) -> list[Evidence]
  • Use @retry decorator from tenacity for resilience
  • Rate limiting: Implement _rate_limit() for APIs with limits (e.g., PubMed)
  • Error handling: Raise SearchError or RateLimitError on failures

Example pattern:

class MySearchTool:
+    @property
+    def name(self) -> str:
+        return "mytool"
+    
+    @retry(stop=stop_after_attempt(3), wait=wait_exponential(...))
+    async def search(self, query: str, max_results: int = 10) -> list[Evidence]:
+        # Implementation
+        return evidence_list
+

Judge Handlers

  • Implement JudgeHandlerProtocol (async def assess(question, evidence) -> JudgeAssessment)
  • Use pydantic-ai Agent with output_type=JudgeAssessment
  • System prompts in src/prompts/judge.py
  • Support fallback handlers: MockJudgeHandler, HFInferenceJudgeHandler
  • Always return valid JudgeAssessment (never raise exceptions)

Agent Factory Pattern

  • Use factory functions for creating agents (src/agent_factory/)
  • Lazy initialization for optional dependencies (e.g., embeddings, Modal)
  • Check requirements before initialization:

State Management

  • Magentic Mode: Use ContextVar for thread-safe state (src/agents/state.py)
  • Simple Mode: Pass state via function parameters
  • Never use global mutable state (except singletons via @lru_cache)

Singleton Pattern

Use @lru_cache(maxsize=1) for singletons:

  • Lazy initialization to avoid requiring dependencies at import time

See Also

\ No newline at end of file diff --git a/site/contributing/index.html b/site/contributing/index.html index 06c9c33..3d16a3f 100644 --- a/site/contributing/index.html +++ b/site/contributing/index.html @@ -1,15 +1,52 @@ - Contributing to DeepCritical - The DETERMINATOR
Skip to content

Contributing to DeepCritical

Thank you for your interest in contributing to DeepCritical! This guide will help you get started.

Git Workflow

  • main: Production-ready (GitHub)
  • dev: Development integration (GitHub)
  • Use feature branches: yourname-dev
  • NEVER push directly to main or dev on HuggingFace
  • GitHub is source of truth; HuggingFace is for deployment

Development Commands

make install      # Install dependencies + pre-commit
-make check        # Lint + typecheck + test (MUST PASS)
-make test         # Run unit tests
-make lint         # Run ruff
-make format       # Format with ruff
-make typecheck    # Run mypy
-make test-cov     # Test with coverage
-

Getting Started

  1. Fork the repository on GitHub
  2. Clone your fork: 
    git clone https://github.com/yourusername/GradioDemo.git
-cd GradioDemo
-
  3. Install dependencies: 
    make install
-
  4. Create a feature branch: 
    git checkout -b yourname-feature-name
-
  5. Make your changes following the guidelines below
  6. Run checks: 
    make check
-
  7. Commit and push: 
    git commit -m "Description of changes"
-git push origin yourname-feature-name
-
  8. Create a pull request on GitHub

Development Guidelines

Code Style

  • Follow Code Style Guidelines
  • All code must pass mypy --strict
  • Use ruff for linting and formatting
  • Line length: 100 characters

Error Handling

  • Follow Error Handling Guidelines
  • Always chain exceptions: raise SearchError(...) from e
  • Use structured logging with structlog
  • Never silently swallow exceptions

Testing

  • Follow Testing Guidelines
  • Write tests before implementation (TDD)
  • Aim for >80% coverage on critical paths
  • Use markers: unit, integration, slow

Implementation Patterns

  • Follow Implementation Patterns
  • Use factory functions for agent/tool creation
  • Implement protocols for extensibility
  • Use singleton pattern with @lru_cache(maxsize=1)

Prompt Engineering

  • Follow Prompt Engineering Guidelines
  • Always validate citations
  • Use diverse evidence selection
  • Never trust LLM-generated citations without validation

Code Quality

  • Follow Code Quality Guidelines
  • Google-style docstrings for all public functions
  • Explain WHY, not WHAT in comments
  • Mark critical sections: # CRITICAL: ...

MCP Integration

MCP Tools

  • Functions in src/mcp_tools.py for Claude Desktop
  • Full type hints required
  • Google-style docstrings with Args/Returns sections
  • Formatted string returns (markdown)

Gradio MCP Server

  • Enable with mcp_server=True in demo.launch()
  • Endpoint: /gradio_api/mcp/
  • Use ssr_mode=False to fix hydration issues in HF Spaces

Common Pitfalls

  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Hallucinated citations: Always validate references
  4. Global mutable state: Use ContextVar or pass via parameters
  5. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
  6. Rate limiting: Always implement for external APIs
  7. Error chaining: Always use from e when raising exceptions

Key Principles

  1. Type Safety First: All code must pass mypy --strict
  2. Async Everything: All I/O must be async
  3. Test-Driven: Write tests before implementation
  4. No Hallucinations: Validate all citations
  5. Graceful Degradation: Support free tier (HF Inference) when no API keys
  6. Lazy Loading: Don't require optional dependencies at import time
  7. Structured Logging: Use structlog, never print()
  8. Error Chaining: Always preserve exception context

Pull Request Process

  1. Ensure all checks pass: make check
  2. Update documentation if needed
  3. Add tests for new features
  4. Update CHANGELOG if applicable
  5. Request review from maintainers
  6. Address review feedback
  7. Wait for approval before merging

Questions?

  • Open an issue on GitHub
  • Check existing documentation
  • Review code examples in the codebase

Thank you for contributing to DeepCritical!

\ No newline at end of file + Contributing to The DETERMINATOR - The DETERMINATOR
Skip to content

Contributing to The DETERMINATOR

Thank you for your interest in contributing to The DETERMINATOR! This guide will help you get started.

Note on Project Names: "The DETERMINATOR" is the product name, "DeepCritical" is the organization/project name, and "determinator" is the Python package name.

Git Workflow

  • main: Production-ready (GitHub)
  • dev: Development integration (GitHub)
  • Use feature branches: yourname-dev
  • NEVER push directly to main or dev on HuggingFace
  • GitHub is source of truth; HuggingFace is for deployment

Repository Information

Dual Repository Setup

This project uses a dual repository setup:

  • GitHub (DeepCritical/GradioDemo): Source of truth for code, PRs, and code review
  • HuggingFace (DataQuests/DeepCritical): Deployment target for the Gradio demo

Remote Configuration

When cloning, set up remotes as follows:

# Clone from GitHub
+git clone https://github.com/DeepCritical/GradioDemo.git
+cd GradioDemo
+
+# Add HuggingFace remote (optional, for deployment)
+git remote add huggingface-upstream https://huggingface.co/spaces/DataQuests/DeepCritical
+

Important: Never push directly to main or dev on HuggingFace. Always work through GitHub PRs. GitHub is the source of truth; HuggingFace is for deployment/demo only.

Package Manager

This project uses uv as the package manager. All commands should be prefixed with uv run to ensure they run in the correct environment.

Installation

# Install uv if you haven't already (recommended: standalone installer)
+# Unix/macOS/Linux:
+curl -LsSf https://astral.sh/uv/install.sh | sh
+
+# Windows (PowerShell):
+powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
+
+# Alternative: pipx install uv
+# Or: pip install uv
+
+# Sync all dependencies including dev extras
+uv sync --all-extras
+
+# Install pre-commit hooks
+uv run pre-commit install
+

Development Commands

# Installation
+uv sync --all-extras              # Install all dependencies including dev
+uv run pre-commit install          # Install pre-commit hooks
+
+# Code Quality Checks (run all before committing)
+uv run ruff check src tests       # Lint with ruff
+uv run ruff format src tests      # Format with ruff
+uv run mypy src                   # Type checking
+uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m "not openai" -p no:logfire  # Tests with coverage
+
+# Testing Commands
+uv run pytest tests/unit/ -v -m "not openai" -p no:logfire              # Run unit tests (excludes OpenAI tests)
+uv run pytest tests/ -v -m "huggingface" -p no:logfire                 # Run HuggingFace tests
+uv run pytest tests/ -v -p no:logfire                                  # Run all tests
+uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m "not openai" -p no:logfire  # Tests with terminal coverage
+uv run pytest --cov=src --cov-report=html -p no:logfire                # Generate HTML coverage report (opens htmlcov/index.html)
+
+# Documentation Commands
+uv run mkdocs build                # Build documentation
+uv run mkdocs serve                # Serve documentation locally (http://127.0.0.1:8000)
+

Test Markers

The project uses pytest markers to categorize tests. See Testing Guidelines for details:

  • unit: Unit tests (mocked, fast)
  • integration: Integration tests (real APIs)
  • slow: Slow tests
  • openai: Tests requiring OpenAI API key
  • huggingface: Tests requiring HuggingFace API key
  • embedding_provider: Tests requiring API-based embedding providers
  • local_embeddings: Tests using local embeddings

Note: The -p no:logfire flag disables the logfire plugin to avoid conflicts during testing.

Getting Started

  1. Fork the repository on GitHub: DeepCritical/GradioDemo

  2. Clone your fork:

git clone https://github.com/yourusername/GradioDemo.git
+cd GradioDemo
+
  1. Install dependencies:
uv sync --all-extras
+uv run pre-commit install
+
  1. Create a feature branch:
git checkout -b yourname-feature-name
+

  1. Make your changes following the guidelines below

  2. Run checks:

uv run ruff check src tests
+uv run mypy src
+uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m "not openai" -p no:logfire
+
  1. Commit and push:
git commit -m "Description of changes"
+git push origin yourname-feature-name
+
  1. Create a pull request on GitHub

Development Guidelines

Code Style

  • Follow Code Style Guidelines
  • All code must pass mypy --strict
  • Use ruff for linting and formatting
  • Line length: 100 characters

Error Handling

  • Follow Error Handling Guidelines
  • Always chain exceptions: raise SearchError(...) from e
  • Use structured logging with structlog
  • Never silently swallow exceptions

Testing

  • Follow Testing Guidelines
  • Write tests before implementation (TDD)
  • Aim for >80% coverage on critical paths
  • Use markers: unit, integration, slow

Implementation Patterns

  • Follow Implementation Patterns
  • Use factory functions for agent/tool creation
  • Implement protocols for extensibility
  • Use singleton pattern with @lru_cache(maxsize=1)

Prompt Engineering

  • Follow Prompt Engineering Guidelines
  • Always validate citations
  • Use diverse evidence selection
  • Never trust LLM-generated citations without validation

Code Quality

  • Follow Code Quality Guidelines
  • Google-style docstrings for all public functions
  • Explain WHY, not WHAT in comments
  • Mark critical sections: # CRITICAL: ...

MCP Integration

MCP Tools

  • Functions in src/mcp_tools.py for Claude Desktop
  • Full type hints required
  • Google-style docstrings with Args/Returns sections
  • Formatted string returns (markdown)

Gradio MCP Server

  • Enable with mcp_server=True in demo.launch()
  • Endpoint: /gradio_api/mcp/
  • Use ssr_mode=False to fix hydration issues in HF Spaces

Common Pitfalls

  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Hallucinated citations: Always validate references
  4. Global mutable state: Use ContextVar or pass via parameters
  5. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
  6. Rate limiting: Always implement for external APIs
  7. Error chaining: Always use from e when raising exceptions

Key Principles

  1. Type Safety First: All code must pass mypy --strict
  2. Async Everything: All I/O must be async
  3. Test-Driven: Write tests before implementation
  4. No Hallucinations: Validate all citations
  5. Graceful Degradation: Support free tier (HF Inference) when no API keys
  6. Lazy Loading: Don't require optional dependencies at import time
  7. Structured Logging: Use structlog, never print()
  8. Error Chaining: Always preserve exception context

Pull Request Process

  1. Ensure all checks pass: uv run ruff check src tests && uv run mypy src && uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m "not openai" -p no:logfire
  2. Update documentation if needed
  3. Add tests for new features
  4. Update CHANGELOG if applicable
  5. Request review from maintainers
  6. Address review feedback
  7. Wait for approval before merging

Project Structure

  • src/: Main source code
  • tests/: Test files (unit/ and integration/)
  • docs/: Documentation source files (MkDocs)
  • examples/: Example usage scripts
  • pyproject.toml: Project configuration and dependencies
  • .pre-commit-config.yaml: Pre-commit hook configuration

Questions?

Thank you for contributing to The DETERMINATOR!

\ No newline at end of file diff --git a/site/contributing/prompt-engineering/index.html b/site/contributing/prompt-engineering/index.html index af3f28e..b1298ff 100644 --- a/site/contributing/prompt-engineering/index.html +++ b/site/contributing/prompt-engineering/index.html @@ -1 +1 @@ - Prompt Engineering & Citation Validation - The DETERMINATOR
Skip to content

Prompt Engineering & Citation Validation

This document outlines prompt engineering guidelines and citation validation rules.

Judge Prompts

  • System prompt in src/prompts/judge.py
  • Format evidence with truncation (1500 chars per item)
  • Handle empty evidence case separately
  • Always request structured JSON output
  • Use format_user_prompt() and format_empty_evidence_prompt() helpers

Hypothesis Prompts

  • Use diverse evidence selection (MMR algorithm)
  • Sentence-aware truncation (truncate_at_sentence())
  • Format: Drug → Target → Pathway → Effect
  • System prompt emphasizes mechanistic reasoning
  • Use format_hypothesis_prompt() with embeddings for diversity

Report Prompts

  • Include full citation details for validation
  • Use diverse evidence selection (n=20)
  • CRITICAL: Emphasize citation validation rules
  • Format hypotheses with support/contradiction counts
  • System prompt includes explicit JSON structure requirements

Citation Validation

  • ALWAYS validate references before returning reports
  • Use validate_references() from src/utils/citation_validator.py
  • Remove hallucinated citations (URLs not in evidence)
  • Log warnings for removed citations
  • Never trust LLM-generated citations without validation

Citation Validation Rules

  1. Every reference URL must EXACTLY match a provided evidence URL
  2. Do NOT invent, fabricate, or hallucinate any references
  3. Do NOT modify paper titles, authors, dates, or URLs
  4. If unsure about a citation, OMIT it rather than guess
  5. Copy URLs exactly as provided - do not create similar-looking URLs

Evidence Selection

  • Use select_diverse_evidence() for MMR-based selection
  • Balance relevance vs diversity (lambda=0.7 default)
  • Sentence-aware truncation preserves meaning
  • Limit evidence per prompt to avoid context overflow

See Also

\ No newline at end of file + Prompt Engineering & Citation Validation - The DETERMINATOR
Skip to content

Prompt Engineering & Citation Validation

This document outlines prompt engineering guidelines and citation validation rules.

Judge Prompts

  • System prompt in src/prompts/judge.py
  • Format evidence with truncation (1500 chars per item)
  • Handle empty evidence case separately
  • Always request structured JSON output
  • Use format_user_prompt() and format_empty_evidence_prompt() helpers

Hypothesis Prompts

  • Use diverse evidence selection (MMR algorithm)
  • Sentence-aware truncation (truncate_at_sentence())
  • Format: Drug → Target → Pathway → Effect
  • System prompt emphasizes mechanistic reasoning
  • Use format_hypothesis_prompt() with embeddings for diversity

Report Prompts

  • Include full citation details for validation
  • Use diverse evidence selection (n=20)
  • CRITICAL: Emphasize citation validation rules
  • Format hypotheses with support/contradiction counts
  • System prompt includes explicit JSON structure requirements

Citation Validation

  • ALWAYS validate references before returning reports
  • Use validate_references() from src/utils/citation_validator.py
  • Remove hallucinated citations (URLs not in evidence)
  • Log warnings for removed citations
  • Never trust LLM-generated citations without validation

Citation Validation Rules

  1. Every reference URL must EXACTLY match a provided evidence URL
  2. Do NOT invent, fabricate, or hallucinate any references
  3. Do NOT modify paper titles, authors, dates, or URLs
  4. If unsure about a citation, OMIT it rather than guess
  5. Copy URLs exactly as provided - do not create similar-looking URLs

Evidence Selection

  • Use select_diverse_evidence() for MMR-based selection
  • Balance relevance vs diversity (lambda=0.7 default)
  • Sentence-aware truncation preserves meaning
  • Limit evidence per prompt to avoid context overflow

See Also

\ No newline at end of file diff --git a/site/contributing/testing/index.html b/site/contributing/testing/index.html index d34e59d..b840011 100644 --- a/site/contributing/testing/index.html +++ b/site/contributing/testing/index.html @@ -1,13 +1,37 @@ - Testing Requirements - The DETERMINATOR
Skip to content

Testing Requirements

This document outlines testing requirements and guidelines for DeepCritical.

Test Structure

  • Unit tests in tests/unit/ (mocked, fast)
  • Integration tests in tests/integration/ (real APIs, marked @pytest.mark.integration)
  • Use markers: unit, integration, slow

Mocking

  • Use respx for httpx mocking
  • Use pytest-mock for general mocking
  • Mock LLM calls in unit tests (use MockJudgeHandler)
  • Fixtures in tests/conftest.py: mock_httpx_client, mock_llm_response

TDD Workflow

  1. Write failing test in tests/unit/
  2. Implement in src/
  3. Ensure test passes
  4. Run make check (lint + typecheck + test)

Test Examples

@pytest.mark.unit
-async def test_pubmed_search(mock_httpx_client):
-    tool = PubMedTool()
-    results = await tool.search("metformin", max_results=5)
-    assert len(results) > 0
-    assert all(isinstance(r, Evidence) for r in results)
-
-@pytest.mark.integration
-async def test_real_pubmed_search():
-    tool = PubMedTool()
-    results = await tool.search("metformin", max_results=3)
-    assert len(results) <= 3
-

Test Coverage

  • Run make test-cov for coverage report
  • Aim for >80% coverage on critical paths
  • Exclude: __init__.py, TYPE_CHECKING blocks

See Also

\ No newline at end of file + Testing Requirements - The DETERMINATOR
Skip to content

Testing Requirements

This document outlines testing requirements and guidelines for The DETERMINATOR.

Test Structure

  • Unit tests in tests/unit/ (mocked, fast)
  • Integration tests in tests/integration/ (real APIs, marked @pytest.mark.integration)
  • Use markers: unit, integration, slow, openai, huggingface, embedding_provider, local_embeddings

Test Markers

The project uses pytest markers to categorize tests. These markers are defined in pyproject.toml:

  • @pytest.mark.unit: Unit tests (mocked, fast) - Run with -m "unit"
  • @pytest.mark.integration: Integration tests (real APIs) - Run with -m "integration"
  • @pytest.mark.slow: Slow tests - Run with -m "slow"
  • @pytest.mark.openai: Tests requiring OpenAI API key - Run with -m "openai" or exclude with -m "not openai"
  • @pytest.mark.huggingface: Tests requiring HuggingFace API key or using HuggingFace models - Run with -m "huggingface"
  • @pytest.mark.embedding_provider: Tests requiring API-based embedding providers (OpenAI, etc.) - Run with -m "embedding_provider"
  • @pytest.mark.local_embeddings: Tests using local embeddings (sentence-transformers, ChromaDB) - Run with -m "local_embeddings"

Running Tests by Marker

# Run only unit tests (excludes OpenAI tests by default)
+uv run pytest tests/unit/ -v -m "not openai" -p no:logfire
+
+# Run HuggingFace tests
+uv run pytest tests/ -v -m "huggingface" -p no:logfire
+
+# Run all tests
+uv run pytest tests/ -v -p no:logfire
+
+# Run only local embedding tests
+uv run pytest tests/ -v -m "local_embeddings" -p no:logfire
+
+# Exclude slow tests
+uv run pytest tests/ -v -m "not slow" -p no:logfire
+

Note: The -p no:logfire flag disables the logfire plugin to avoid conflicts during testing.

Mocking

  • Use respx for httpx mocking
  • Use pytest-mock for general mocking
  • Mock LLM calls in unit tests (use MockJudgeHandler)
  • Fixtures in tests/conftest.py: mock_httpx_client, mock_llm_response

TDD Workflow

  1. Write failing test in tests/unit/
  2. Implement in src/
  3. Ensure test passes
  4. Run checks: uv run ruff check src tests && uv run mypy src && uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m "not openai" -p no:logfire

Test Command Examples

# Run unit tests (default, excludes OpenAI tests)
+uv run pytest tests/unit/ -v -m "not openai" -p no:logfire
+
+# Run HuggingFace tests
+uv run pytest tests/ -v -m "huggingface" -p no:logfire
+
+# Run all tests
+uv run pytest tests/ -v -p no:logfire
+

Test Examples

@pytest.mark.unit
+async def test_pubmed_search(mock_httpx_client):
+    tool = PubMedTool()
+    results = await tool.search("metformin", max_results=5)
+    assert len(results) > 0
+    assert all(isinstance(r, Evidence) for r in results)
+
+@pytest.mark.integration
+async def test_real_pubmed_search():
+    tool = PubMedTool()
+    results = await tool.search("metformin", max_results=3)
+    assert len(results) <= 3
+

Test Coverage

Terminal Coverage Report

uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m "not openai" -p no:logfire
+

This shows coverage with missing lines highlighted in the terminal output.

HTML Coverage Report

uv run pytest --cov=src --cov-report=html -p no:logfire
+

This generates an HTML coverage report in htmlcov/index.html. Open this file in your browser to see detailed coverage information.

Coverage Goals

  • Aim for >80% coverage on critical paths
  • Exclude: __init__.py, TYPE_CHECKING blocks
  • Coverage configuration is in pyproject.toml under [tool.coverage.*]

See Also

\ No newline at end of file diff --git a/site/getting-started/examples/index.html b/site/getting-started/examples/index.html index 6ac4769..88d4916 100644 --- a/site/getting-started/examples/index.html +++ b/site/getting-started/examples/index.html @@ -1,49 +1,49 @@ - Examples - The DETERMINATOR
Skip to content

Examples

This page provides examples of using DeepCritical for various research tasks.

Basic Research Query

Example 1: Drug Information

Query: 

What are the latest treatments for Alzheimer's disease?
-

What DeepCritical Does: 1. Searches PubMed for recent papers 2. Searches ClinicalTrials.gov for active trials 3. Evaluates evidence quality 4. Synthesizes findings into a comprehensive report

Query: 

What clinical trials are investigating metformin for cancer prevention?
-

What DeepCritical Does: 1. Searches ClinicalTrials.gov for relevant trials 2. Searches PubMed for supporting literature 3. Provides trial details and status 4. Summarizes findings

Advanced Research Queries

Example 3: Comprehensive Review

Query: 

Review the evidence for using metformin as an anti-aging intervention, 
-including clinical trials, mechanisms of action, and safety profile.
-

What DeepCritical Does: 1. Uses deep research mode (multi-section) 2. Searches multiple sources in parallel 3. Generates sections on: - Clinical trials - Mechanisms of action - Safety profile 4. Synthesizes comprehensive report

Example 4: Hypothesis Testing

Query: 

Test the hypothesis that regular exercise reduces Alzheimer's disease risk.
-

What DeepCritical Does: 1. Generates testable hypotheses 2. Searches for supporting/contradicting evidence 3. Performs statistical analysis (if Modal configured) 4. Provides verdict: SUPPORTED, REFUTED, or INCONCLUSIVE

MCP Tool Examples

Using search_pubmed

Search PubMed for "CRISPR gene editing cancer therapy"
-

Using search_clinical_trials

Find active clinical trials for "diabetes type 2 treatment"
-

Using search_all

Search all sources for "COVID-19 vaccine side effects"
-

Using analyze_hypothesis

Analyze whether vitamin D supplementation reduces COVID-19 severity
-

Code Examples

Python API Usage

from src.orchestrator_factory import create_orchestrator
-from src.tools.search_handler import SearchHandler
-from src.agent_factory.judges import create_judge_handler
-
-# Create orchestrator
-search_handler = SearchHandler()
-judge_handler = create_judge_handler()
-
# Run research query
-query = "What are the latest treatments for Alzheimer's disease?"
-async for event in orchestrator.run(query):
-    print(f"Event: {event.type} - {event.data}")
-

Gradio UI Integration

import gradio as gr
-from src.app import create_research_interface
-
-# Create interface
-interface = create_research_interface()
-
-# Launch
-interface.launch(server_name="0.0.0.0", server_port=7860)
-

Research Patterns

Iterative Research

Single-loop research with search-judge-synthesize cycles:

from src.orchestrator.research_flow import IterativeResearchFlow
-
async for event in flow.run(query):
-    # Handle events
-    pass
-

Deep Research

Multi-section parallel research:

from src.orchestrator.research_flow import DeepResearchFlow
-
async for event in flow.run(query):
-    # Handle events
-    pass
-

Configuration Examples

Basic Configuration

# .env file
-LLM_PROVIDER=openai
-OPENAI_API_KEY=your_key_here
-MAX_ITERATIONS=10
-

Advanced Configuration

# .env file
-LLM_PROVIDER=anthropic
-ANTHROPIC_API_KEY=your_key_here
-EMBEDDING_PROVIDER=local
-WEB_SEARCH_PROVIDER=duckduckgo
-MAX_ITERATIONS=20
-DEFAULT_TOKEN_LIMIT=200000
-USE_GRAPH_EXECUTION=true
-

Next Steps

\ No newline at end of file + Examples - The DETERMINATOR
Skip to content

Examples

This page provides examples of using The DETERMINATOR for various research tasks.

Basic Research Query

Example 1: Drug Information

Query: 

What are the latest treatments for Alzheimer's disease?
+

What The DETERMINATOR Does: 1. Searches PubMed for recent papers 2. Searches ClinicalTrials.gov for active trials 3. Evaluates evidence quality 4. Synthesizes findings into a comprehensive report

Query: 

What clinical trials are investigating metformin for cancer prevention?
+

What The DETERMINATOR Does:

  1. Searches ClinicalTrials.gov for relevant trials
  2. Searches PubMed for supporting literature
  3. Provides trial details and status
  4. Summarizes findings

Advanced Research Queries

Example 3: Comprehensive Review

Query:

Review the evidence for using metformin as an anti-aging intervention, 
+including clinical trials, mechanisms of action, and safety profile.
+

What The DETERMINATOR Does: 1. Uses deep research mode (multi-section) 2. Searches multiple sources in parallel 3. Generates sections on: - Clinical trials - Mechanisms of action - Safety profile 4. Synthesizes comprehensive report

Example 4: Hypothesis Testing

Query: 

Test the hypothesis that regular exercise reduces Alzheimer's disease risk.
+

What The DETERMINATOR Does: 1. Generates testable hypotheses 2. Searches for supporting/contradicting evidence 3. Performs statistical analysis (if Modal configured) 4. Provides verdict: SUPPORTED, REFUTED, or INCONCLUSIVE

MCP Tool Examples

Using search_pubmed

Search PubMed for "CRISPR gene editing cancer therapy"
+

Using search_clinical_trials

Find active clinical trials for "diabetes type 2 treatment"
+

Using search_all

Search all sources for "COVID-19 vaccine side effects"
+

Using analyze_hypothesis

Analyze whether vitamin D supplementation reduces COVID-19 severity
+

Code Examples

Python API Usage

from src.orchestrator_factory import create_orchestrator
+from src.tools.search_handler import SearchHandler
+from src.agent_factory.judges import create_judge_handler
+
+# Create orchestrator
+search_handler = SearchHandler()
+judge_handler = create_judge_handler()
+
# Run research query
+query = "What are the latest treatments for Alzheimer's disease?"
+async for event in orchestrator.run(query):
+    print(f"Event: {event.type} - {event.data}")
+

Gradio UI Integration

import gradio as gr
+from src.app import create_research_interface
+
+# Create interface
+interface = create_research_interface()
+
+# Launch
+interface.launch(server_name="0.0.0.0", server_port=7860)
+

Research Patterns

Iterative Research

Single-loop research with search-judge-synthesize cycles:

from src.orchestrator.research_flow import IterativeResearchFlow
+
async for event in flow.run(query):
+    # Handle events
+    pass
+

Deep Research

Multi-section parallel research:

from src.orchestrator.research_flow import DeepResearchFlow
+
async for event in flow.run(query):
+    # Handle events
+    pass
+

Configuration Examples

Basic Configuration

# .env file
+LLM_PROVIDER=openai
+OPENAI_API_KEY=your_key_here
+MAX_ITERATIONS=10
+

Advanced Configuration

# .env file
+LLM_PROVIDER=anthropic
+ANTHROPIC_API_KEY=your_key_here
+EMBEDDING_PROVIDER=local
+WEB_SEARCH_PROVIDER=duckduckgo
+MAX_ITERATIONS=20
+DEFAULT_TOKEN_LIMIT=200000
+USE_GRAPH_EXECUTION=true
+

Next Steps

\ No newline at end of file diff --git a/site/getting-started/installation/index.html b/site/getting-started/installation/index.html index 77fdfbc..c702d58 100644 --- a/site/getting-started/installation/index.html +++ b/site/getting-started/installation/index.html @@ -1,21 +1,27 @@ - Installation - The DETERMINATOR
Skip to content

Installation

This guide will help you install and set up DeepCritical on your system.

Prerequisites

  • Python 3.11 or higher
  • uv package manager (recommended) or pip
  • At least one LLM API key (OpenAI, Anthropic, or HuggingFace)

Installation Steps

uv is a fast Python package installer and resolver. Install it with:

pip install uv
-

2. Clone the Repository

git clone https://github.com/DeepCritical/GradioDemo.git
-cd GradioDemo
-

3. Install Dependencies

Using uv (recommended):

uv sync
-

Using pip:

pip install -e .
-

4. Install Optional Dependencies

For embeddings support (local sentence-transformers):

uv sync --extra embeddings
-

For Modal sandbox execution:

uv sync --extra modal
-

For Magentic orchestration:

uv sync --extra magentic
-

Install all extras:

uv sync --all-extras
-

5. Configure Environment Variables

Create a .env file in the project root:

# Required: At least one LLM provider
-LLM_PROVIDER=openai  # or "anthropic" or "huggingface"
-OPENAI_API_KEY=your_openai_api_key_here
-
-# Optional: Other services
-NCBI_API_KEY=your_ncbi_api_key_here  # For higher PubMed rate limits
-MODAL_TOKEN_ID=your_modal_token_id
-MODAL_TOKEN_SECRET=your_modal_token_secret
-

See the Configuration Guide for all available options.

6. Verify Installation

Run the application:

uv run gradio run src/app.py
-

Open your browser to http://localhost:7860 to verify the installation.

Development Setup

For development, install dev dependencies:

uv sync --all-extras --dev
-

Install pre-commit hooks:

uv run pre-commit install
-

Troubleshooting

Common Issues

Import Errors: - Ensure you've installed all required dependencies - Check that Python 3.11+ is being used

API Key Errors: - Verify your .env file is in the project root - Check that API keys are correctly formatted - Ensure at least one LLM provider is configured

Module Not Found: - Run uv sync or pip install -e . again - Check that you're in the correct virtual environment

Port Already in Use: - Change the port in src/app.py or use environment variable - Kill the process using port 7860

Next Steps

<<<<<<< Updated upstream

<<<<<<< Updated upstream

=======

Stashed changes

Stashed changes

\ No newline at end of file + Installation - The DETERMINATOR
Skip to content

Installation

This guide will help you install and set up DeepCritical on your system.

Prerequisites

  • Python 3.11 or higher
  • uv package manager (recommended) or pip
  • At least one LLM API key (OpenAI, Anthropic, or HuggingFace)

Installation Steps

uv is a fast Python package installer and resolver. Install it using the standalone installer (recommended):

Unix/macOS/Linux: 

curl -LsSf https://astral.sh/uv/install.sh | sh
+

Windows (PowerShell): 

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
+

Alternative methods: 

# Using pipx (recommended if you have pipx installed)
+pipx install uv
+
+# Or using pip
+pip install uv
+

After installation, restart your terminal or add ~/.cargo/bin to your PATH.

2. Clone the Repository

git clone https://github.com/DeepCritical/GradioDemo.git
+cd GradioDemo
+

3. Install Dependencies

Using uv (recommended):

uv sync
+

Using pip:

pip install -e .
+

4. Install Optional Dependencies

For embeddings support (local sentence-transformers):

uv sync --extra embeddings
+

For Modal sandbox execution:

uv sync --extra modal
+

For Magentic orchestration:

uv sync --extra magentic
+

Install all extras:

uv sync --all-extras
+

5. Configure Environment Variables

Create a .env file in the project root:

# Required: At least one LLM provider
+LLM_PROVIDER=openai  # or "anthropic" or "huggingface"
+OPENAI_API_KEY=your_openai_api_key_here
+
+# Optional: Other services
+NCBI_API_KEY=your_ncbi_api_key_here  # For higher PubMed rate limits
+MODAL_TOKEN_ID=your_modal_token_id
+MODAL_TOKEN_SECRET=your_modal_token_secret
+

See the Configuration Guide for all available options.

6. Verify Installation

Run the application:

uv run gradio run src/app.py
+

Open your browser to http://localhost:7860 to verify the installation.

Development Setup

For development, install dev dependencies:

uv sync --all-extras --dev
+

Install pre-commit hooks:

uv run pre-commit install
+

Troubleshooting

Common Issues

Import Errors: - Ensure you've installed all required dependencies - Check that Python 3.11+ is being used

API Key Errors: - Verify your .env file is in the project root - Check that API keys are correctly formatted - Ensure at least one LLM provider is configured

Module Not Found: - Run uv sync or pip install -e . again - Check that you're in the correct virtual environment

Port Already in Use: - Change the port in src/app.py or use environment variable - Kill the process using port 7860

Next Steps

\ No newline at end of file diff --git a/site/getting-started/mcp-integration/index.html b/site/getting-started/mcp-integration/index.html index 157a588..9c98dcd 100644 --- a/site/getting-started/mcp-integration/index.html +++ b/site/getting-started/mcp-integration/index.html @@ -1,35 +1,35 @@ - MCP Integration - The DETERMINATOR
Skip to content

MCP Integration

DeepCritical exposes a Model Context Protocol (MCP) server, allowing you to use its search tools directly from Claude Desktop or other MCP clients.

What is MCP?

The Model Context Protocol (MCP) is a standard for connecting AI assistants to external tools and data sources. DeepCritical implements an MCP server that exposes its search capabilities as MCP tools.

MCP Server URL

When running locally:

http://localhost:7860/gradio_api/mcp/
-

Claude Desktop Configuration

1. Locate Configuration File

macOS: 

~/Library/Application Support/Claude/claude_desktop_config.json
-

Windows: 

%APPDATA%\Claude\claude_desktop_config.json
-

Linux: 

~/.config/Claude/claude_desktop_config.json
-

2. Add DeepCritical Server

Edit claude_desktop_config.json and add:

{
-  "mcpServers": {
-    "deepcritical": {
-      "url": "http://localhost:7860/gradio_api/mcp/"
-    }
-  }
-}
-

3. Restart Claude Desktop

Close and restart Claude Desktop for changes to take effect.

4. Verify Connection

In Claude Desktop, you should see DeepCritical tools available: - search_pubmed - search_clinical_trials - search_biorxiv - search_all - analyze_hypothesis

Available Tools

search_pubmed

Search peer-reviewed biomedical literature from PubMed.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search PubMed for "metformin diabetes"
-

search_clinical_trials

Search ClinicalTrials.gov for interventional studies.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search clinical trials for "Alzheimer's disease treatment"
-

search_biorxiv

Search bioRxiv/medRxiv preprints via Europe PMC.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search bioRxiv for "CRISPR gene editing"
-

search_all

Search all sources simultaneously (PubMed, ClinicalTrials.gov, Europe PMC).

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results per source (default: 10)

Example: 

Search all sources for "COVID-19 vaccine efficacy"
-

analyze_hypothesis

Perform secure statistical analysis using Modal sandboxes.

Parameters: - hypothesis (string): Hypothesis to analyze - data (string, optional): Data description or code

Example: 

Analyze the hypothesis that metformin reduces cancer risk
-

Using Tools in Claude Desktop

Once configured, you can ask Claude to use DeepCritical tools:

Use DeepCritical to search PubMed for recent papers on Alzheimer's disease treatments.
-

Claude will automatically: 1. Call the appropriate DeepCritical tool 2. Retrieve results 3. Use the results in its response

Troubleshooting

Connection Issues

Server Not Found: - Ensure DeepCritical is running (uv run gradio run src/app.py) - Verify the URL in claude_desktop_config.json is correct - Check that port 7860 is not blocked by firewall

Tools Not Appearing: - Restart Claude Desktop after configuration changes - Check Claude Desktop logs for errors - Verify MCP server is accessible at the configured URL

Authentication

If DeepCritical requires authentication: - Configure API keys in DeepCritical settings - Use HuggingFace OAuth login - Ensure API keys are valid

Advanced Configuration

Custom Port

If running on a different port, update the URL:

{
-  "mcpServers": {
-    "deepcritical": {
-      "url": "http://localhost:8080/gradio_api/mcp/"
-    }
-  }
-}
-

Multiple Instances

You can configure multiple DeepCritical instances:

{
-  "mcpServers": {
-    "deepcritical-local": {
-      "url": "http://localhost:7860/gradio_api/mcp/"
-    },
-    "deepcritical-remote": {
-      "url": "https://your-server.com/gradio_api/mcp/"
-    }
-  }
-}
-

Next Steps

\ No newline at end of file + MCP Integration - The DETERMINATOR
Skip to content

MCP Integration

The DETERMINATOR exposes a Model Context Protocol (MCP) server, allowing you to use its search tools directly from Claude Desktop or other MCP clients.

What is MCP?

The Model Context Protocol (MCP) is a standard for connecting AI assistants to external tools and data sources. The DETERMINATOR implements an MCP server that exposes its search capabilities as MCP tools.

MCP Server URL

When running locally:

http://localhost:7860/gradio_api/mcp/
+

Claude Desktop Configuration

1. Locate Configuration File

macOS: 

~/Library/Application Support/Claude/claude_desktop_config.json
+

Windows: 

%APPDATA%\Claude\claude_desktop_config.json
+

Linux: 

~/.config/Claude/claude_desktop_config.json
+

2. Add The DETERMINATOR Server

Edit claude_desktop_config.json and add:

{
+  "mcpServers": {
+    "determinator": {
+      "url": "http://localhost:7860/gradio_api/mcp/"
+    }
+  }
+}
+

3. Restart Claude Desktop

Close and restart Claude Desktop for changes to take effect.

4. Verify Connection

In Claude Desktop, you should see The DETERMINATOR tools available: - search_pubmed - search_clinical_trials - search_biorxiv - search_all - analyze_hypothesis

Available Tools

search_pubmed

Search peer-reviewed biomedical literature from PubMed.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search PubMed for "metformin diabetes"
+

search_clinical_trials

Search ClinicalTrials.gov for interventional studies.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search clinical trials for "Alzheimer's disease treatment"
+

search_biorxiv

Search bioRxiv/medRxiv preprints via Europe PMC.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search bioRxiv for "CRISPR gene editing"
+

search_all

Search all sources simultaneously (PubMed, ClinicalTrials.gov, Europe PMC).

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results per source (default: 10)

Example: 

Search all sources for "COVID-19 vaccine efficacy"
+

analyze_hypothesis

Perform secure statistical analysis using Modal sandboxes.

Parameters: - hypothesis (string): Hypothesis to analyze - data (string, optional): Data description or code

Example: 

Analyze the hypothesis that metformin reduces cancer risk
+

Using Tools in Claude Desktop

Once configured, you can ask Claude to use DeepCritical tools:

Use DeepCritical to search PubMed for recent papers on Alzheimer's disease treatments.
+

Claude will automatically: 1. Call the appropriate DeepCritical tool 2. Retrieve results 3. Use the results in its response

Troubleshooting

Connection Issues

Server Not Found: - Ensure DeepCritical is running (uv run gradio run src/app.py) - Verify the URL in claude_desktop_config.json is correct - Check that port 7860 is not blocked by firewall

Tools Not Appearing: - Restart Claude Desktop after configuration changes - Check Claude Desktop logs for errors - Verify MCP server is accessible at the configured URL

Authentication

If DeepCritical requires authentication: - Configure API keys in DeepCritical settings - Use HuggingFace OAuth login - Ensure API keys are valid

Advanced Configuration

Custom Port

If running on a different port, update the URL:

{
+  "mcpServers": {
+    "deepcritical": {
+      "url": "http://localhost:8080/gradio_api/mcp/"
+    }
+  }
+}
+

Multiple Instances

You can configure multiple DeepCritical instances:

{
+  "mcpServers": {
+    "deepcritical-local": {
+      "url": "http://localhost:7860/gradio_api/mcp/"
+    },
+    "deepcritical-remote": {
+      "url": "https://your-server.com/gradio_api/mcp/"
+    }
+  }
+}
+

Next Steps

\ No newline at end of file diff --git a/site/getting-started/quick-start/index.html b/site/getting-started/quick-start/index.html index 2b66ea1..f3fadf4 100644 --- a/site/getting-started/quick-start/index.html +++ b/site/getting-started/quick-start/index.html @@ -1,6 +1,44 @@ - Quick Start Guide - The DETERMINATOR
Skip to content

Quick Start Guide

Get up and running with DeepCritical in minutes.

Start the Application

uv run gradio run src/app.py
-

Open your browser to http://localhost:7860.

First Research Query

  1. Enter a Research Question

Type your research question in the chat interface, for example: - "What are the latest treatments for Alzheimer's disease?" - "Review the evidence for metformin in cancer prevention" - "What clinical trials are investigating COVID-19 vaccines?"

  1. Submit the Query

Click "Submit" or press Enter. The system will: - Generate observations about your query - Identify knowledge gaps - Search multiple sources (PubMed, ClinicalTrials.gov, Europe PMC) - Evaluate evidence quality - Synthesize findings into a report

  1. Review Results

Watch the real-time progress in the chat interface: - Search operations and results - Evidence evaluation - Report generation - Final research report with citations

Authentication

  1. Click "Sign in with HuggingFace" at the top of the app
  2. Authorize the application
  3. Your HuggingFace API token will be automatically used
  4. No need to manually enter API keys

Manual API Key

  1. Open the Settings accordion
  2. Enter your API key:
  3. OpenAI API key
  4. Anthropic API key
  5. HuggingFace API key
  6. Click "Save Settings"
  7. Manual keys take priority over OAuth tokens

Understanding the Interface

Chat Interface

  • Input: Enter your research questions here
  • Messages: View conversation history and research progress
  • Streaming: Real-time updates as research progresses

Status Indicators

  • Searching: Active search operations
  • Evaluating: Evidence quality assessment
  • Synthesizing: Report generation
  • Complete: Research finished

Settings

  • API Keys: Configure LLM providers
  • Research Mode: Choose iterative or deep research
  • Budget Limits: Set token, time, and iteration limits

Example Queries

Simple Query

What are the side effects of metformin?
-

Complex Query

Review the evidence for using metformin as an anti-aging intervention, 
-including clinical trials, mechanisms of action, and safety profile.
-

Clinical Trial Query

What are the active clinical trials investigating Alzheimer's disease treatments?
-

Next Steps

\ No newline at end of file + Single Command Deploy - The DETERMINATOR
Skip to content

Single Command Deploy

Deploy with docker instandly with a single command :

docker run -it -p 7860:7860 --platform=linux/amd64 \
+    -e DB_KEY="YOUR_VALUE_HERE" \
+    -e SERP_API="YOUR_VALUE_HERE" \
+    -e INFERENCE_API="YOUR_VALUE_HERE" \
+    -e MODAL_TOKEN_ID="YOUR_VALUE_HERE" \
+    -e MODAL_TOKEN_SECRET="YOUR_VALUE_HERE" \
+    -e NCBI_API_KEY="YOUR_VALUE_HERE" \
+    -e SERPER_API_KEY="YOUR_VALUE_HERE" \
+    -e CHROMA_DB_PATH="./chroma_db" \
+    -e CHROMA_DB_HOST="localhost" \
+    -e CHROMA_DB_PORT="8000" \
+    -e RAG_COLLECTION_NAME="deepcritical_evidence" \
+    -e RAG_SIMILARITY_TOP_K="5" \
+    -e RAG_AUTO_INGEST="true" \
+    -e USE_GRAPH_EXECUTION="false" \
+    -e DEFAULT_TOKEN_LIMIT="100000" \
+    -e DEFAULT_TIME_LIMIT_MINUTES="10" \
+    -e DEFAULT_ITERATIONS_LIMIT="10" \
+    -e WEB_SEARCH_PROVIDER="duckduckgo" \
+    -e MAX_ITERATIONS="10" \
+    -e SEARCH_TIMEOUT="30" \
+    -e LOG_LEVEL="DEBUG" \
+    -e EMBEDDING_PROVIDER="local" \
+    -e OPENAI_EMBEDDING_MODEL="text-embedding-3-small" \
+    -e LOCAL_EMBEDDING_MODEL="BAAI/bge-small-en-v1.5" \
+    -e HUGGINGFACE_EMBEDDING_MODEL="sentence-transformers/all-MiniLM-L6-v2" \
+    -e HF_FALLBACK_MODELS="Qwen/Qwen3-Next-80B-A3B-Thinking,Qwen/Qwen3-Next-80B-A3B-Instruct,meta-llama/Llama-3.3-70B-Instruct,meta-llama/Llama-3.1-8B-Instruct,HuggingFaceH4/zephyr-7b-beta,Qwen/Qwen2-7B-Instruct" \
+    -e HUGGINGFACE_MODEL="Qwen/Qwen3-Next-80B-A3B-Thinking" \
+    registry.hf.space/dataquests-deepcritical:latest python src/app.py
+   ```
+
+## Quick start guide
+
+Get up and running with The DETERMINATOR in minutes.
+
+## Start the Application
+
+```bash
+gradio src/app.py
+

Open your browser to http://localhost:7860.

First Research Query

  1. Enter a Research Question

Type your research question in the chat interface, for example: - "What are the latest treatments for Alzheimer's disease?" - "Review the evidence for metformin in cancer prevention" - "What clinical trials are investigating COVID-19 vaccines?"

  1. Submit the Query

Click "Submit" or press Enter. The system will: - Generate observations about your query - Identify knowledge gaps - Search multiple sources (PubMed, ClinicalTrials.gov, Europe PMC) - Evaluate evidence quality - Synthesize findings into a report

  1. Review Results

Watch the real-time progress in the chat interface: - Search operations and results - Evidence evaluation - Report generation - Final research report with citations

Authentication

  1. Click "Sign in with HuggingFace" at the top of the app
  2. Authorize the application
  3. Your HuggingFace API token will be automatically used
  4. No need to manually enter API keys

Manual API Key

  1. Open the Settings accordion
  2. Enter your API key:
  3. OpenAI API key
  4. Anthropic API key
  5. HuggingFace API key
  6. Click "Save Settings"
  7. Manual keys take priority over OAuth tokens

Understanding the Interface

Chat Interface

  • Input: Enter your research questions here
  • Messages: View conversation history and research progress
  • Streaming: Real-time updates as research progresses

Status Indicators

  • Searching: Active search operations
  • Evaluating: Evidence quality assessment
  • Synthesizing: Report generation
  • Complete: Research finished

Settings

  • API Keys: Configure LLM providers
  • Research Mode: Choose iterative or deep research
  • Budget Limits: Set token, time, and iteration limits

Example Queries

Simple Query

What are the side effects of metformin?
+

Complex Query

Review the evidence for using metformin as an anti-aging intervention, 
+including clinical trials, mechanisms of action, and safety profile.
+

Clinical Trial Query

What are the active clinical trials investigating Alzheimer's disease treatments?
+

Next Steps

\ No newline at end of file diff --git a/site/index.html b/site/index.html index 7ce7805..5e2f78d 100644 --- a/site/index.html +++ b/site/index.html @@ -1,9 +1,16 @@ - The DETERMINATOR
Skip to content

DeepCritical

AI-Native Drug Repurposing Research Agent

DeepCritical is a deep research agent system that uses iterative search-and-judge loops to comprehensively answer research questions. The system supports multiple orchestration patterns, graph-based execution, parallel research workflows, and long-running task management with real-time streaming.

Features

  • Multi-Source Search: PubMed, ClinicalTrials.gov, Europe PMC (includes bioRxiv/medRxiv)
  • MCP Integration: Use our tools from Claude Desktop or any MCP client
  • HuggingFace OAuth: Sign in with your HuggingFace account to automatically use your API token
  • Modal Sandbox: Secure execution of AI-generated statistical code
  • LlamaIndex RAG: Semantic search and evidence synthesis
  • HuggingFace Inference: Free tier support with automatic fallback
  • Strongly Typed Composable Graphs: Graph-based orchestration with Pydantic AI
  • Specialized Research Teams of Agents: Multi-agent coordination for complex research tasks

Quick Start

# Install uv if you haven't already
-pip install uv
-
-# Sync dependencies
-uv sync
-
-# Start the Gradio app
-uv run gradio run src/app.py
-

Open your browser to http://localhost:7860.

For detailed installation and setup instructions, see the Getting Started Guide.

Architecture

DeepCritical uses a Vertical Slice Architecture:

  1. Search Slice: Retrieving evidence from PubMed, ClinicalTrials.gov, and Europe PMC
  2. Judge Slice: Evaluating evidence quality using LLMs
  3. Orchestrator Slice: Managing the research loop and UI

The system supports three main research patterns:

  • Iterative Research: Single research loop with search-judge-synthesize cycles
  • Deep Research: Multi-section parallel research with planning and synthesis
  • Research Team: Multi-agent coordination using Magentic framework

Learn more about the Architecture.

Documentation

\ No newline at end of file + The DETERMINATOR
Skip to content

The DETERMINATOR

Generalist Deep Research Agent - Stops at Nothing Until Finding Precise Answers

The DETERMINATOR is a powerful generalist deep research agent system that uses iterative search-and-judge loops to comprehensively investigate any research question. It stops at nothing until finding precise answers, only stopping at configured limits (budget, time, iterations).

Key Features: - Generalist: Handles queries from any domain (medical, technical, business, scientific, etc.) - Automatic Source Selection: Automatically determines if medical knowledge sources (PubMed, ClinicalTrials.gov) are needed - Multi-Source Search: Web search, PubMed, ClinicalTrials.gov, Europe PMC, RAG - Iterative Refinement: Continues searching and refining until precise answers are found - Evidence Synthesis: Comprehensive reports with proper citations

Important: The DETERMINATOR is a research tool that synthesizes evidence. It cannot provide medical advice or answer medical questions directly.

Features

  • Generalist Research: Handles any research question from any domain
  • Automatic Medical Detection: Automatically determines if medical knowledge sources are needed
  • Multi-Source Search: Web search, PubMed, ClinicalTrials.gov, Europe PMC (includes bioRxiv/medRxiv), RAG
  • Iterative Until Precise: Stops at nothing until finding precise answers (only stops at configured limits)
  • MCP Integration: Use our tools from Claude Desktop or any MCP client
  • HuggingFace OAuth: Sign in with your HuggingFace account to automatically use your API token
  • Modal Sandbox: Secure execution of AI-generated statistical code
  • LlamaIndex RAG: Semantic search and evidence synthesis
  • HuggingFace Inference: Free tier support with automatic fallback
  • Strongly Typed Composable Graphs: Graph-based orchestration with Pydantic AI
  • Specialized Research Teams of Agents: Multi-agent coordination for complex research tasks

Quick Start

# Install uv if you haven't already (recommended: standalone installer)
+# Unix/macOS/Linux:
+curl -LsSf https://astral.sh/uv/install.sh | sh
+
+# Windows (PowerShell):
+powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
+
+# Alternative: pipx install uv
+# Or: pip install uv
+
+# Sync dependencies
+uv sync
+
+# Start the Gradio app
+uv run gradio run src/app.py
+

Open your browser to http://localhost:7860.

For detailed installation and setup instructions, see the Getting Started Guide.

Architecture

The DETERMINATOR uses a Vertical Slice Architecture:

  1. Search Slice: Retrieving evidence from multiple sources (web, PubMed, ClinicalTrials.gov, Europe PMC, RAG) based on query analysis
  2. Judge Slice: Evaluating evidence quality using LLMs
  3. Orchestrator Slice: Managing the research loop and UI

The system supports three main research patterns:

  • Iterative Research: Single research loop with search-judge-synthesize cycles
  • Deep Research: Multi-section parallel research with planning and synthesis
  • Research Team: Multi-agent coordination using Magentic framework

Learn more about the Architecture.

Documentation

\ No newline at end of file diff --git a/site/license/index.html b/site/license/index.html index 277a05d..a352fec 100644 --- a/site/license/index.html +++ b/site/license/index.html @@ -1 +1 @@ - License - The DETERMINATOR
Skip to content

License

DeepCritical is licensed under the MIT License.

MIT License

Copyright (c) 2024 DeepCritical Team

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

\ No newline at end of file + License - The DETERMINATOR
Skip to content

License

DeepCritical is licensed under the MIT License.

MIT License

Copyright (c) 2024 DeepCritical Team

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

\ No newline at end of file diff --git a/site/overview/architecture/index.html b/site/overview/architecture/index.html index 8a6eedd..b7a56c1 100644 --- a/site/overview/architecture/index.html +++ b/site/overview/architecture/index.html @@ -1 +1 @@ - Architecture Overview - The DETERMINATOR
Skip to content

Architecture Overview

DeepCritical is a deep research agent system that uses iterative search-and-judge loops to comprehensively answer research questions. The system supports multiple orchestration patterns, graph-based execution, parallel research workflows, and long-running task management with real-time streaming.

Core Architecture

Orchestration Patterns

  1. Graph Orchestrator (src/orchestrator/graph_orchestrator.py):
  2. Graph-based execution using Pydantic AI agents as nodes
  3. Supports both iterative and deep research patterns
  4. Node types: Agent, State, Decision, Parallel
  5. Edge types: Sequential, Conditional, Parallel
  6. Conditional routing based on knowledge gaps, budget, and iterations
  7. Parallel execution for concurrent research loops
  8. Event streaming via AsyncGenerator[AgentEvent] for real-time UI updates

  9. Fallback to agent chains when graph execution is disabled

  10. Deep Research Flow (src/orchestrator/research_flow.py):

  11. Pattern: Planner → Parallel Iterative Loops (one per section) → Synthesis
  12. Uses PlannerAgent to break query into report sections
  13. Runs IterativeResearchFlow instances in parallel per section via WorkflowManager
  14. Synthesizes results using LongWriterAgent or ProofreaderAgent
  15. Supports both graph execution (use_graph=True) and agent chains (use_graph=False)
  16. Budget tracking per section and globally

  17. State synchronization across parallel loops

  18. Iterative Research Flow (src/orchestrator/research_flow.py):

  19. Pattern: Generate observations → Evaluate gaps → Select tools → Execute → Judge → Continue/Complete
  20. Uses KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent, WriterAgent
  21. JudgeHandler assesses evidence sufficiency
  22. Iterates until research complete or constraints met (iterations, time, tokens)

  23. Supports graph execution and agent chains

  24. Magentic Orchestrator (src/orchestrator_magentic.py):

  25. Multi-agent coordination using agent-framework-core
  26. ChatAgent pattern with internal LLMs per agent
  27. Uses MagenticBuilder with participants: searcher, hypothesizer, judge, reporter
  28. Manager orchestrates agents via OpenAIChatClient
  29. Requires OpenAI API key (function calling support)
  30. Event-driven: converts Magentic events to AgentEvent for UI streaming

  31. Supports long-running workflows with max rounds and stall/reset handling

  32. Hierarchical Orchestrator (src/orchestrator_hierarchical.py):

  33. Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge
  34. Adapts Magentic ChatAgent to SubIterationTeam protocol
  35. Event-driven via asyncio.Queue for coordination

  36. Supports sub-iteration patterns for complex research tasks

  37. Legacy Simple Mode (src/legacy_orchestrator.py):

  38. Linear search-judge-synthesize loop
  39. Uses SearchHandlerProtocol and JudgeHandlerProtocol
  40. Generator-based design yielding AgentEvent objects
  41. Backward compatibility for simple use cases

Long-Running Task Support

The system is designed for long-running research tasks with comprehensive state management and streaming:

  1. Event Streaming:
  2. All orchestrators yield AgentEvent objects via AsyncGenerator
  3. Real-time UI updates through Gradio chat interface
  4. Event types: started, searching, search_complete, judging, judge_complete, looping, synthesizing, hypothesizing, complete, error

  5. Metadata includes iteration numbers, tool names, result counts, durations

  6. Budget Tracking (src/middleware/budget_tracker.py):

  7. Per-loop and global budget management
  8. Tracks: tokens, time (seconds), iterations
  9. Budget enforcement at decision nodes
  10. Token estimation (~4 chars per token)
  11. Early termination when budgets exceeded

  12. Budget summaries for monitoring

  13. Workflow Manager (src/middleware/workflow_manager.py):

  14. Coordinates parallel research loops
  15. Tracks loop status: pending, running, completed, failed, cancelled
  16. Synchronizes evidence between loops and global state
  17. Handles errors per loop (doesn't fail all if one fails)
  18. Supports loop cancellation and timeout handling

  19. Evidence deduplication across parallel loops

  20. State Management (src/middleware/state_machine.py):

  21. Thread-safe isolation using ContextVar for concurrent requests
  22. WorkflowState tracks: evidence, conversation history, embedding service
  23. Evidence deduplication by URL
  24. Semantic search via embedding service
  25. State persistence across long-running workflows

  26. Supports both iterative and deep research patterns

  27. Gradio UI (src/app.py):

  28. Real-time streaming of research progress
  29. Accordion-based UI for pending/done operations
  30. OAuth integration (HuggingFace)
  31. Multiple backend support (API keys, free tier)
  32. Handles long-running tasks with progress indicators
  33. Event accumulation for pending operations

Graph Architecture

The graph orchestrator (src/orchestrator/graph_orchestrator.py) implements a flexible graph-based execution model:

Node Types:

  • Agent Nodes: Execute Pydantic AI agents (e.g., KnowledgeGapAgent, ToolSelectorAgent)
  • State Nodes: Update or read workflow state (evidence, conversation)
  • Decision Nodes: Make routing decisions (research complete?, budget exceeded?)
  • Parallel Nodes: Execute multiple nodes concurrently (parallel research loops)

Edge Types:

  • Sequential Edges: Always traversed (no condition)
  • Conditional Edges: Traversed based on condition (e.g., if research complete → writer, else → tool selector)
  • Parallel Edges: Used for parallel execution branches

Graph Patterns:

  • Iterative Graph: [Input] → [Thinking] → [Knowledge Gap] → [Decision: Complete?] → [Tool Selector] or [Writer]
  • Deep Research Graph: [Input] → [Planner] → [Parallel Iterative Loops] → [Synthesizer]

Execution Flow:

  1. Graph construction from nodes and edges
  2. Graph validation (no cycles, all nodes reachable)
  3. Graph execution from entry node
  4. Node execution based on type
  5. Edge evaluation for next node(s)
  6. Parallel execution via asyncio.gather()
  7. State updates at state nodes
  8. Event streaming for UI

Key Components

  • Orchestrators: Multiple orchestration patterns (src/orchestrator/, src/orchestrator_*.py)
  • Research Flows: Iterative and deep research patterns (src/orchestrator/research_flow.py)
  • Graph Builder: Graph construction utilities (src/agent_factory/graph_builder.py)
  • Agents: Pydantic AI agents (src/agents/, src/agent_factory/agents.py)
  • Search Tools: PubMed, ClinicalTrials.gov, Europe PMC, RAG (src/tools/)
  • Judge Handler: LLM-based evidence assessment (src/agent_factory/judges.py)
  • Embeddings: Semantic search & deduplication (src/services/embeddings.py)
  • Statistical Analyzer: Modal sandbox execution (src/services/statistical_analyzer.py)
  • Middleware: State management, budget tracking, workflow coordination (src/middleware/)
  • MCP Tools: Claude Desktop integration (src/mcp_tools.py)
  • Gradio UI: Web interface with MCP server and streaming (src/app.py)

Research Team & Parallel Execution

The system supports complex research workflows through:

  1. WorkflowManager: Coordinates multiple parallel research loops
  2. Creates and tracks ResearchLoop instances
  3. Runs loops in parallel via asyncio.gather()
  4. Synchronizes evidence to global state

  5. Handles loop failures gracefully

  6. Deep Research Pattern: Breaks complex queries into sections

  7. Planner creates report outline with sections
  8. Each section runs as independent iterative research loop
  9. Loops execute in parallel
  10. Evidence shared across loops via global state

  11. Final synthesis combines all section results

  12. State Synchronization: Thread-safe evidence sharing

  13. Evidence deduplication by URL
  14. Global state accessible to all loops
  15. Semantic search across all collected evidence
  16. Conversation history tracking per iteration

Configuration & Modes

  • Orchestrator Factory (src/orchestrator_factory.py):
  • Auto-detects mode: "advanced" if OpenAI key available, else "simple"
  • Supports explicit mode selection: "simple", "magentic", "advanced"

  • Lazy imports for optional dependencies

  • Research Modes:

  • iterative: Single research loop
  • deep: Multi-section parallel research

  • auto: Auto-detect based on query complexity

  • Execution Modes:

  • use_graph=True: Graph-based execution (parallel, conditional routing)
  • use_graph=False: Agent chains (sequential, backward compatible)
\ No newline at end of file + Architecture Overview - The DETERMINATOR
Skip to content

Architecture Overview

The DETERMINATOR is a powerful generalist deep research agent system that uses iterative search-and-judge loops to comprehensively investigate any research question. It stops at nothing until finding precise answers, only stopping at configured limits (budget, time, iterations). The system automatically determines if medical knowledge sources are needed and adapts its search strategy accordingly. It supports multiple orchestration patterns, graph-based execution, parallel research workflows, and long-running task management with real-time streaming.

Core Architecture

Orchestration Patterns

  1. Graph Orchestrator (src/orchestrator/graph_orchestrator.py):
  2. Graph-based execution using Pydantic AI agents as nodes
  3. Supports both iterative and deep research patterns
  4. Node types: Agent, State, Decision, Parallel
  5. Edge types: Sequential, Conditional, Parallel
  6. Conditional routing based on knowledge gaps, budget, and iterations
  7. Parallel execution for concurrent research loops
  8. Event streaming via AsyncGenerator[AgentEvent] for real-time UI updates

  9. Fallback to agent chains when graph execution is disabled

  10. Deep Research Flow (src/orchestrator/research_flow.py):

  11. Pattern: Planner → Parallel Iterative Loops (one per section) → Synthesis
  12. Uses PlannerAgent to break query into report sections
  13. Runs IterativeResearchFlow instances in parallel per section via WorkflowManager
  14. Synthesizes results using LongWriterAgent or ProofreaderAgent
  15. Supports both graph execution (use_graph=True) and agent chains (use_graph=False)
  16. Budget tracking per section and globally

  17. State synchronization across parallel loops

  18. Iterative Research Flow (src/orchestrator/research_flow.py):

  19. Pattern: Generate observations → Evaluate gaps → Select tools → Execute → Judge → Continue/Complete
  20. Uses KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent, WriterAgent
  21. JudgeHandler assesses evidence sufficiency
  22. Iterates until research complete or constraints met (iterations, time, tokens)

  23. Supports graph execution and agent chains

  24. Magentic Orchestrator (src/orchestrator_magentic.py):

  25. Multi-agent coordination using agent-framework-core
  26. ChatAgent pattern with internal LLMs per agent
  27. Uses MagenticBuilder with participants: searcher, hypothesizer, judge, reporter
  28. Manager orchestrates agents via OpenAIChatClient
  29. Requires OpenAI API key (function calling support)
  30. Event-driven: converts Magentic events to AgentEvent for UI streaming

  31. Supports long-running workflows with max rounds and stall/reset handling

  32. Hierarchical Orchestrator (src/orchestrator_hierarchical.py):

  33. Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge
  34. Adapts Magentic ChatAgent to SubIterationTeam protocol
  35. Event-driven via asyncio.Queue for coordination

  36. Supports sub-iteration patterns for complex research tasks

  37. Legacy Simple Mode (src/legacy_orchestrator.py):

  38. Linear search-judge-synthesize loop
  39. Uses SearchHandlerProtocol and JudgeHandlerProtocol
  40. Generator-based design yielding AgentEvent objects
  41. Backward compatibility for simple use cases

Long-Running Task Support

The system is designed for long-running research tasks with comprehensive state management and streaming:

  1. Event Streaming:
  2. All orchestrators yield AgentEvent objects via AsyncGenerator
  3. Real-time UI updates through Gradio chat interface
  4. Event types: started, searching, search_complete, judging, judge_complete, looping, synthesizing, hypothesizing, complete, error

  5. Metadata includes iteration numbers, tool names, result counts, durations

  6. Budget Tracking (src/middleware/budget_tracker.py):

  7. Per-loop and global budget management
  8. Tracks: tokens, time (seconds), iterations
  9. Budget enforcement at decision nodes
  10. Token estimation (~4 chars per token)
  11. Early termination when budgets exceeded

  12. Budget summaries for monitoring

  13. Workflow Manager (src/middleware/workflow_manager.py):

  14. Coordinates parallel research loops
  15. Tracks loop status: pending, running, completed, failed, cancelled
  16. Synchronizes evidence between loops and global state
  17. Handles errors per loop (doesn't fail all if one fails)
  18. Supports loop cancellation and timeout handling

  19. Evidence deduplication across parallel loops

  20. State Management (src/middleware/state_machine.py):

  21. Thread-safe isolation using ContextVar for concurrent requests
  22. WorkflowState tracks: evidence, conversation history, embedding service
  23. Evidence deduplication by URL
  24. Semantic search via embedding service
  25. State persistence across long-running workflows

  26. Supports both iterative and deep research patterns

  27. Gradio UI (src/app.py):

  28. Real-time streaming of research progress
  29. Accordion-based UI for pending/done operations
  30. OAuth integration (HuggingFace)
  31. Multiple backend support (API keys, free tier)
  32. Handles long-running tasks with progress indicators
  33. Event accumulation for pending operations

Graph Architecture

The graph orchestrator (src/orchestrator/graph_orchestrator.py) implements a flexible graph-based execution model:

Node Types:

  • Agent Nodes: Execute Pydantic AI agents (e.g., KnowledgeGapAgent, ToolSelectorAgent)
  • State Nodes: Update or read workflow state (evidence, conversation)
  • Decision Nodes: Make routing decisions (research complete?, budget exceeded?)
  • Parallel Nodes: Execute multiple nodes concurrently (parallel research loops)

Edge Types:

  • Sequential Edges: Always traversed (no condition)
  • Conditional Edges: Traversed based on condition (e.g., if research complete → writer, else → tool selector)
  • Parallel Edges: Used for parallel execution branches

Graph Patterns:

  • Iterative Graph: [Input] → [Thinking] → [Knowledge Gap] → [Decision: Complete?] → [Tool Selector] or [Writer]
  • Deep Research Graph: [Input] → [Planner] → [Parallel Iterative Loops] → [Synthesizer]

Execution Flow:

  1. Graph construction from nodes and edges
  2. Graph validation (no cycles, all nodes reachable)
  3. Graph execution from entry node
  4. Node execution based on type
  5. Edge evaluation for next node(s)
  6. Parallel execution via asyncio.gather()
  7. State updates at state nodes
  8. Event streaming for UI

Key Components

  • Orchestrators: Multiple orchestration patterns (src/orchestrator/, src/orchestrator_*.py)
  • Research Flows: Iterative and deep research patterns (src/orchestrator/research_flow.py)
  • Graph Builder: Graph construction utilities (src/agent_factory/graph_builder.py)
  • Agents: Pydantic AI agents (src/agents/, src/agent_factory/agents.py)
  • Search Tools: Neo4j knowledge graph, PubMed, ClinicalTrials.gov, Europe PMC, Web search, RAG (src/tools/)
  • Judge Handler: LLM-based evidence assessment (src/agent_factory/judges.py)
  • Embeddings: Semantic search & deduplication (src/services/embeddings.py)
  • Statistical Analyzer: Modal sandbox execution (src/services/statistical_analyzer.py)
  • Multimodal Processing: Image OCR and audio STT/TTS services (src/services/multimodal_processing.py, src/services/audio_processing.py)
  • Middleware: State management, budget tracking, workflow coordination (src/middleware/)
  • MCP Tools: Claude Desktop integration (src/mcp_tools.py)
  • Gradio UI: Web interface with MCP server and streaming (src/app.py)

Research Team & Parallel Execution

The system supports complex research workflows through:

  1. WorkflowManager: Coordinates multiple parallel research loops
  2. Creates and tracks ResearchLoop instances
  3. Runs loops in parallel via asyncio.gather()
  4. Synchronizes evidence to global state

  5. Handles loop failures gracefully

  6. Deep Research Pattern: Breaks complex queries into sections

  7. Planner creates report outline with sections
  8. Each section runs as independent iterative research loop
  9. Loops execute in parallel
  10. Evidence shared across loops via global state

  11. Final synthesis combines all section results

  12. State Synchronization: Thread-safe evidence sharing

  13. Evidence deduplication by URL
  14. Global state accessible to all loops
  15. Semantic search across all collected evidence
  16. Conversation history tracking per iteration

Configuration & Modes

  • Orchestrator Factory (src/orchestrator_factory.py):
  • Auto-detects mode: "advanced" if OpenAI key available, else "simple"
  • Supports explicit mode selection: "simple", "magentic" (alias for "advanced"), "advanced", "iterative", "deep", "auto"

  • Lazy imports for optional dependencies

  • Orchestrator Modes (selected in UI or via factory):

  • simple: Legacy linear search-judge loop (Free Tier)
  • advanced or magentic: Multi-agent coordination using Microsoft Agent Framework (requires OpenAI API key)
  • iterative: Knowledge-gap-driven research with single loop (Free Tier)
  • deep: Parallel section-based research with planning (Free Tier)

  • auto: Intelligent mode detection based on query complexity (Free Tier)

  • Graph Research Modes (used within graph orchestrator, separate from orchestrator mode):

  • iterative: Single research loop pattern
  • deep: Multi-section parallel research pattern

  • auto: Auto-detect pattern based on query complexity

  • Execution Modes:

  • use_graph=True: Graph-based execution (parallel, conditional routing)
  • use_graph=False: Agent chains (sequential, backward compatible)

Note: The UI provides separate controls for orchestrator mode and graph research mode. When using graph-based orchestrators (iterative/deep/auto), the graph research mode determines the specific pattern used within the graph execution.

\ No newline at end of file diff --git a/site/overview/features/index.html b/site/overview/features/index.html index 26bc6ad..de754ff 100644 --- a/site/overview/features/index.html +++ b/site/overview/features/index.html @@ -1 +1 @@ - Features - The DETERMINATOR
Skip to content

Features

DeepCritical provides a comprehensive set of features for AI-assisted research:

Core Features

  • PubMed: Search peer-reviewed biomedical literature via NCBI E-utilities
  • ClinicalTrials.gov: Search interventional clinical trials
  • Europe PMC: Search preprints and peer-reviewed articles (includes bioRxiv/medRxiv)
  • RAG: Semantic search within collected evidence using LlamaIndex

MCP Integration

  • Model Context Protocol: Expose search tools via MCP server
  • Claude Desktop: Use DeepCritical tools directly from Claude Desktop
  • MCP Clients: Compatible with any MCP-compatible client

Authentication

  • HuggingFace OAuth: Sign in with HuggingFace account for automatic API token usage
  • Manual API Keys: Support for OpenAI, Anthropic, and HuggingFace API keys
  • Free Tier Support: Automatic fallback to HuggingFace Inference API

Secure Code Execution

  • Modal Sandbox: Secure execution of AI-generated statistical code
  • Isolated Environment: Network isolation and package version pinning
  • Safe Execution: Prevents malicious code execution

Semantic Search & RAG

  • LlamaIndex Integration: Advanced RAG capabilities
  • Vector Storage: ChromaDB for embedding storage
  • Semantic Deduplication: Automatic detection of similar evidence
  • Embedding Service: Local sentence-transformers (no API key required)

Orchestration Patterns

  • Graph-Based Execution: Flexible graph orchestration with conditional routing
  • Parallel Research Loops: Run multiple research tasks concurrently
  • Iterative Research: Single-loop research with search-judge-synthesize cycles
  • Deep Research: Multi-section parallel research with planning and synthesis
  • Magentic Orchestration: Multi-agent coordination using Microsoft Agent Framework

Real-Time Streaming

  • Event Streaming: Real-time updates via AsyncGenerator[AgentEvent]
  • Progress Tracking: Monitor research progress with detailed event metadata
  • UI Integration: Seamless integration with Gradio chat interface

Budget Management

  • Token Budget: Track and limit LLM token usage
  • Time Budget: Enforce time limits per research loop
  • Iteration Budget: Limit maximum iterations
  • Per-Loop Budgets: Independent budgets for parallel research loops

State Management

  • Thread-Safe Isolation: ContextVar-based state management
  • Evidence Deduplication: Automatic URL-based deduplication
  • Conversation History: Track iteration history and agent interactions
  • State Synchronization: Share evidence across parallel loops

Advanced Features

Agent System

  • Pydantic AI Agents: Type-safe agent implementation
  • Structured Output: Pydantic models for agent responses
  • Agent Factory: Centralized agent creation with fallback support
  • Specialized Agents: Knowledge gap, tool selector, writer, proofreader, and more

Search Tools

  • Rate Limiting: Built-in rate limiting for external APIs
  • Retry Logic: Automatic retry with exponential backoff
  • Query Preprocessing: Automatic query enhancement and synonym expansion
  • Evidence Conversion: Automatic conversion to structured Evidence objects

Error Handling

  • Custom Exceptions: Hierarchical exception system
  • Error Chaining: Preserve exception context
  • Structured Logging: Comprehensive logging with structlog
  • Graceful Degradation: Fallback handlers for missing dependencies

Configuration

  • Pydantic Settings: Type-safe configuration management
  • Environment Variables: Support for .env files
  • Validation: Automatic configuration validation
  • Flexible Providers: Support for multiple LLM and embedding providers

Testing

  • Unit Tests: Comprehensive unit test coverage
  • Integration Tests: Real API integration tests
  • Mock Support: Extensive mocking utilities
  • Coverage Reports: Code coverage tracking

UI Features

Gradio Interface

  • Real-Time Chat: Interactive chat interface
  • Streaming Updates: Live progress updates
  • Accordion UI: Organized display of pending/done operations
  • OAuth Integration: Seamless HuggingFace authentication

MCP Server

  • RESTful API: HTTP-based MCP server
  • Tool Discovery: Automatic tool registration
  • Request Handling: Async request processing
  • Error Responses: Structured error responses

Development Features

Code Quality

  • Type Safety: Full type hints with mypy strict mode
  • Linting: Ruff for code quality
  • Formatting: Automatic code formatting
  • Pre-commit Hooks: Automated quality checks

Documentation

  • Comprehensive Docs: Detailed documentation for all components
  • Code Examples: Extensive code examples
  • Architecture Diagrams: Visual architecture documentation
  • API Reference: Complete API documentation
\ No newline at end of file + Features - The DETERMINATOR
Skip to content

Features

The DETERMINATOR provides a comprehensive set of features for AI-assisted research:

Core Features

  • General Web Search: Search general knowledge sources for any domain
  • Neo4j Knowledge Graph: Search structured knowledge graph for papers and disease relationships
  • PubMed: Search peer-reviewed biomedical literature via NCBI E-utilities (automatically used when medical knowledge needed)
  • ClinicalTrials.gov: Search interventional clinical trials (automatically used when medical knowledge needed)
  • Europe PMC: Search preprints and peer-reviewed articles (includes bioRxiv/medRxiv)
  • RAG: Semantic search within collected evidence using LlamaIndex
  • Automatic Source Selection: Automatically determines which sources are needed based on query analysis

MCP Integration

  • Model Context Protocol: Expose search tools via MCP server
  • Claude Desktop: Use The DETERMINATOR tools directly from Claude Desktop
  • MCP Clients: Compatible with any MCP-compatible client

Authentication

  • REQUIRED: Authentication is mandatory before using the application
  • HuggingFace OAuth: Sign in with HuggingFace account for automatic API token usage (recommended)
  • Manual API Keys: Support for HuggingFace API keys via environment variables (HF_TOKEN or HUGGINGFACE_API_KEY)
  • Free Tier Support: Automatic fallback to HuggingFace Inference API (public models) when no API key is available
  • Authentication Check: The application will display an error message if authentication is not provided

Secure Code Execution

  • Modal Sandbox: Secure execution of AI-generated statistical code
  • Isolated Environment: Network isolation and package version pinning
  • Safe Execution: Prevents malicious code execution

Semantic Search & RAG

  • LlamaIndex Integration: Advanced RAG capabilities
  • Vector Storage: ChromaDB for embedding storage
  • Semantic Deduplication: Automatic detection of similar evidence
  • Embedding Service: Local sentence-transformers (no API key required)

Orchestration Patterns

  • Graph-Based Execution: Flexible graph orchestration with conditional routing
  • Parallel Research Loops: Run multiple research tasks concurrently
  • Iterative Research: Single-loop research with search-judge-synthesize cycles that continues until precise answers are found
  • Deep Research: Multi-section parallel research with planning and synthesis
  • Magentic Orchestration: Multi-agent coordination using Microsoft Agent Framework (alias: "advanced" mode)
  • Stops at Nothing: Only stops at configured limits (budget, time, iterations), otherwise continues until finding precise answers

Orchestrator Modes: - simple: Legacy linear search-judge loop - advanced (or magentic): Multi-agent coordination (requires OpenAI API key) - iterative: Knowledge-gap-driven research with single loop - deep: Parallel section-based research with planning - auto: Intelligent mode detection based on query complexity

Graph Research Modes (used within graph orchestrator): - iterative: Single research loop pattern - deep: Multi-section parallel research pattern - auto: Auto-detect pattern based on query complexity

Execution Modes: - use_graph=True: Graph-based execution with parallel and conditional routing - use_graph=False: Agent chains with sequential execution (backward compatible)

Real-Time Streaming

  • Event Streaming: Real-time updates via AsyncGenerator[AgentEvent]
  • Progress Tracking: Monitor research progress with detailed event metadata
  • UI Integration: Seamless integration with Gradio chat interface

Budget Management

  • Token Budget: Track and limit LLM token usage
  • Time Budget: Enforce time limits per research loop
  • Iteration Budget: Limit maximum iterations
  • Per-Loop Budgets: Independent budgets for parallel research loops

State Management

  • Thread-Safe Isolation: ContextVar-based state management
  • Evidence Deduplication: Automatic URL-based deduplication
  • Conversation History: Track iteration history and agent interactions
  • State Synchronization: Share evidence across parallel loops

Multimodal Input & Output

  • Image Input (OCR): Upload images and extract text using optical character recognition
  • Audio Input (STT): Record or upload audio files and transcribe to text using speech-to-text
  • Audio Output (TTS): Generate audio responses with text-to-speech synthesis
  • Configurable Settings: Enable/disable multimodal features via sidebar settings
  • Voice Selection: Choose from multiple TTS voices (American English: af_, am_)
  • Speech Speed Control: Adjust TTS speech speed (0.5x to 2.0x)
  • Multimodal Processing Service: Integrated service for processing images and audio files

Advanced Features

Agent System

  • Pydantic AI Agents: Type-safe agent implementation
  • Structured Output: Pydantic models for agent responses
  • Agent Factory: Centralized agent creation with fallback support
  • Specialized Agents: Knowledge gap, tool selector, writer, proofreader, and more

Search Tools

  • Rate Limiting: Built-in rate limiting for external APIs
  • Retry Logic: Automatic retry with exponential backoff
  • Query Preprocessing: Automatic query enhancement and synonym expansion
  • Evidence Conversion: Automatic conversion to structured Evidence objects

Error Handling

  • Custom Exceptions: Hierarchical exception system
  • Error Chaining: Preserve exception context
  • Structured Logging: Comprehensive logging with structlog
  • Graceful Degradation: Fallback handlers for missing dependencies

Configuration

  • Pydantic Settings: Type-safe configuration management
  • Environment Variables: Support for .env files
  • Validation: Automatic configuration validation
  • Flexible Providers: Support for multiple LLM and embedding providers

Testing

  • Unit Tests: Comprehensive unit test coverage
  • Integration Tests: Real API integration tests
  • Mock Support: Extensive mocking utilities
  • Coverage Reports: Code coverage tracking

UI Features

Gradio Interface

  • Real-Time Chat: Interactive chat interface with multimodal support
  • Streaming Updates: Live progress updates
  • Accordion UI: Organized display of pending/done operations
  • OAuth Integration: Seamless HuggingFace authentication
  • Multimodal Input: Support for text, images, and audio input in the same interface
  • Sidebar Settings: Configuration accordions for research, multimodal, and audio settings

MCP Server

  • RESTful API: HTTP-based MCP server
  • Tool Discovery: Automatic tool registration
  • Request Handling: Async request processing
  • Error Responses: Structured error responses

Development Features

Code Quality

  • Type Safety: Full type hints with mypy strict mode
  • Linting: Ruff for code quality
  • Formatting: Automatic code formatting
  • Pre-commit Hooks: Automated quality checks

Documentation

  • Comprehensive Docs: Detailed documentation for all components
  • Code Examples: Extensive code examples
  • Architecture Diagrams: Visual architecture documentation
  • API Reference: Complete API documentation
\ No newline at end of file diff --git a/site/overview/quick-start/index.html b/site/overview/quick-start/index.html index b8629df..4a1744a 100644 --- a/site/overview/quick-start/index.html +++ b/site/overview/quick-start/index.html @@ -1,15 +1,22 @@ - Quick Start - The DETERMINATOR
Skip to content

Quick Start

Get started with DeepCritical in minutes.

Installation

# Install uv if you haven't already
-pip install uv
-
-# Sync dependencies
-uv sync
-

Run the UI

# Start the Gradio app
-uv run gradio run src/app.py
-

Open your browser to http://localhost:7860.

Basic Usage

1. Authentication (Optional)

HuggingFace OAuth Login: - Click the "Sign in with HuggingFace" button at the top of the app - Your HuggingFace API token will be automatically used for AI inference - No need to manually enter API keys when logged in

Manual API Key (BYOK): - Provide your own API key in the Settings accordion - Supports HuggingFace, OpenAI, or Anthropic API keys - Manual keys take priority over OAuth tokens

2. Start a Research Query

  1. Enter your research question in the chat interface
  2. Click "Submit" or press Enter
  3. Watch the real-time progress as the system:
  4. Generates observations
  5. Identifies knowledge gaps
  6. Searches multiple sources
  7. Evaluates evidence
  8. Synthesizes findings
  9. Review the final research report

3. MCP Integration (Optional)

Connect DeepCritical to Claude Desktop:

  1. Add to your claude_desktop_config.json: 

    {
-  "mcpServers": {
-    "deepcritical": {
-      "url": "http://localhost:7860/gradio_api/mcp/"
-    }
-  }
-}
-

  2. Restart Claude Desktop

  3. Use DeepCritical tools directly from Claude Desktop

Available Tools

  • search_pubmed: Search peer-reviewed biomedical literature
  • search_clinical_trials: Search ClinicalTrials.gov
  • search_biorxiv: Search bioRxiv/medRxiv preprints
  • search_all: Search all sources simultaneously
  • analyze_hypothesis: Secure statistical analysis using Modal sandboxes

Next Steps

\ No newline at end of file + Quick Start - The DETERMINATOR
Skip to content

Quick Start

Get started with DeepCritical in minutes.

Installation

# Install uv if you haven't already (recommended: standalone installer)
+# Unix/macOS/Linux:
+curl -LsSf https://astral.sh/uv/install.sh | sh
+
+# Windows (PowerShell):
+powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
+
+# Alternative: pipx install uv
+# Or: pip install uv
+
+# Sync dependencies
+uv sync
+

Run the UI

# Start the Gradio app
+uv run gradio run src/app.py
+

Open your browser to http://localhost:7860.

Basic Usage

1. Authentication (REQUIRED)

Authentication is mandatory - you must authenticate before using the application. The app will display an error message if you try to use it without authentication.

HuggingFace OAuth Login (Recommended): - Click the "Sign in with HuggingFace" button at the top of the app - Your HuggingFace API token will be automatically used for AI inference - No need to manually enter API keys when logged in

Manual API Key (Alternative): - Set environment variable HF_TOKEN or HUGGINGFACE_API_KEY before starting the app - The app will automatically use these tokens if OAuth login is not available - Supports HuggingFace API keys only (OpenAI/Anthropic keys are not used in the current implementation)

2. Start a Research Query

  1. Enter your research question in the chat interface
  2. Text Input: Type your question directly
  3. Image Input: Click the 📷 icon to upload images (OCR will extract text)
  4. Audio Input: Click the 🎤 icon to record or upload audio (STT will transcribe to text)
  5. Click "Submit" or press Enter
  6. Watch the real-time progress as the system:
  7. Generates observations
  8. Identifies knowledge gaps
  9. Searches multiple sources
  10. Evaluates evidence
  11. Synthesizes findings
  12. Review the final research report
  13. Audio Output: If enabled, the final response will include audio synthesis (TTS)

Multimodal Features: - Configure image/audio input and output in the sidebar settings - Image OCR and audio STT/TTS can be enabled/disabled independently - TTS voice and speed can be customized in the Audio Output settings

3. MCP Integration (Optional)

Connect DeepCritical to Claude Desktop:

  1. Add to your claude_desktop_config.json: 

    {
+  "mcpServers": {
+    "deepcritical": {
+      "url": "http://localhost:7860/gradio_api/mcp/"
+    }
+  }
+}
+

  2. Restart Claude Desktop

  3. Use DeepCritical tools directly from Claude Desktop

Available Tools

  • search_pubmed: Search peer-reviewed biomedical literature
  • search_clinical_trials: Search ClinicalTrials.gov
  • search_biorxiv: Search bioRxiv/medRxiv preprints
  • search_neo4j: Search Neo4j knowledge graph for papers and disease relationships
  • search_all: Search all sources simultaneously
  • analyze_hypothesis: Secure statistical analysis using Modal sandboxes

Note: The application automatically uses all available search tools (Neo4j, PubMed, ClinicalTrials.gov, Europe PMC, Web search, RAG) based on query analysis. Neo4j knowledge graph search is included by default for biomedical queries.

Next Steps

\ No newline at end of file diff --git a/site/search/search_index.json b/site/search/search_index.json index 992482b..306ac34 100644 --- a/site/search/search_index.json +++ b/site/search/search_index.json @@ -1 +1 @@ -{"config":{"lang":["en"],"separator":"[\\s\\-]+","pipeline":["stopWordFilter"],"fields":{"title":{"boost":1000.0},"text":{"boost":1.0},"tags":{"boost":1000000.0}}},"docs":[{"location":"","title":"DeepCritical","text":"

AI-Native Drug Repurposing Research Agent

DeepCritical is a deep research agent system that uses iterative search-and-judge loops to comprehensively answer research questions. The system supports multiple orchestration patterns, graph-based execution, parallel research workflows, and long-running task management with real-time streaming.

"},{"location":"#features","title":"Features","text":""},{"location":"#quick-start","title":"Quick Start","text":"
# Install uv if you haven't already\npip install uv\n\n# Sync dependencies\nuv sync\n\n# Start the Gradio app\nuv run gradio run src/app.py\n

Open your browser to http://localhost:7860.

For detailed installation and setup instructions, see the Getting Started Guide.

"},{"location":"#architecture","title":"Architecture","text":"

DeepCritical uses a Vertical Slice Architecture:

  1. Search Slice: Retrieving evidence from PubMed, ClinicalTrials.gov, and Europe PMC
  2. Judge Slice: Evaluating evidence quality using LLMs
  3. Orchestrator Slice: Managing the research loop and UI

The system supports three main research patterns:

Learn more about the Architecture.

"},{"location":"#documentation","title":"Documentation","text":""},{"location":"#links","title":"Links","text":""},{"location":"CONTRIBUTING/","title":"Contributing to DeepCritical","text":"

Thank you for your interest in contributing to DeepCritical! This guide will help you get started.

"},{"location":"CONTRIBUTING/#table-of-contents","title":"Table of Contents","text":"

Note: Additional sections (Code Style, Error Handling, Testing, Implementation Patterns, Code Quality, and Prompt Engineering) are available as separate pages in the navigation sidebar.

"},{"location":"CONTRIBUTING/#git-workflow","title":"Git Workflow","text":""},{"location":"CONTRIBUTING/#getting-started","title":"Getting Started","text":"
  1. Fork the repository on GitHub
  2. Clone your fork:
git clone https://github.com/yourusername/GradioDemo.git\ncd GradioDemo\n
  1. Install dependencies:
make install\n
  1. Create a feature branch:
git checkout -b yourname-feature-name\n
  1. Make your changes following the guidelines below
  2. Run checks:
make check\n
  1. Commit and push:

git commit -m \"Description of changes\"\ngit push origin yourname-feature-name\n
8. Create a pull request on GitHub

"},{"location":"CONTRIBUTING/#development-commands","title":"Development Commands","text":"
make install      # Install dependencies + pre-commit\nmake check        # Lint + typecheck + test (MUST PASS)\nmake test         # Run unit tests\nmake lint         # Run ruff\nmake format       # Format with ruff\nmake typecheck    # Run mypy\nmake test-cov     # Test with coverage\nmake docs-build  # Build documentation\nmake docs-serve  # Serve documentation locally\n
"},{"location":"CONTRIBUTING/#code-style-conventions","title":"Code Style & Conventions","text":""},{"location":"CONTRIBUTING/#type-safety","title":"Type Safety","text":""},{"location":"CONTRIBUTING/#pydantic-models","title":"Pydantic Models","text":""},{"location":"CONTRIBUTING/#async-patterns","title":"Async Patterns","text":" 
loop = asyncio.get_running_loop()\nresult = await loop.run_in_executor(None, cpu_bound_function, args)\n
"},{"location":"CONTRIBUTING/#linting","title":"Linting","text":""},{"location":"CONTRIBUTING/#pre-commit","title":"Pre-commit","text":""},{"location":"CONTRIBUTING/#error-handling-logging","title":"Error Handling & Logging","text":""},{"location":"CONTRIBUTING/#exception-hierarchy","title":"Exception Hierarchy","text":"

Use custom exception hierarchy (src/utils/exceptions.py):

"},{"location":"CONTRIBUTING/#error-handling-rules","title":"Error Handling Rules","text":" 
logger.error(\"Operation failed\", error=str(e), context=value)\n
"},{"location":"CONTRIBUTING/#logging","title":"Logging","text":""},{"location":"CONTRIBUTING/#logging-examples","title":"Logging Examples","text":"
logger.info(\"Starting search\", query=query, tools=[t.name for t in tools])\nlogger.warning(\"Search tool failed\", tool=tool.name, error=str(result))\nlogger.error(\"Assessment failed\", error=str(e))\n
"},{"location":"CONTRIBUTING/#error-chaining","title":"Error Chaining","text":"

Always preserve exception context:

try:\n    result = await api_call()\nexcept httpx.HTTPError as e:\n    raise SearchError(f\"API call failed: {e}\") from e\n
"},{"location":"CONTRIBUTING/#testing-requirements","title":"Testing Requirements","text":""},{"location":"CONTRIBUTING/#test-structure","title":"Test Structure","text":""},{"location":"CONTRIBUTING/#mocking","title":"Mocking","text":""},{"location":"CONTRIBUTING/#tdd-workflow","title":"TDD Workflow","text":"
  1. Write failing test in tests/unit/
  2. Implement in src/
  3. Ensure test passes
  4. Run make check (lint + typecheck + test)
"},{"location":"CONTRIBUTING/#test-examples","title":"Test Examples","text":"
@pytest.mark.unit\nasync def test_pubmed_search(mock_httpx_client):\n    tool = PubMedTool()\n    results = await tool.search(\"metformin\", max_results=5)\n    assert len(results) > 0\n    assert all(isinstance(r, Evidence) for r in results)\n\n@pytest.mark.integration\nasync def test_real_pubmed_search():\n    tool = PubMedTool()\n    results = await tool.search(\"metformin\", max_results=3)\n    assert len(results) <= 3\n
"},{"location":"CONTRIBUTING/#test-coverage","title":"Test Coverage","text":""},{"location":"CONTRIBUTING/#implementation-patterns","title":"Implementation Patterns","text":""},{"location":"CONTRIBUTING/#search-tools","title":"Search Tools","text":"

All tools implement SearchTool protocol (src/tools/base.py):

Example pattern:

class MySearchTool:\n    @property\n    def name(self) -> str:\n        return \"mytool\"\n    \n    @retry(stop=stop_after_attempt(3), wait=wait_exponential(...))\n    async def search(self, query: str, max_results: int = 10) -> list[Evidence]:\n        # Implementation\n        return evidence_list\n
"},{"location":"CONTRIBUTING/#judge-handlers","title":"Judge Handlers","text":""},{"location":"CONTRIBUTING/#agent-factory-pattern","title":"Agent Factory Pattern","text":""},{"location":"CONTRIBUTING/#state-management","title":"State Management","text":""},{"location":"CONTRIBUTING/#singleton-pattern","title":"Singleton Pattern","text":"

Use @lru_cache(maxsize=1) for singletons:

"},{"location":"CONTRIBUTING/#code-quality-documentation","title":"Code Quality & Documentation","text":""},{"location":"CONTRIBUTING/#docstrings","title":"Docstrings","text":"

Example:

"},{"location":"CONTRIBUTING/#code-comments","title":"Code Comments","text":""},{"location":"CONTRIBUTING/#prompt-engineering-citation-validation","title":"Prompt Engineering & Citation Validation","text":""},{"location":"CONTRIBUTING/#judge-prompts","title":"Judge Prompts","text":""},{"location":"CONTRIBUTING/#hypothesis-prompts","title":"Hypothesis Prompts","text":""},{"location":"CONTRIBUTING/#report-prompts","title":"Report Prompts","text":""},{"location":"CONTRIBUTING/#citation-validation","title":"Citation Validation","text":""},{"location":"CONTRIBUTING/#citation-validation-rules","title":"Citation Validation Rules","text":"
  1. Every reference URL must EXACTLY match a provided evidence URL
  2. Do NOT invent, fabricate, or hallucinate any references
  3. Do NOT modify paper titles, authors, dates, or URLs
  4. If unsure about a citation, OMIT it rather than guess
  5. Copy URLs exactly as provided - do not create similar-looking URLs
"},{"location":"CONTRIBUTING/#evidence-selection","title":"Evidence Selection","text":""},{"location":"CONTRIBUTING/#mcp-integration","title":"MCP Integration","text":""},{"location":"CONTRIBUTING/#mcp-tools","title":"MCP Tools","text":""},{"location":"CONTRIBUTING/#gradio-mcp-server","title":"Gradio MCP Server","text":""},{"location":"CONTRIBUTING/#common-pitfalls","title":"Common Pitfalls","text":"
  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Hallucinated citations: Always validate references
  4. Global mutable state: Use ContextVar or pass via parameters
  5. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
  6. Rate limiting: Always implement for external APIs
  7. Error chaining: Always use from e when raising exceptions
"},{"location":"CONTRIBUTING/#key-principles","title":"Key Principles","text":"
  1. Type Safety First: All code must pass mypy --strict
  2. Async Everything: All I/O must be async
  3. Test-Driven: Write tests before implementation
  4. No Hallucinations: Validate all citations
  5. Graceful Degradation: Support free tier (HF Inference) when no API keys
  6. Lazy Loading: Don't require optional dependencies at import time
  7. Structured Logging: Use structlog, never print()
  8. Error Chaining: Always preserve exception context
"},{"location":"CONTRIBUTING/#pull-request-process","title":"Pull Request Process","text":"
  1. Ensure all checks pass: make check
  2. Update documentation if needed
  3. Add tests for new features
  4. Update CHANGELOG if applicable
  5. Request review from maintainers
  6. Address review feedback
  7. Wait for approval before merging
"},{"location":"CONTRIBUTING/#questions","title":"Questions?","text":"

Thank you for contributing to DeepCritical!

"},{"location":"LICENSE/","title":"License","text":"

DeepCritical is licensed under the MIT License.

"},{"location":"LICENSE/#mit-license","title":"MIT License","text":"

Copyright (c) 2024 DeepCritical Team

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

"},{"location":"team/","title":"Team","text":"

DeepCritical is developed by a team of researchers and developers working on AI-assisted research.

"},{"location":"team/#team-members","title":"Team Members","text":""},{"location":"team/#zj","title":"ZJ","text":""},{"location":"team/#mario-aderman","title":"Mario Aderman","text":""},{"location":"team/#joseph-pollack","title":"Joseph Pollack","text":""},{"location":"team/#virat-chauran","title":"Virat Chauran","text":""},{"location":"team/#anna-bossler","title":"Anna Bossler","text":""},{"location":"team/#about","title":"About","text":"

The DeepCritical team met online in the Alzheimer's Critical Literature Review Group in the Hugging Science initiative. We're building the agent framework we want to use for AI-assisted research to turn the vast amounts of clinical data into cures.

"},{"location":"team/#contributing","title":"Contributing","text":"

We welcome contributions! See the Contributing Guide for details.

"},{"location":"team/#links","title":"Links","text":""},{"location":"api/agents/","title":"Agents API Reference","text":"

This page documents the API for DeepCritical agents.

"},{"location":"api/agents/#knowledgegapagent","title":"KnowledgeGapAgent","text":"

Module: src.agents.knowledge_gap

Purpose: Evaluates research state and identifies knowledge gaps.

"},{"location":"api/agents/#methods","title":"Methods","text":""},{"location":"api/agents/#evaluate","title":"evaluate","text":"
async def evaluate(\n    self,\n    query: str,\n    background_context: str,\n    conversation_history: Conversation,\n    iteration: int,\n    time_elapsed_minutes: float,\n    max_time_minutes: float\n) -> KnowledgeGapOutput\n

Evaluates research completeness and identifies outstanding knowledge gaps.

Parameters: - query: Research query string - background_context: Background context for the query - conversation_history: Conversation history with previous iterations - iteration: Current iteration number - time_elapsed_minutes: Elapsed time in minutes - max_time_minutes: Maximum time limit in minutes

Returns: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

"},{"location":"api/agents/#toolselectoragent","title":"ToolSelectorAgent","text":"

Module: src.agents.tool_selector

Purpose: Selects appropriate tools for addressing knowledge gaps.

"},{"location":"api/agents/#methods_1","title":"Methods","text":""},{"location":"api/agents/#select_tools","title":"select_tools","text":"
async def select_tools(\n    self,\n    query: str,\n    knowledge_gaps: list[str],\n    available_tools: list[str]\n) -> AgentSelectionPlan\n

Selects tools for addressing knowledge gaps.

Parameters: - query: Research query string - knowledge_gaps: List of knowledge gaps to address - available_tools: List of available tool names

Returns: AgentSelectionPlan with list of AgentTask objects.

"},{"location":"api/agents/#writeragent","title":"WriterAgent","text":"

Module: src.agents.writer

Purpose: Generates final reports from research findings.

"},{"location":"api/agents/#methods_2","title":"Methods","text":""},{"location":"api/agents/#write_report","title":"write_report","text":"
async def write_report(\n    self,\n    query: str,\n    findings: str,\n    output_length: str = \"medium\",\n    output_instructions: str | None = None\n) -> str\n

Generates a markdown report from research findings.

Parameters: - query: Research query string - findings: Research findings to include in report - output_length: Desired output length (\"short\", \"medium\", \"long\") - output_instructions: Additional instructions for report generation

Returns: Markdown string with numbered citations.

"},{"location":"api/agents/#longwriteragent","title":"LongWriterAgent","text":"

Module: src.agents.long_writer

Purpose: Long-form report generation with section-by-section writing.

"},{"location":"api/agents/#methods_3","title":"Methods","text":""},{"location":"api/agents/#write_next_section","title":"write_next_section","text":"
async def write_next_section(\n    self,\n    query: str,\n    draft: ReportDraft,\n    section_title: str,\n    section_content: str\n) -> LongWriterOutput\n

Writes the next section of a long-form report.

Parameters: - query: Research query string - draft: Current report draft - section_title: Title of the section to write - section_content: Content/guidance for the section

Returns: LongWriterOutput with updated draft.

"},{"location":"api/agents/#write_report_1","title":"write_report","text":"
async def write_report(\n    self,\n    query: str,\n    report_title: str,\n    report_draft: ReportDraft\n) -> str\n

Generates final report from draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Complete report draft

Returns: Final markdown report string.

"},{"location":"api/agents/#proofreaderagent","title":"ProofreaderAgent","text":"

Module: src.agents.proofreader

Purpose: Proofreads and polishes report drafts.

"},{"location":"api/agents/#methods_4","title":"Methods","text":""},{"location":"api/agents/#proofread","title":"proofread","text":"
async def proofread(\n    self,\n    query: str,\n    report_title: str,\n    report_draft: ReportDraft\n) -> str\n

Proofreads and polishes a report draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Report draft to proofread

Returns: Polished markdown string.

"},{"location":"api/agents/#thinkingagent","title":"ThinkingAgent","text":"

Module: src.agents.thinking

Purpose: Generates observations from conversation history.

"},{"location":"api/agents/#methods_5","title":"Methods","text":""},{"location":"api/agents/#generate_observations","title":"generate_observations","text":"
async def generate_observations(\n    self,\n    query: str,\n    background_context: str,\n    conversation_history: Conversation\n) -> str\n

Generates observations from conversation history.

Parameters: - query: Research query string - background_context: Background context - conversation_history: Conversation history

Returns: Observation string.

"},{"location":"api/agents/#inputparseragent","title":"InputParserAgent","text":"

Module: src.agents.input_parser

Purpose: Parses and improves user queries, detects research mode.

"},{"location":"api/agents/#methods_6","title":"Methods","text":""},{"location":"api/agents/#parse_query","title":"parse_query","text":"
async def parse_query(\n    self,\n    query: str\n) -> ParsedQuery\n

Parses and improves a user query.

Parameters: - query: Original query string

Returns: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: \"iterative\" or \"deep\" - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"api/agents/#factory-functions","title":"Factory Functions","text":"

All agents have factory functions in src.agent_factory.agents:

Parameters: - model: Optional Pydantic AI model. If None, uses get_model() from settings.

Returns: Agent instance.

"},{"location":"api/agents/#see-also","title":"See Also","text":""},{"location":"api/models/","title":"Models API Reference","text":"

This page documents the Pydantic models used throughout DeepCritical.

"},{"location":"api/models/#evidence","title":"Evidence","text":"

Module: src.utils.models

Purpose: Represents evidence from search results.

Fields: - citation: Citation information (title, URL, date, authors) - content: Evidence text content - relevance_score: Relevance score (0.0-1.0) - metadata: Additional metadata dictionary

"},{"location":"api/models/#citation","title":"Citation","text":"

Module: src.utils.models

Purpose: Citation information for evidence.

Fields: - title: Article/trial title - url: Source URL - date: Publication date (optional) - authors: List of authors (optional)

"},{"location":"api/models/#knowledgegapoutput","title":"KnowledgeGapOutput","text":"

Module: src.utils.models

Purpose: Output from knowledge gap evaluation.

Fields: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

"},{"location":"api/models/#agentselectionplan","title":"AgentSelectionPlan","text":"

Module: src.utils.models

Purpose: Plan for tool/agent selection.

Fields: - tasks: List of agent tasks to execute

"},{"location":"api/models/#agenttask","title":"AgentTask","text":"

Module: src.utils.models

Purpose: Individual agent task.

Fields: - agent_name: Name of agent to use - query: Task query - context: Additional context dictionary

"},{"location":"api/models/#reportdraft","title":"ReportDraft","text":"

Module: src.utils.models

Purpose: Draft structure for long-form reports.

Fields: - title: Report title - sections: List of report sections - references: List of citations

"},{"location":"api/models/#reportsection","title":"ReportSection","text":"

Module: src.utils.models

Purpose: Individual section in a report draft.

Fields: - title: Section title - content: Section content - order: Section order number

"},{"location":"api/models/#parsedquery","title":"ParsedQuery","text":"

Module: src.utils.models

Purpose: Parsed and improved query.

Fields: - original_query: Original query string - improved_query: Refined query string - research_mode: Research mode (\"iterative\" or \"deep\") - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"api/models/#conversation","title":"Conversation","text":"

Module: src.utils.models

Purpose: Conversation history with iterations.

Fields: - iterations: List of iteration data

"},{"location":"api/models/#iterationdata","title":"IterationData","text":"

Module: src.utils.models

Purpose: Data for a single iteration.

Fields: - iteration: Iteration number - observations: Generated observations - knowledge_gaps: Identified knowledge gaps - tool_calls: Tool calls made - findings: Findings from tools - thoughts: Agent thoughts

"},{"location":"api/models/#agentevent","title":"AgentEvent","text":"

Module: src.utils.models

Purpose: Event emitted during research execution.

Fields: - type: Event type (e.g., \"started\", \"search_complete\", \"complete\") - iteration: Iteration number (optional) - data: Event data dictionary

"},{"location":"api/models/#budgetstatus","title":"BudgetStatus","text":"

Module: src.utils.models

Purpose: Current budget status.

Fields: - tokens_used: Tokens used so far - tokens_limit: Token limit - time_elapsed_seconds: Elapsed time in seconds - time_limit_seconds: Time limit in seconds - iterations: Current iteration count - iterations_limit: Iteration limit

"},{"location":"api/models/#see-also","title":"See Also","text":""},{"location":"api/orchestrators/","title":"Orchestrators API Reference","text":"

This page documents the API for DeepCritical orchestrators.

"},{"location":"api/orchestrators/#iterativeresearchflow","title":"IterativeResearchFlow","text":"

Module: src.orchestrator.research_flow

Purpose: Single-loop research with search-judge-synthesize cycles.

"},{"location":"api/orchestrators/#methods","title":"Methods","text":""},{"location":"api/orchestrators/#run","title":"run","text":"
async def run(\n    self,\n    query: str,\n    background_context: str = \"\",\n    max_iterations: int | None = None,\n    max_time_minutes: float | None = None,\n    token_budget: int | None = None\n) -> AsyncGenerator[AgentEvent, None]\n

Runs iterative research flow.

Parameters: - query: Research query string - background_context: Background context (default: \"\") - max_iterations: Maximum iterations (default: from settings) - max_time_minutes: Maximum time in minutes (default: from settings) - token_budget: Token budget (default: from settings)

Yields: AgentEvent objects for: - started: Research started - search_complete: Search completed - judge_complete: Evidence evaluation completed - synthesizing: Generating report - complete: Research completed - error: Error occurred

"},{"location":"api/orchestrators/#deepresearchflow","title":"DeepResearchFlow","text":"

Module: src.orchestrator.research_flow

Purpose: Multi-section parallel research with planning and synthesis.

"},{"location":"api/orchestrators/#methods_1","title":"Methods","text":""},{"location":"api/orchestrators/#run_1","title":"run","text":"
async def run(\n    self,\n    query: str,\n    background_context: str = \"\",\n    max_iterations_per_section: int | None = None,\n    max_time_minutes: float | None = None,\n    token_budget: int | None = None\n) -> AsyncGenerator[AgentEvent, None]\n

Runs deep research flow.

Parameters: - query: Research query string - background_context: Background context (default: \"\") - max_iterations_per_section: Maximum iterations per section (default: from settings) - max_time_minutes: Maximum time in minutes (default: from settings) - token_budget: Token budget (default: from settings)

Yields: AgentEvent objects for: - started: Research started - planning: Creating research plan - looping: Running parallel research loops - synthesizing: Synthesizing results - complete: Research completed - error: Error occurred

"},{"location":"api/orchestrators/#graphorchestrator","title":"GraphOrchestrator","text":"

Module: src.orchestrator.graph_orchestrator

Purpose: Graph-based execution using Pydantic AI agents as nodes.

"},{"location":"api/orchestrators/#methods_2","title":"Methods","text":""},{"location":"api/orchestrators/#run_2","title":"run","text":"
async def run(\n    self,\n    query: str,\n    research_mode: str = \"auto\",\n    use_graph: bool = True\n) -> AsyncGenerator[AgentEvent, None]\n

Runs graph-based research orchestration.

Parameters: - query: Research query string - research_mode: Research mode (\"iterative\", \"deep\", or \"auto\") - use_graph: Whether to use graph execution (default: True)

Yields: AgentEvent objects during graph execution.

"},{"location":"api/orchestrators/#orchestrator-factory","title":"Orchestrator Factory","text":"

Module: src.orchestrator_factory

Purpose: Factory for creating orchestrators.

"},{"location":"api/orchestrators/#functions","title":"Functions","text":""},{"location":"api/orchestrators/#create_orchestrator","title":"create_orchestrator","text":"
def create_orchestrator(\n    search_handler: SearchHandlerProtocol,\n    judge_handler: JudgeHandlerProtocol,\n    config: dict[str, Any],\n    mode: str | None = None\n) -> Any\n

Creates an orchestrator instance.

Parameters: - search_handler: Search handler protocol implementation - judge_handler: Judge handler protocol implementation - config: Configuration dictionary - mode: Orchestrator mode (\"simple\", \"advanced\", \"magentic\", or None for auto-detect)

Returns: Orchestrator instance.

Raises: - ValueError: If requirements not met

Modes: - \"simple\": Legacy orchestrator - \"advanced\" or \"magentic\": Magentic orchestrator (requires OpenAI API key) - None: Auto-detect based on API key availability

"},{"location":"api/orchestrators/#magenticorchestrator","title":"MagenticOrchestrator","text":"

Module: src.orchestrator_magentic

Purpose: Multi-agent coordination using Microsoft Agent Framework.

"},{"location":"api/orchestrators/#methods_3","title":"Methods","text":""},{"location":"api/orchestrators/#run_3","title":"run","text":"
async def run(\n    self,\n    query: str,\n    max_rounds: int = 15,\n    max_stalls: int = 3\n) -> AsyncGenerator[AgentEvent, None]\n

Runs Magentic orchestration.

Parameters: - query: Research query string - max_rounds: Maximum rounds (default: 15) - max_stalls: Maximum stalls before reset (default: 3)

Yields: AgentEvent objects converted from Magentic events.

Requirements: - agent-framework-core package - OpenAI API key

"},{"location":"api/orchestrators/#see-also","title":"See Also","text":""},{"location":"api/services/","title":"Services API Reference","text":"

This page documents the API for DeepCritical services.

"},{"location":"api/services/#embeddingservice","title":"EmbeddingService","text":"

Module: src.services.embeddings

Purpose: Local sentence-transformers for semantic search and deduplication.

"},{"location":"api/services/#methods","title":"Methods","text":""},{"location":"api/services/#embed","title":"embed","text":"
async def embed(self, text: str) -> list[float]\n

Generates embedding for a text string.

Parameters: - text: Text to embed

Returns: Embedding vector as list of floats.

"},{"location":"api/services/#embed_batch","title":"embed_batch","text":"
async def embed_batch(self, texts: list[str]) -> list[list[float]]\n

Generates embeddings for multiple texts.

Parameters: - texts: List of texts to embed

Returns: List of embedding vectors.

"},{"location":"api/services/#similarity","title":"similarity","text":"
async def similarity(self, text1: str, text2: str) -> float\n

Calculates similarity between two texts.

Parameters: - text1: First text - text2: Second text

Returns: Similarity score (0.0-1.0).

"},{"location":"api/services/#find_duplicates","title":"find_duplicates","text":"
async def find_duplicates(\n    self,\n    texts: list[str],\n    threshold: float = 0.85\n) -> list[tuple[int, int]]\n

Finds duplicate texts based on similarity threshold.

Parameters: - texts: List of texts to check - threshold: Similarity threshold (default: 0.85)

Returns: List of (index1, index2) tuples for duplicate pairs.

"},{"location":"api/services/#factory-function","title":"Factory Function","text":""},{"location":"api/services/#get_embedding_service","title":"get_embedding_service","text":"
@lru_cache(maxsize=1)\ndef get_embedding_service() -> EmbeddingService\n

Returns singleton EmbeddingService instance.

"},{"location":"api/services/#llamaindexragservice","title":"LlamaIndexRAGService","text":"

Module: src.services.rag

Purpose: Retrieval-Augmented Generation using LlamaIndex.

"},{"location":"api/services/#methods_1","title":"Methods","text":""},{"location":"api/services/#ingest_evidence","title":"ingest_evidence","text":"
async def ingest_evidence(self, evidence: list[Evidence]) -> None\n

Ingests evidence into RAG service.

Parameters: - evidence: List of Evidence objects to ingest

Note: Requires OpenAI API key for embeddings.

"},{"location":"api/services/#retrieve","title":"retrieve","text":"
async def retrieve(\n    self,\n    query: str,\n    top_k: int = 5\n) -> list[Document]\n

Retrieves relevant documents for a query.

Parameters: - query: Search query string - top_k: Number of top results to return (default: 5)

Returns: List of Document objects with metadata.

"},{"location":"api/services/#query","title":"query","text":"
async def query(\n    self,\n    query: str,\n    top_k: int = 5\n) -> str\n

Queries RAG service and returns formatted results.

Parameters: - query: Search query string - top_k: Number of top results to return (default: 5)

Returns: Formatted query results as string.

"},{"location":"api/services/#factory-function_1","title":"Factory Function","text":""},{"location":"api/services/#get_rag_service","title":"get_rag_service","text":"
@lru_cache(maxsize=1)\ndef get_rag_service() -> LlamaIndexRAGService | None\n

Returns singleton LlamaIndexRAGService instance, or None if OpenAI key not available.

"},{"location":"api/services/#statisticalanalyzer","title":"StatisticalAnalyzer","text":"

Module: src.services.statistical_analyzer

Purpose: Secure execution of AI-generated statistical code.

"},{"location":"api/services/#methods_2","title":"Methods","text":""},{"location":"api/services/#analyze","title":"analyze","text":"
async def analyze(\n    self,\n    hypothesis: str,\n    evidence: list[Evidence],\n    data_description: str | None = None\n) -> AnalysisResult\n

Analyzes a hypothesis using statistical methods.

Parameters: - hypothesis: Hypothesis to analyze - evidence: List of Evidence objects - data_description: Optional data description

Returns: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - code: Generated analysis code - output: Execution output - error: Error message if execution failed

Note: Requires Modal credentials for sandbox execution.

"},{"location":"api/services/#see-also","title":"See Also","text":""},{"location":"api/tools/","title":"Tools API Reference","text":"

This page documents the API for DeepCritical search tools.

"},{"location":"api/tools/#searchtool-protocol","title":"SearchTool Protocol","text":"

All tools implement the SearchTool protocol:

class SearchTool(Protocol):\n    @property\n    def name(self) -> str: ...\n    \n    async def search(\n        self, \n        query: str, \n        max_results: int = 10\n    ) -> list[Evidence]: ...\n
"},{"location":"api/tools/#pubmedtool","title":"PubMedTool","text":"

Module: src.tools.pubmed

Purpose: Search peer-reviewed biomedical literature from PubMed.

"},{"location":"api/tools/#properties","title":"Properties","text":""},{"location":"api/tools/#name","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"pubmed\"

"},{"location":"api/tools/#methods","title":"Methods","text":""},{"location":"api/tools/#search","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches PubMed for articles.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with PubMed articles.

Raises: - SearchError: If search fails - RateLimitError: If rate limit is exceeded

"},{"location":"api/tools/#clinicaltrialstool","title":"ClinicalTrialsTool","text":"

Module: src.tools.clinicaltrials

Purpose: Search ClinicalTrials.gov for interventional studies.

"},{"location":"api/tools/#properties_1","title":"Properties","text":""},{"location":"api/tools/#name_1","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"clinicaltrials\"

"},{"location":"api/tools/#methods_1","title":"Methods","text":""},{"location":"api/tools/#search_1","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches ClinicalTrials.gov for trials.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with clinical trials.

Note: Only returns interventional studies with status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION

Raises: - SearchError: If search fails

"},{"location":"api/tools/#europepmctool","title":"EuropePMCTool","text":"

Module: src.tools.europepmc

Purpose: Search Europe PMC for preprints and peer-reviewed articles.

"},{"location":"api/tools/#properties_2","title":"Properties","text":""},{"location":"api/tools/#name_2","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"europepmc\"

"},{"location":"api/tools/#methods_2","title":"Methods","text":""},{"location":"api/tools/#search_2","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches Europe PMC for articles and preprints.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with articles/preprints.

Note: Includes both preprints (marked with [PREPRINT - Not peer-reviewed]) and peer-reviewed articles.

Raises: - SearchError: If search fails

"},{"location":"api/tools/#ragtool","title":"RAGTool","text":"

Module: src.tools.rag_tool

Purpose: Semantic search within collected evidence.

"},{"location":"api/tools/#properties_3","title":"Properties","text":""},{"location":"api/tools/#name_3","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"rag\"

"},{"location":"api/tools/#methods_3","title":"Methods","text":""},{"location":"api/tools/#search_3","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches collected evidence using semantic similarity.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects from collected evidence.

Note: Requires evidence to be ingested into RAG service first.

"},{"location":"api/tools/#searchhandler","title":"SearchHandler","text":"

Module: src.tools.search_handler

Purpose: Orchestrates parallel searches across multiple tools.

"},{"location":"api/tools/#methods_4","title":"Methods","text":""},{"location":"api/tools/#search_4","title":"search","text":"
async def search(\n    self,\n    query: str,\n    tools: list[SearchTool] | None = None,\n    max_results_per_tool: int = 10\n) -> SearchResult\n

Searches multiple tools in parallel.

Parameters: - query: Search query string - tools: List of tools to use (default: all available tools) - max_results_per_tool: Maximum results per tool (default: 10)

Returns: SearchResult with: - evidence: Aggregated list of evidence - tool_results: Results per tool - total_count: Total number of results

Note: Uses asyncio.gather() for parallel execution. Handles tool failures gracefully.

"},{"location":"api/tools/#see-also","title":"See Also","text":""},{"location":"architecture/agents/","title":"Agents Architecture","text":"

DeepCritical uses Pydantic AI agents for all AI-powered operations. All agents follow a consistent pattern and use structured output types.

"},{"location":"architecture/agents/#agent-pattern","title":"Agent Pattern","text":"

All agents use the Pydantic AI Agent class with the following structure:

"},{"location":"architecture/agents/#model-initialization","title":"Model Initialization","text":"

Agents use get_model() from src/agent_factory/judges.py if no model is provided. This supports:

The model selection is based on the configured LLM_PROVIDER in settings.

"},{"location":"architecture/agents/#error-handling","title":"Error Handling","text":"

Agents return fallback values on failure rather than raising exceptions:

All errors are logged with context using structlog.

"},{"location":"architecture/agents/#input-validation","title":"Input Validation","text":"

All agents validate inputs:

"},{"location":"architecture/agents/#output-types","title":"Output Types","text":"

Agents use structured output types from src/utils/models.py:

For text output (writer agents), agents return str directly.

"},{"location":"architecture/agents/#agent-types","title":"Agent Types","text":""},{"location":"architecture/agents/#knowledge-gap-agent","title":"Knowledge Gap Agent","text":"

File: src/agents/knowledge_gap.py

Purpose: Evaluates research state and identifies knowledge gaps.

Output: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

Methods: - async def evaluate(query, background_context, conversation_history, iteration, time_elapsed_minutes, max_time_minutes) -> KnowledgeGapOutput

"},{"location":"architecture/agents/#tool-selector-agent","title":"Tool Selector Agent","text":"

File: src/agents/tool_selector.py

Purpose: Selects appropriate tools for addressing knowledge gaps.

Output: AgentSelectionPlan with list of AgentTask objects.

Available Agents: - WebSearchAgent: General web search for fresh information - SiteCrawlerAgent: Research specific entities/companies - RAGAgent: Semantic search within collected evidence

"},{"location":"architecture/agents/#writer-agent","title":"Writer Agent","text":"

File: src/agents/writer.py

Purpose: Generates final reports from research findings.

Output: Markdown string with numbered citations.

Methods: - async def write_report(query, findings, output_length, output_instructions) -> str

Features: - Validates inputs - Truncates very long findings (max 50000 chars) with warning - Retry logic for transient failures (3 retries) - Citation validation before returning

"},{"location":"architecture/agents/#long-writer-agent","title":"Long Writer Agent","text":"

File: src/agents/long_writer.py

Purpose: Long-form report generation with section-by-section writing.

Input/Output: Uses ReportDraft models.

Methods: - async def write_next_section(query, draft, section_title, section_content) -> LongWriterOutput - async def write_report(query, report_title, report_draft) -> str

Features: - Writes sections iteratively - Aggregates references across sections - Reformats section headings and references - Deduplicates and renumbers references

"},{"location":"architecture/agents/#proofreader-agent","title":"Proofreader Agent","text":"

File: src/agents/proofreader.py

Purpose: Proofreads and polishes report drafts.

Input: ReportDraft Output: Polished markdown string

Methods: - async def proofread(query, report_title, report_draft) -> str

Features: - Removes duplicate content across sections - Adds executive summary if multiple sections - Preserves all references and citations - Improves flow and readability

"},{"location":"architecture/agents/#thinking-agent","title":"Thinking Agent","text":"

File: src/agents/thinking.py

Purpose: Generates observations from conversation history.

Output: Observation string

Methods: - async def generate_observations(query, background_context, conversation_history) -> str

"},{"location":"architecture/agents/#input-parser-agent","title":"Input Parser Agent","text":"

File: src/agents/input_parser.py

Purpose: Parses and improves user queries, detects research mode.

Output: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: \"iterative\" or \"deep\" - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"architecture/agents/#factory-functions","title":"Factory Functions","text":"

All agents have factory functions in src/agent_factory/agents.py:

Factory functions: - Use get_model() if no model provided - Raise ConfigurationError if creation fails - Log agent creation

"},{"location":"architecture/agents/#see-also","title":"See Also","text":""},{"location":"architecture/graph-orchestration/","title":"Graph Orchestration Architecture","text":""},{"location":"architecture/graph-orchestration/#overview","title":"Overview","text":"

Phase 4 implements a graph-based orchestration system for research workflows using Pydantic AI agents as nodes. This enables better parallel execution, conditional routing, and state management compared to simple agent chains.

"},{"location":"architecture/graph-orchestration/#graph-structure","title":"Graph Structure","text":""},{"location":"architecture/graph-orchestration/#nodes","title":"Nodes","text":"

Graph nodes represent different stages in the research workflow:

  1. Agent Nodes: Execute Pydantic AI agents
  2. Input: Prompt/query
  3. Output: Structured or unstructured response

  4. Examples: KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent

  5. State Nodes: Update or read workflow state

  6. Input: Current state
  7. Output: Updated state

  8. Examples: Update evidence, update conversation history

  9. Decision Nodes: Make routing decisions based on conditions

  10. Input: Current state/results
  11. Output: Next node ID

  12. Examples: Continue research vs. complete research

  13. Parallel Nodes: Execute multiple nodes concurrently

  14. Input: List of node IDs
  15. Output: Aggregated results
  16. Examples: Parallel iterative research loops
"},{"location":"architecture/graph-orchestration/#edges","title":"Edges","text":"

Edges define transitions between nodes:

  1. Sequential Edges: Always traversed (no condition)
  2. From: Source node
  3. To: Target node

  4. Condition: None (always True)

  5. Conditional Edges: Traversed based on condition

  6. From: Source node
  7. To: Target node
  8. Condition: Callable that returns bool

  9. Example: If research complete \u2192 go to writer, else \u2192 continue loop

  10. Parallel Edges: Used for parallel execution branches

  11. From: Parallel node
  12. To: Multiple target nodes
  13. Execution: All targets run concurrently
"},{"location":"architecture/graph-orchestration/#graph-patterns","title":"Graph Patterns","text":""},{"location":"architecture/graph-orchestration/#iterative-research-graph","title":"Iterative Research Graph","text":"
[Input] \u2192 [Thinking] \u2192 [Knowledge Gap] \u2192 [Decision: Complete?]\n                                              \u2193 No          \u2193 Yes\n                                    [Tool Selector]    [Writer]\n                                              \u2193\n                                    [Execute Tools] \u2192 [Loop Back]\n
"},{"location":"architecture/graph-orchestration/#deep-research-graph","title":"Deep Research Graph","text":"
[Input] \u2192 [Planner] \u2192 [Parallel Iterative Loops] \u2192 [Synthesizer]\n                           \u2193         \u2193         \u2193\n                        [Loop1]  [Loop2]  [Loop3]\n
"},{"location":"architecture/graph-orchestration/#state-management","title":"State Management","text":"

State is managed via WorkflowState using ContextVar for thread-safe isolation:

State transitions occur at state nodes, which update the global workflow state.

"},{"location":"architecture/graph-orchestration/#execution-flow","title":"Execution Flow","text":"
  1. Graph Construction: Build graph from nodes and edges
  2. Graph Validation: Ensure graph is valid (no cycles, all nodes reachable)
  3. Graph Execution: Traverse graph from entry node
  4. Node Execution: Execute each node based on type
  5. Edge Evaluation: Determine next node(s) based on edges
  6. Parallel Execution: Use asyncio.gather() for parallel nodes
  7. State Updates: Update state at state nodes
  8. Event Streaming: Yield events during execution for UI
"},{"location":"architecture/graph-orchestration/#conditional-routing","title":"Conditional Routing","text":"

Decision nodes evaluate conditions and return next node IDs:

"},{"location":"architecture/graph-orchestration/#parallel-execution","title":"Parallel Execution","text":"

Parallel nodes execute multiple nodes concurrently:

"},{"location":"architecture/graph-orchestration/#budget-enforcement","title":"Budget Enforcement","text":"

Budget constraints are enforced at decision nodes:

If any budget is exceeded, execution routes to exit node.

"},{"location":"architecture/graph-orchestration/#error-handling","title":"Error Handling","text":"

Errors are handled at multiple levels:

  1. Node Level: Catch errors in individual node execution
  2. Graph Level: Handle errors during graph traversal
  3. State Level: Rollback state changes on error

Errors are logged and yield error events for UI.

"},{"location":"architecture/graph-orchestration/#backward-compatibility","title":"Backward Compatibility","text":"

Graph execution is optional via feature flag:

This allows gradual migration and fallback if needed.

"},{"location":"architecture/graph_orchestration/","title":"Graph Orchestration Architecture","text":""},{"location":"architecture/graph_orchestration/#graph-patterns","title":"Graph Patterns","text":""},{"location":"architecture/graph_orchestration/#iterative-research-graph","title":"Iterative Research Graph","text":"
[Input] \u2192 [Thinking] \u2192 [Knowledge Gap] \u2192 [Decision: Complete?]\n                                              \u2193 No          \u2193 Yes\n                                    [Tool Selector]    [Writer]\n                                              \u2193\n                                    [Execute Tools] \u2192 [Loop Back]\n
"},{"location":"architecture/graph_orchestration/#deep-research-graph","title":"Deep Research Graph","text":"
[Input] \u2192 [Planner] \u2192 [Parallel Iterative Loops] \u2192 [Synthesizer]\n                           \u2193         \u2193         \u2193\n                        [Loop1]  [Loop2]  [Loop3]\n
"},{"location":"architecture/graph_orchestration/#deep-research","title":"Deep Research","text":"
\nsequenceDiagram\n    actor User\n    participant GraphOrchestrator\n    participant InputParser\n    participant GraphBuilder\n    participant GraphExecutor\n    participant Agent\n    participant BudgetTracker\n    participant WorkflowState\n\n    User->>GraphOrchestrator: run(query)\n    GraphOrchestrator->>InputParser: detect_research_mode(query)\n    InputParser-->>GraphOrchestrator: mode (iterative/deep)\n    GraphOrchestrator->>GraphBuilder: build_graph(mode)\n    GraphBuilder-->>GraphOrchestrator: ResearchGraph\n    GraphOrchestrator->>WorkflowState: init_workflow_state()\n    GraphOrchestrator->>BudgetTracker: create_budget()\n    GraphOrchestrator->>GraphExecutor: _execute_graph(graph)\n    \n    loop For each node in graph\n        GraphExecutor->>Agent: execute_node(agent_node)\n        Agent->>Agent: process_input\n        Agent-->>GraphExecutor: result\n        GraphExecutor->>WorkflowState: update_state(result)\n        GraphExecutor->>BudgetTracker: add_tokens(used)\n        GraphExecutor->>BudgetTracker: check_budget()\n        alt Budget exceeded\n            GraphExecutor->>GraphOrchestrator: emit(error_event)\n        else Continue\n            GraphExecutor->>GraphOrchestrator: emit(progress_event)\n        end\n    end\n    \n    GraphOrchestrator->>User: AsyncGenerator[AgentEvent]\n
"},{"location":"architecture/graph_orchestration/#iterative-research","title":"Iterative Research","text":"
sequenceDiagram\n    participant IterativeFlow\n    participant ThinkingAgent\n    participant KnowledgeGapAgent\n    participant ToolSelector\n    participant ToolExecutor\n    participant JudgeHandler\n    participant WriterAgent\n\n    IterativeFlow->>IterativeFlow: run(query)\n    \n    loop Until complete or max_iterations\n        IterativeFlow->>ThinkingAgent: generate_observations()\n        ThinkingAgent-->>IterativeFlow: observations\n        \n        IterativeFlow->>KnowledgeGapAgent: evaluate_gaps()\n        KnowledgeGapAgent-->>IterativeFlow: KnowledgeGapOutput\n        \n        alt Research complete\n            IterativeFlow->>WriterAgent: create_final_report()\n            WriterAgent-->>IterativeFlow: final_report\n        else Gaps remain\n            IterativeFlow->>ToolSelector: select_agents(gap)\n            ToolSelector-->>IterativeFlow: AgentSelectionPlan\n            \n            IterativeFlow->>ToolExecutor: execute_tool_tasks()\n            ToolExecutor-->>IterativeFlow: ToolAgentOutput[]\n            \n            IterativeFlow->>JudgeHandler: assess_evidence()\n            JudgeHandler-->>IterativeFlow: should_continue\n        end\n    end
"},{"location":"architecture/graph_orchestration/#graph-structure","title":"Graph Structure","text":""},{"location":"architecture/graph_orchestration/#nodes","title":"Nodes","text":"

Graph nodes represent different stages in the research workflow:

  1. Agent Nodes: Execute Pydantic AI agents
  2. Input: Prompt/query
  3. Output: Structured or unstructured response

  4. Examples: KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent

  5. State Nodes: Update or read workflow state

  6. Input: Current state
  7. Output: Updated state

  8. Examples: Update evidence, update conversation history

  9. Decision Nodes: Make routing decisions based on conditions

  10. Input: Current state/results
  11. Output: Next node ID

  12. Examples: Continue research vs. complete research

  13. Parallel Nodes: Execute multiple nodes concurrently

  14. Input: List of node IDs
  15. Output: Aggregated results
  16. Examples: Parallel iterative research loops
"},{"location":"architecture/graph_orchestration/#edges","title":"Edges","text":"

Edges define transitions between nodes:

  1. Sequential Edges: Always traversed (no condition)
  2. From: Source node
  3. To: Target node

  4. Condition: None (always True)

  5. Conditional Edges: Traversed based on condition

  6. From: Source node
  7. To: Target node
  8. Condition: Callable that returns bool

  9. Example: If research complete \u2192 go to writer, else \u2192 continue loop

  10. Parallel Edges: Used for parallel execution branches

  11. From: Parallel node
  12. To: Multiple target nodes
  13. Execution: All targets run concurrently
"},{"location":"architecture/graph_orchestration/#state-management","title":"State Management","text":"

State is managed via WorkflowState using ContextVar for thread-safe isolation:

State transitions occur at state nodes, which update the global workflow state.

"},{"location":"architecture/graph_orchestration/#execution-flow","title":"Execution Flow","text":"
  1. Graph Construction: Build graph from nodes and edges
  2. Graph Validation: Ensure graph is valid (no cycles, all nodes reachable)
  3. Graph Execution: Traverse graph from entry node
  4. Node Execution: Execute each node based on type
  5. Edge Evaluation: Determine next node(s) based on edges
  6. Parallel Execution: Use asyncio.gather() for parallel nodes
  7. State Updates: Update state at state nodes
  8. Event Streaming: Yield events during execution for UI
"},{"location":"architecture/graph_orchestration/#conditional-routing","title":"Conditional Routing","text":"

Decision nodes evaluate conditions and return next node IDs:

"},{"location":"architecture/graph_orchestration/#parallel-execution","title":"Parallel Execution","text":"

Parallel nodes execute multiple nodes concurrently:

"},{"location":"architecture/graph_orchestration/#budget-enforcement","title":"Budget Enforcement","text":"

Budget constraints are enforced at decision nodes:

If any budget is exceeded, execution routes to exit node.

"},{"location":"architecture/graph_orchestration/#error-handling","title":"Error Handling","text":"

Errors are handled at multiple levels:

  1. Node Level: Catch errors in individual node execution
  2. Graph Level: Handle errors during graph traversal
  3. State Level: Rollback state changes on error

Errors are logged and yield error events for UI.

"},{"location":"architecture/graph_orchestration/#backward-compatibility","title":"Backward Compatibility","text":"

Graph execution is optional via feature flag:

This allows gradual migration and fallback if needed.

"},{"location":"architecture/graph_orchestration/#see-also","title":"See Also","text":""},{"location":"architecture/middleware/","title":"Middleware Architecture","text":"

DeepCritical uses middleware for state management, budget tracking, and workflow coordination.

"},{"location":"architecture/middleware/#state-management","title":"State Management","text":""},{"location":"architecture/middleware/#workflowstate","title":"WorkflowState","text":"

File: src/middleware/state_machine.py

Purpose: Thread-safe state management for research workflows

Implementation: Uses ContextVar for thread-safe isolation

State Components: - evidence: list[Evidence]: Collected evidence from searches - conversation: Conversation: Iteration history (gaps, tool calls, findings, thoughts) - embedding_service: Any: Embedding service for semantic search

Methods: - add_evidence(evidence: Evidence): Adds evidence with URL-based deduplication - async search_related(query: str, top_k: int = 5) -> list[Evidence]: Semantic search

Initialization:

Access:

"},{"location":"architecture/middleware/#workflow-manager","title":"Workflow Manager","text":"

File: src/middleware/workflow_manager.py

Purpose: Coordinates parallel research loops

Methods: - add_loop(loop: ResearchLoop): Add a research loop to manage - async run_loops_parallel() -> list[ResearchLoop]: Run all loops in parallel - update_loop_status(loop_id: str, status: str): Update loop status - sync_loop_evidence_to_state(): Synchronize evidence from loops to global state

Features: - Uses asyncio.gather() for parallel execution - Handles errors per loop (doesn't fail all if one fails) - Tracks loop status: pending, running, completed, failed, cancelled - Evidence deduplication across parallel loops

Usage: 

from src.middleware.workflow_manager import WorkflowManager\n\nmanager = WorkflowManager()\nmanager.add_loop(loop1)\nmanager.add_loop(loop2)\ncompleted_loops = await manager.run_loops_parallel()\n

"},{"location":"architecture/middleware/#budget-tracker","title":"Budget Tracker","text":"

File: src/middleware/budget_tracker.py

Purpose: Tracks and enforces resource limits

Budget Components: - Tokens: LLM token usage - Time: Elapsed time in seconds - Iterations: Number of iterations

Methods: - create_budget(token_limit, time_limit_seconds, iterations_limit) -> BudgetStatus - add_tokens(tokens: int): Add token usage - start_timer(): Start time tracking - update_timer(): Update elapsed time - increment_iteration(): Increment iteration count - check_budget() -> BudgetStatus: Check current budget status - can_continue() -> bool: Check if research can continue

Token Estimation: - estimate_tokens(text: str) -> int: ~4 chars per token - estimate_llm_call_tokens(prompt: str, response: str) -> int: Estimate LLM call tokens

Usage: 

from src.middleware.budget_tracker import BudgetTracker\n\ntracker = BudgetTracker()\nbudget = tracker.create_budget(\n    token_limit=100000,\n    time_limit_seconds=600,\n    iterations_limit=10\n)\ntracker.start_timer()\n# ... research operations ...\nif not tracker.can_continue():\n    # Budget exceeded, stop research\n    pass\n

"},{"location":"architecture/middleware/#models","title":"Models","text":"

All middleware models are defined in src/utils/models.py:

"},{"location":"architecture/middleware/#thread-safety","title":"Thread Safety","text":"

All middleware components use ContextVar for thread-safe isolation:

"},{"location":"architecture/middleware/#see-also","title":"See Also","text":""},{"location":"architecture/orchestrators/","title":"Orchestrators Architecture","text":"

DeepCritical supports multiple orchestration patterns for research workflows.

"},{"location":"architecture/orchestrators/#research-flows","title":"Research Flows","text":""},{"location":"architecture/orchestrators/#iterativeresearchflow","title":"IterativeResearchFlow","text":"

File: src/orchestrator/research_flow.py

Pattern: Generate observations \u2192 Evaluate gaps \u2192 Select tools \u2192 Execute \u2192 Judge \u2192 Continue/Complete

Agents Used: - KnowledgeGapAgent: Evaluates research completeness - ToolSelectorAgent: Selects tools for addressing gaps - ThinkingAgent: Generates observations - WriterAgent: Creates final report - JudgeHandler: Assesses evidence sufficiency

Features: - Tracks iterations, time, budget - Supports graph execution (use_graph=True) and agent chains (use_graph=False) - Iterates until research complete or constraints met

Usage:

"},{"location":"architecture/orchestrators/#deepresearchflow","title":"DeepResearchFlow","text":"

File: src/orchestrator/research_flow.py

Pattern: Planner \u2192 Parallel iterative loops per section \u2192 Synthesizer

Agents Used: - PlannerAgent: Breaks query into report sections - IterativeResearchFlow: Per-section research (parallel) - LongWriterAgent or ProofreaderAgent: Final synthesis

Features: - Uses WorkflowManager for parallel execution - Budget tracking per section and globally - State synchronization across parallel loops - Supports graph execution and agent chains

Usage:

"},{"location":"architecture/orchestrators/#graph-orchestrator","title":"Graph Orchestrator","text":"

File: src/orchestrator/graph_orchestrator.py

Purpose: Graph-based execution using Pydantic AI agents as nodes

Features: - Uses Pydantic AI Graphs (when available) or agent chains (fallback) - Routes based on research mode (iterative/deep/auto) - Streams AgentEvent objects for UI

Node Types: - Agent Nodes: Execute Pydantic AI agents - State Nodes: Update or read workflow state - Decision Nodes: Make routing decisions - Parallel Nodes: Execute multiple nodes concurrently

Edge Types: - Sequential Edges: Always traversed - Conditional Edges: Traversed based on condition - Parallel Edges: Used for parallel execution branches

"},{"location":"architecture/orchestrators/#orchestrator-factory","title":"Orchestrator Factory","text":"

File: src/orchestrator_factory.py

Purpose: Factory for creating orchestrators

Modes: - Simple: Legacy orchestrator (backward compatible) - Advanced: Magentic orchestrator (requires OpenAI API key) - Auto-detect: Chooses based on API key availability

Usage:

"},{"location":"architecture/orchestrators/#magentic-orchestrator","title":"Magentic Orchestrator","text":"

File: src/orchestrator_magentic.py

Purpose: Multi-agent coordination using Microsoft Agent Framework

Features: - Uses agent-framework-core - ChatAgent pattern with internal LLMs per agent - MagenticBuilder with participants: searcher, hypothesizer, judge, reporter - Manager orchestrates agents via OpenAIChatClient - Requires OpenAI API key (function calling support) - Event-driven: converts Magentic events to AgentEvent for UI streaming

Requirements: - agent-framework-core package - OpenAI API key

"},{"location":"architecture/orchestrators/#hierarchical-orchestrator","title":"Hierarchical Orchestrator","text":"

File: src/orchestrator_hierarchical.py

Purpose: Hierarchical orchestrator using middleware and sub-teams

Features: - Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge - Adapts Magentic ChatAgent to SubIterationTeam protocol - Event-driven via asyncio.Queue for coordination - Supports sub-iteration patterns for complex research tasks

"},{"location":"architecture/orchestrators/#legacy-simple-mode","title":"Legacy Simple Mode","text":"

File: src/legacy_orchestrator.py

Purpose: Linear search-judge-synthesize loop

Features: - Uses SearchHandlerProtocol and JudgeHandlerProtocol - Generator-based design yielding AgentEvent objects - Backward compatibility for simple use cases

"},{"location":"architecture/orchestrators/#state-initialization","title":"State Initialization","text":"

All orchestrators must initialize workflow state:

"},{"location":"architecture/orchestrators/#event-streaming","title":"Event Streaming","text":"

All orchestrators yield AgentEvent objects:

Event Types: - started: Research started - search_complete: Search completed - judge_complete: Evidence evaluation completed - hypothesizing: Generating hypotheses - synthesizing: Synthesizing results - complete: Research completed - error: Error occurred

Event Structure:

"},{"location":"architecture/orchestrators/#see-also","title":"See Also","text":""},{"location":"architecture/services/","title":"Services Architecture","text":"

DeepCritical provides several services for embeddings, RAG, and statistical analysis.

"},{"location":"architecture/services/#embedding-service","title":"Embedding Service","text":"

File: src/services/embeddings.py

Purpose: Local sentence-transformers for semantic search and deduplication

Features: - No API Key Required: Uses local sentence-transformers models - Async-Safe: All operations use run_in_executor() to avoid blocking - ChromaDB Storage: Vector storage for embeddings - Deduplication: 0.85 similarity threshold (85% similarity = duplicate)

Model: Configurable via settings.local_embedding_model (default: all-MiniLM-L6-v2)

Methods: - async def embed(text: str) -> list[float]: Generate embeddings - async def embed_batch(texts: list[str]) -> list[list[float]]: Batch embedding - async def similarity(text1: str, text2: str) -> float: Calculate similarity - async def find_duplicates(texts: list[str], threshold: float = 0.85) -> list[tuple[int, int]]: Find duplicates

Usage: 

from src.services.embeddings import get_embedding_service\n\nservice = get_embedding_service()\nembedding = await service.embed(\"text to embed\")\n

"},{"location":"architecture/services/#llamaindex-rag-service","title":"LlamaIndex RAG Service","text":"

File: src/services/rag.py

Purpose: Retrieval-Augmented Generation using LlamaIndex

Features: - OpenAI Embeddings: Requires OPENAI_API_KEY - ChromaDB Storage: Vector database for document storage - Metadata Preservation: Preserves source, title, URL, date, authors - Lazy Initialization: Graceful fallback if OpenAI key not available

Methods: - async def ingest_evidence(evidence: list[Evidence]) -> None: Ingest evidence into RAG - async def retrieve(query: str, top_k: int = 5) -> list[Document]: Retrieve relevant documents - async def query(query: str, top_k: int = 5) -> str: Query with RAG

Usage: 

from src.services.rag import get_rag_service\n\nservice = get_rag_service()\nif service:\n    documents = await service.retrieve(\"query\", top_k=5)\n

"},{"location":"architecture/services/#statistical-analyzer","title":"Statistical Analyzer","text":"

File: src/services/statistical_analyzer.py

Purpose: Secure execution of AI-generated statistical code

Features: - Modal Sandbox: Secure, isolated execution environment - Code Generation: Generates Python code via LLM - Library Pinning: Version-pinned libraries in SANDBOX_LIBRARIES - Network Isolation: block_network=True by default

Libraries Available: - pandas, numpy, scipy - matplotlib, scikit-learn - statsmodels

Output: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - code: Generated analysis code - output: Execution output - error: Error message if execution failed

Usage: 

from src.services.statistical_analyzer import StatisticalAnalyzer\n\nanalyzer = StatisticalAnalyzer()\nresult = await analyzer.analyze(\n    hypothesis=\"Metformin reduces cancer risk\",\n    evidence=evidence_list\n)\n

"},{"location":"architecture/services/#singleton-pattern","title":"Singleton Pattern","text":"

All services use the singleton pattern with @lru_cache(maxsize=1):

@lru_cache(maxsize=1)\ndef get_embedding_service() -> EmbeddingService:\n    return EmbeddingService()\n

This ensures: - Single instance per process - Lazy initialization - No dependencies required at import time

"},{"location":"architecture/services/#service-availability","title":"Service Availability","text":"

Services check availability before use:

from src.utils.config import settings\n\nif settings.modal_available:\n    # Use Modal sandbox\n    pass\n\nif settings.has_openai_key:\n    # Use OpenAI embeddings for RAG\n    pass\n
"},{"location":"architecture/services/#see-also","title":"See Also","text":""},{"location":"architecture/tools/","title":"Tools Architecture","text":"

DeepCritical implements a protocol-based search tool system for retrieving evidence from multiple sources.

"},{"location":"architecture/tools/#searchtool-protocol","title":"SearchTool Protocol","text":"

All tools implement the SearchTool protocol from src/tools/base.py:

"},{"location":"architecture/tools/#rate-limiting","title":"Rate Limiting","text":"

All tools use the @retry decorator from tenacity:

@retry(\n    stop=stop_after_attempt(3), \n    wait=wait_exponential(...)\n)\nasync def search(self, query: str, max_results: int = 10) -> list[Evidence]:\n    # Implementation\n

Tools with API rate limits implement _rate_limit() method and use shared rate limiters from src/tools/rate_limiter.py.

"},{"location":"architecture/tools/#error-handling","title":"Error Handling","text":"

Tools raise custom exceptions:

Tools handle HTTP errors (429, 500, timeout) and return empty lists on non-critical errors (with warning logs).

"},{"location":"architecture/tools/#query-preprocessing","title":"Query Preprocessing","text":"

Tools use preprocess_query() from src/tools/query_utils.py to:

"},{"location":"architecture/tools/#evidence-conversion","title":"Evidence Conversion","text":"

All tools convert API responses to Evidence objects with:

Missing fields are handled gracefully with defaults.

"},{"location":"architecture/tools/#tool-implementations","title":"Tool Implementations","text":""},{"location":"architecture/tools/#pubmed-tool","title":"PubMed Tool","text":"

File: src/tools/pubmed.py

API: NCBI E-utilities (ESearch \u2192 EFetch)

Rate Limiting: - 0.34s between requests (3 req/sec without API key) - 0.1s between requests (10 req/sec with NCBI API key)

Features: - XML parsing with xmltodict - Handles single vs. multiple articles - Query preprocessing - Evidence conversion with metadata extraction

"},{"location":"architecture/tools/#clinicaltrials-tool","title":"ClinicalTrials Tool","text":"

File: src/tools/clinicaltrials.py

API: ClinicalTrials.gov API v2

Important: Uses requests library (NOT httpx) because WAF blocks httpx TLS fingerprint.

Execution: Runs in thread pool: await asyncio.to_thread(requests.get, ...)

Filtering: - Only interventional studies - Status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION

Features: - Parses nested JSON structure - Extracts trial metadata - Evidence conversion

"},{"location":"architecture/tools/#europe-pmc-tool","title":"Europe PMC Tool","text":"

File: src/tools/europepmc.py

API: Europe PMC REST API

Features: - Handles preprint markers: [PREPRINT - Not peer-reviewed] - Builds URLs from DOI or PMID - Checks pubTypeList for preprint detection - Includes both preprints and peer-reviewed articles

"},{"location":"architecture/tools/#rag-tool","title":"RAG Tool","text":"

File: src/tools/rag_tool.py

Purpose: Semantic search within collected evidence

Implementation: Wraps LlamaIndexRAGService

Features: - Returns Evidence from RAG results - Handles evidence ingestion - Semantic similarity search - Metadata preservation

"},{"location":"architecture/tools/#search-handler","title":"Search Handler","text":"

File: src/tools/search_handler.py

Purpose: Orchestrates parallel searches across multiple tools

Features: - Uses asyncio.gather() with return_exceptions=True - Aggregates results into SearchResult - Handles tool failures gracefully - Deduplicates results by URL

"},{"location":"architecture/tools/#tool-registration","title":"Tool Registration","text":"

Tools are registered in the search handler:

from src.tools.pubmed import PubMedTool\nfrom src.tools.clinicaltrials import ClinicalTrialsTool\nfrom src.tools.europepmc import EuropePMCTool\n\nsearch_handler = SearchHandler(\n    tools=[\n        PubMedTool(),\n        ClinicalTrialsTool(),\n        EuropePMCTool(),\n    ]\n)\n
"},{"location":"architecture/tools/#see-also","title":"See Also","text":""},{"location":"architecture/workflow-diagrams/","title":"DeepCritical Workflow - Simplified Magentic Architecture","text":"

Architecture Pattern: Microsoft Magentic Orchestration Design Philosophy: Simple, dynamic, manager-driven coordination Key Innovation: Intelligent manager replaces rigid sequential phases

"},{"location":"architecture/workflow-diagrams/#1-high-level-magentic-workflow","title":"1. High-Level Magentic Workflow","text":"
flowchart TD\n    Start([User Query]) --> Manager[Magentic Manager<br/>Plan \u2022 Select \u2022 Assess \u2022 Adapt]\n\n    Manager -->|Plans| Task1[Task Decomposition]\n    Task1 --> Manager\n\n    Manager -->|Selects & Executes| HypAgent[Hypothesis Agent]\n    Manager -->|Selects & Executes| SearchAgent[Search Agent]\n    Manager -->|Selects & Executes| AnalysisAgent[Analysis Agent]\n    Manager -->|Selects & Executes| ReportAgent[Report Agent]\n\n    HypAgent -->|Results| Manager\n    SearchAgent -->|Results| Manager\n    AnalysisAgent -->|Results| Manager\n    ReportAgent -->|Results| Manager\n\n    Manager -->|Assesses Quality| Decision{Good Enough?}\n    Decision -->|No - Refine| Manager\n    Decision -->|No - Different Agent| Manager\n    Decision -->|No - Stalled| Replan[Reset Plan]\n    Replan --> Manager\n\n    Decision -->|Yes| Synthesis[Synthesize Final Result]\n    Synthesis --> Output([Research Report])\n\n    style Start fill:#e1f5e1\n    style Manager fill:#ffe6e6\n    style HypAgent fill:#fff4e6\n    style SearchAgent fill:#fff4e6\n    style AnalysisAgent fill:#fff4e6\n    style ReportAgent fill:#fff4e6\n    style Decision fill:#ffd6d6\n    style Synthesis fill:#d4edda\n    style Output fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#2-magentic-manager-the-6-phase-cycle","title":"2. Magentic Manager: The 6-Phase Cycle","text":"
flowchart LR\n    P1[1. Planning<br/>Analyze task<br/>Create strategy] --> P2[2. Agent Selection<br/>Pick best agent<br/>for subtask]\n    P2 --> P3[3. Execution<br/>Run selected<br/>agent with tools]\n    P3 --> P4[4. Assessment<br/>Evaluate quality<br/>Check progress]\n    P4 --> Decision{Quality OK?<br/>Progress made?}\n    Decision -->|Yes| P6[6. Synthesis<br/>Combine results<br/>Generate report]\n    Decision -->|No| P5[5. Iteration<br/>Adjust plan<br/>Try again]\n    P5 --> P2\n    P6 --> Done([Complete])\n\n    style P1 fill:#fff4e6\n    style P2 fill:#ffe6e6\n    style P3 fill:#e6f3ff\n    style P4 fill:#ffd6d6\n    style P5 fill:#fff3cd\n    style P6 fill:#d4edda\n    style Done fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#3-simplified-agent-architecture","title":"3. Simplified Agent Architecture","text":"
graph TB\n    subgraph \"Orchestration Layer\"\n        Manager[Magentic Manager<br/>\u2022 Plans workflow<br/>\u2022 Selects agents<br/>\u2022 Assesses quality<br/>\u2022 Adapts strategy]\n        SharedContext[(Shared Context<br/>\u2022 Hypotheses<br/>\u2022 Search Results<br/>\u2022 Analysis<br/>\u2022 Progress)]\n        Manager <--> SharedContext\n    end\n\n    subgraph \"Specialist Agents\"\n        HypAgent[Hypothesis Agent<br/>\u2022 Domain understanding<br/>\u2022 Hypothesis generation<br/>\u2022 Testability refinement]\n        SearchAgent[Search Agent<br/>\u2022 Multi-source search<br/>\u2022 RAG retrieval<br/>\u2022 Result ranking]\n        AnalysisAgent[Analysis Agent<br/>\u2022 Evidence extraction<br/>\u2022 Statistical analysis<br/>\u2022 Code execution]\n        ReportAgent[Report Agent<br/>\u2022 Report assembly<br/>\u2022 Visualization<br/>\u2022 Citation formatting]\n    end\n\n    subgraph \"MCP Tools\"\n        WebSearch[Web Search<br/>PubMed \u2022 arXiv \u2022 bioRxiv]\n        CodeExec[Code Execution<br/>Sandboxed Python]\n        RAG[RAG Retrieval<br/>Vector DB \u2022 Embeddings]\n        Viz[Visualization<br/>Charts \u2022 Graphs]\n    end\n\n    Manager -->|Selects & Directs| HypAgent\n    Manager -->|Selects & Directs| SearchAgent\n    Manager -->|Selects & Directs| AnalysisAgent\n    Manager -->|Selects & Directs| ReportAgent\n\n    HypAgent --> SharedContext\n    SearchAgent --> SharedContext\n    AnalysisAgent --> SharedContext\n    ReportAgent --> SharedContext\n\n    SearchAgent --> WebSearch\n    SearchAgent --> RAG\n    AnalysisAgent --> CodeExec\n    ReportAgent --> CodeExec\n    ReportAgent --> Viz\n\n    style Manager fill:#ffe6e6\n    style SharedContext fill:#ffe6f0\n    style HypAgent fill:#fff4e6\n    style SearchAgent fill:#fff4e6\n    style AnalysisAgent fill:#fff4e6\n    style ReportAgent fill:#fff4e6\n    style WebSearch fill:#e6f3ff\n    style CodeExec fill:#e6f3ff\n    style RAG fill:#e6f3ff\n    style Viz fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#4-dynamic-workflow-example","title":"4. Dynamic Workflow Example","text":"
sequenceDiagram\n    participant User\n    participant Manager\n    participant HypAgent\n    participant SearchAgent\n    participant AnalysisAgent\n    participant ReportAgent\n\n    User->>Manager: \"Research protein folding in Alzheimer's\"\n\n    Note over Manager: PLAN: Generate hypotheses \u2192 Search \u2192 Analyze \u2192 Report\n\n    Manager->>HypAgent: Generate 3 hypotheses\n    HypAgent-->>Manager: Returns 3 hypotheses\n    Note over Manager: ASSESS: Good quality, proceed\n\n    Manager->>SearchAgent: Search literature for hypothesis 1\n    SearchAgent-->>Manager: Returns 15 papers\n    Note over Manager: ASSESS: Good results, continue\n\n    Manager->>SearchAgent: Search for hypothesis 2\n    SearchAgent-->>Manager: Only 2 papers found\n    Note over Manager: ASSESS: Insufficient, refine search\n\n    Manager->>SearchAgent: Refined query for hypothesis 2\n    SearchAgent-->>Manager: Returns 12 papers\n    Note over Manager: ASSESS: Better, proceed\n\n    Manager->>AnalysisAgent: Analyze evidence for all hypotheses\n    AnalysisAgent-->>Manager: Returns analysis with code\n    Note over Manager: ASSESS: Complete, generate report\n\n    Manager->>ReportAgent: Create comprehensive report\n    ReportAgent-->>Manager: Returns formatted report\n    Note over Manager: SYNTHESIZE: Combine all results\n\n    Manager->>User: Final Research Report
"},{"location":"architecture/workflow-diagrams/#5-manager-decision-logic","title":"5. Manager Decision Logic","text":"
flowchart TD\n    Start([Manager Receives Task]) --> Plan[Create Initial Plan]\n\n    Plan --> Select[Select Agent for Next Subtask]\n    Select --> Execute[Execute Agent]\n    Execute --> Collect[Collect Results]\n\n    Collect --> Assess[Assess Quality & Progress]\n\n    Assess --> Q1{Quality Sufficient?}\n    Q1 -->|No| Q2{Same Agent Can Fix?}\n    Q2 -->|Yes| Feedback[Provide Specific Feedback]\n    Feedback --> Execute\n    Q2 -->|No| Different[Try Different Agent]\n    Different --> Select\n\n    Q1 -->|Yes| Q3{Task Complete?}\n    Q3 -->|No| Q4{Making Progress?}\n    Q4 -->|Yes| Select\n    Q4 -->|No - Stalled| Replan[Reset Plan & Approach]\n    Replan --> Plan\n\n    Q3 -->|Yes| Synth[Synthesize Final Result]\n    Synth --> Done([Return Report])\n\n    style Start fill:#e1f5e1\n    style Plan fill:#fff4e6\n    style Select fill:#ffe6e6\n    style Execute fill:#e6f3ff\n    style Assess fill:#ffd6d6\n    style Q1 fill:#ffe6e6\n    style Q2 fill:#ffe6e6\n    style Q3 fill:#ffe6e6\n    style Q4 fill:#ffe6e6\n    style Synth fill:#d4edda\n    style Done fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#6-hypothesis-agent-workflow","title":"6. Hypothesis Agent Workflow","text":"
flowchart LR\n    Input[Research Query] --> Domain[Identify Domain<br/>& Key Concepts]\n    Domain --> Context[Retrieve Background<br/>Knowledge]\n    Context --> Generate[Generate 3-5<br/>Initial Hypotheses]\n    Generate --> Refine[Refine for<br/>Testability]\n    Refine --> Rank[Rank by<br/>Quality Score]\n    Rank --> Output[Return Top<br/>Hypotheses]\n\n    Output --> Struct[Hypothesis Structure:<br/>\u2022 Statement<br/>\u2022 Rationale<br/>\u2022 Testability Score<br/>\u2022 Data Requirements<br/>\u2022 Expected Outcomes]\n\n    style Input fill:#e1f5e1\n    style Output fill:#fff4e6\n    style Struct fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#7-search-agent-workflow","title":"7. Search Agent Workflow","text":"
flowchart TD\n    Input[Hypotheses] --> Strategy[Formulate Search<br/>Strategy per Hypothesis]\n\n    Strategy --> Multi[Multi-Source Search]\n\n    Multi --> PubMed[PubMed Search<br/>via MCP]\n    Multi --> ArXiv[arXiv Search<br/>via MCP]\n    Multi --> BioRxiv[bioRxiv Search<br/>via MCP]\n\n    PubMed --> Aggregate[Aggregate Results]\n    ArXiv --> Aggregate\n    BioRxiv --> Aggregate\n\n    Aggregate --> Filter[Filter & Rank<br/>by Relevance]\n    Filter --> Dedup[Deduplicate<br/>Cross-Reference]\n    Dedup --> Embed[Embed Documents<br/>via MCP]\n    Embed --> Vector[(Vector DB)]\n    Vector --> RAGRetrieval[RAG Retrieval<br/>Top-K per Hypothesis]\n    RAGRetrieval --> Output[Return Contextualized<br/>Search Results]\n\n    style Input fill:#fff4e6\n    style Multi fill:#ffe6e6\n    style Vector fill:#ffe6f0\n    style Output fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#8-analysis-agent-workflow","title":"8. Analysis Agent Workflow","text":"
flowchart TD\n    Input1[Hypotheses] --> Extract\n    Input2[Search Results] --> Extract[Extract Evidence<br/>per Hypothesis]\n\n    Extract --> Methods[Determine Analysis<br/>Methods Needed]\n\n    Methods --> Branch{Requires<br/>Computation?}\n    Branch -->|Yes| GenCode[Generate Python<br/>Analysis Code]\n    Branch -->|No| Qual[Qualitative<br/>Synthesis]\n\n    GenCode --> Execute[Execute Code<br/>via MCP Sandbox]\n    Execute --> Interpret1[Interpret<br/>Results]\n    Qual --> Interpret2[Interpret<br/>Findings]\n\n    Interpret1 --> Synthesize[Synthesize Evidence<br/>Across Sources]\n    Interpret2 --> Synthesize\n\n    Synthesize --> Verdict[Determine Verdict<br/>per Hypothesis]\n    Verdict --> Support[\u2022 Supported<br/>\u2022 Refuted<br/>\u2022 Inconclusive]\n    Support --> Gaps[Identify Knowledge<br/>Gaps & Limitations]\n    Gaps --> Output[Return Analysis<br/>Report]\n\n    style Input1 fill:#fff4e6\n    style Input2 fill:#e6f3ff\n    style Execute fill:#ffe6e6\n    style Output fill:#e6ffe6
"},{"location":"architecture/workflow-diagrams/#9-report-agent-workflow","title":"9. Report Agent Workflow","text":"
flowchart TD\n    Input1[Query] --> Assemble\n    Input2[Hypotheses] --> Assemble\n    Input3[Search Results] --> Assemble\n    Input4[Analysis] --> Assemble[Assemble Report<br/>Sections]\n\n    Assemble --> Exec[Executive Summary]\n    Assemble --> Intro[Introduction]\n    Assemble --> Methods[Methods]\n    Assemble --> Results[Results per<br/>Hypothesis]\n    Assemble --> Discussion[Discussion]\n    Assemble --> Future[Future Directions]\n    Assemble --> Refs[References]\n\n    Results --> VizCheck{Needs<br/>Visualization?}\n    VizCheck -->|Yes| GenViz[Generate Viz Code]\n    GenViz --> ExecViz[Execute via MCP<br/>Create Charts]\n    ExecViz --> Combine\n    VizCheck -->|No| Combine[Combine All<br/>Sections]\n\n    Exec --> Combine\n    Intro --> Combine\n    Methods --> Combine\n    Discussion --> Combine\n    Future --> Combine\n    Refs --> Combine\n\n    Combine --> Format[Format Output]\n    Format --> MD[Markdown]\n    Format --> PDF[PDF]\n    Format --> JSON[JSON]\n\n    MD --> Output[Return Final<br/>Report]\n    PDF --> Output\n    JSON --> Output\n\n    style Input1 fill:#e1f5e1\n    style Input2 fill:#fff4e6\n    style Input3 fill:#e6f3ff\n    style Input4 fill:#e6ffe6\n    style Output fill:#d4edda
"},{"location":"architecture/workflow-diagrams/#10-data-flow-event-streaming","title":"10. Data Flow & Event Streaming","text":"
flowchart TD\n    User[\ud83d\udc64 User] -->|Research Query| UI[Gradio UI]\n    UI -->|Submit| Manager[Magentic Manager]\n\n    Manager -->|Event: Planning| UI\n    Manager -->|Select Agent| HypAgent[Hypothesis Agent]\n    HypAgent -->|Event: Delta/Message| UI\n    HypAgent -->|Hypotheses| Context[(Shared Context)]\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| SearchAgent[Search Agent]\n    SearchAgent -->|MCP Request| WebSearch[Web Search Tool]\n    WebSearch -->|Results| SearchAgent\n    SearchAgent -->|Event: Delta/Message| UI\n    SearchAgent -->|Documents| Context\n    SearchAgent -->|Embeddings| VectorDB[(Vector DB)]\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| AnalysisAgent[Analysis Agent]\n    AnalysisAgent -->|MCP Request| CodeExec[Code Execution Tool]\n    CodeExec -->|Results| AnalysisAgent\n    AnalysisAgent -->|Event: Delta/Message| UI\n    AnalysisAgent -->|Analysis| Context\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| ReportAgent[Report Agent]\n    ReportAgent -->|MCP Request| CodeExec\n    ReportAgent -->|Event: Delta/Message| UI\n    ReportAgent -->|Report| Context\n\n    Manager -->|Event: Final Result| UI\n    UI -->|Display| User\n\n    style User fill:#e1f5e1\n    style UI fill:#e6f3ff\n    style Manager fill:#ffe6e6\n    style Context fill:#ffe6f0\n    style VectorDB fill:#ffe6f0\n    style WebSearch fill:#f0f0f0\n    style CodeExec fill:#f0f0f0
"},{"location":"architecture/workflow-diagrams/#11-mcp-tool-architecture","title":"11. MCP Tool Architecture","text":"
graph TB\n    subgraph \"Agent Layer\"\n        Manager[Magentic Manager]\n        HypAgent[Hypothesis Agent]\n        SearchAgent[Search Agent]\n        AnalysisAgent[Analysis Agent]\n        ReportAgent[Report Agent]\n    end\n\n    subgraph \"MCP Protocol Layer\"\n        Registry[MCP Tool Registry<br/>\u2022 Discovers tools<br/>\u2022 Routes requests<br/>\u2022 Manages connections]\n    end\n\n    subgraph \"MCP Servers\"\n        Server1[Web Search Server<br/>localhost:8001<br/>\u2022 PubMed<br/>\u2022 arXiv<br/>\u2022 bioRxiv]\n        Server2[Code Execution Server<br/>localhost:8002<br/>\u2022 Sandboxed Python<br/>\u2022 Package management]\n        Server3[RAG Server<br/>localhost:8003<br/>\u2022 Vector embeddings<br/>\u2022 Similarity search]\n        Server4[Visualization Server<br/>localhost:8004<br/>\u2022 Chart generation<br/>\u2022 Plot rendering]\n    end\n\n    subgraph \"External Services\"\n        PubMed[PubMed API]\n        ArXiv[arXiv API]\n        BioRxiv[bioRxiv API]\n        Modal[Modal Sandbox]\n        ChromaDB[(ChromaDB)]\n    end\n\n    SearchAgent -->|Request| Registry\n    AnalysisAgent -->|Request| Registry\n    ReportAgent -->|Request| Registry\n\n    Registry --> Server1\n    Registry --> Server2\n    Registry --> Server3\n    Registry --> Server4\n\n    Server1 --> PubMed\n    Server1 --> ArXiv\n    Server1 --> BioRxiv\n    Server2 --> Modal\n    Server3 --> ChromaDB\n\n    style Manager fill:#ffe6e6\n    style Registry fill:#fff4e6\n    style Server1 fill:#e6f3ff\n    style Server2 fill:#e6f3ff\n    style Server3 fill:#e6f3ff\n    style Server4 fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#12-progress-tracking-stall-detection","title":"12. Progress Tracking & Stall Detection","text":"
stateDiagram-v2\n    [*] --> Initialization: User Query\n\n    Initialization --> Planning: Manager starts\n\n    Planning --> AgentExecution: Select agent\n\n    AgentExecution --> Assessment: Collect results\n\n    Assessment --> QualityCheck: Evaluate output\n\n    QualityCheck --> AgentExecution: Poor quality<br/>(retry < max_rounds)\n    QualityCheck --> Planning: Poor quality<br/>(try different agent)\n    QualityCheck --> NextAgent: Good quality<br/>(task incomplete)\n    QualityCheck --> Synthesis: Good quality<br/>(task complete)\n\n    NextAgent --> AgentExecution: Select next agent\n\n    state StallDetection <<choice>>\n    Assessment --> StallDetection: Check progress\n    StallDetection --> Planning: No progress<br/>(stall count < max)\n    StallDetection --> ErrorRecovery: No progress<br/>(max stalls reached)\n\n    ErrorRecovery --> PartialReport: Generate partial results\n    PartialReport --> [*]\n\n    Synthesis --> FinalReport: Combine all outputs\n    FinalReport --> [*]\n\n    note right of QualityCheck\n        Manager assesses:\n        \u2022 Output completeness\n        \u2022 Quality metrics\n        \u2022 Progress made\n    end note\n\n    note right of StallDetection\n        Stall = no new progress\n        after agent execution\n        Triggers plan reset\n    end note
"},{"location":"architecture/workflow-diagrams/#13-gradio-ui-integration","title":"13. Gradio UI Integration","text":"
graph TD\n    App[Gradio App<br/>DeepCritical Research Agent]\n\n    App --> Input[Input Section]\n    App --> Status[Status Section]\n    App --> Output[Output Section]\n\n    Input --> Query[Research Question<br/>Text Area]\n    Input --> Controls[Controls]\n    Controls --> MaxHyp[Max Hypotheses: 1-10]\n    Controls --> MaxRounds[Max Rounds: 5-20]\n    Controls --> Submit[Start Research Button]\n\n    Status --> Log[Real-time Event Log<br/>\u2022 Manager planning<br/>\u2022 Agent selection<br/>\u2022 Execution updates<br/>\u2022 Quality assessment]\n    Status --> Progress[Progress Tracker<br/>\u2022 Current agent<br/>\u2022 Round count<br/>\u2022 Stall count]\n\n    Output --> Tabs[Tabbed Results]\n    Tabs --> Tab1[Hypotheses Tab<br/>Generated hypotheses with scores]\n    Tabs --> Tab2[Search Results Tab<br/>Papers & sources found]\n    Tabs --> Tab3[Analysis Tab<br/>Evidence & verdicts]\n    Tabs --> Tab4[Report Tab<br/>Final research report]\n    Tab4 --> Download[Download Report<br/>MD / PDF / JSON]\n\n    Submit -.->|Triggers| Workflow[Magentic Workflow]\n    Workflow -.->|MagenticOrchestratorMessageEvent| Log\n    Workflow -.->|MagenticAgentDeltaEvent| Log\n    Workflow -.->|MagenticAgentMessageEvent| Log\n    Workflow -.->|MagenticFinalResultEvent| Tab4\n\n    style App fill:#e1f5e1\n    style Input fill:#fff4e6\n    style Status fill:#e6f3ff\n    style Output fill:#e6ffe6\n    style Workflow fill:#ffe6e6
"},{"location":"architecture/workflow-diagrams/#14-complete-system-context","title":"14. Complete System Context","text":"
graph LR\n    User[\ud83d\udc64 Researcher<br/>Asks research questions] -->|Submits query| DC[DeepCritical<br/>Magentic Workflow]\n\n    DC -->|Literature search| PubMed[PubMed API<br/>Medical papers]\n    DC -->|Preprint search| ArXiv[arXiv API<br/>Scientific preprints]\n    DC -->|Biology search| BioRxiv[bioRxiv API<br/>Biology preprints]\n    DC -->|Agent reasoning| Claude[Claude API<br/>Sonnet 4 / Opus]\n    DC -->|Code execution| Modal[Modal Sandbox<br/>Safe Python env]\n    DC -->|Vector storage| Chroma[ChromaDB<br/>Embeddings & RAG]\n\n    DC -->|Deployed on| HF[HuggingFace Spaces<br/>Gradio 6.0]\n\n    PubMed -->|Results| DC\n    ArXiv -->|Results| DC\n    BioRxiv -->|Results| DC\n    Claude -->|Responses| DC\n    Modal -->|Output| DC\n    Chroma -->|Context| DC\n\n    DC -->|Research report| User\n\n    style User fill:#e1f5e1\n    style DC fill:#ffe6e6\n    style PubMed fill:#e6f3ff\n    style ArXiv fill:#e6f3ff\n    style BioRxiv fill:#e6f3ff\n    style Claude fill:#ffd6d6\n    style Modal fill:#f0f0f0\n    style Chroma fill:#ffe6f0\n    style HF fill:#d4edda
"},{"location":"architecture/workflow-diagrams/#15-workflow-timeline-simplified","title":"15. Workflow Timeline (Simplified)","text":"
gantt\n    title DeepCritical Magentic Workflow - Typical Execution\n    dateFormat mm:ss\n    axisFormat %M:%S\n\n    section Manager Planning\n    Initial planning         :p1, 00:00, 10s\n\n    section Hypothesis Agent\n    Generate hypotheses      :h1, after p1, 30s\n    Manager assessment       :h2, after h1, 5s\n\n    section Search Agent\n    Search hypothesis 1      :s1, after h2, 20s\n    Search hypothesis 2      :s2, after s1, 20s\n    Search hypothesis 3      :s3, after s2, 20s\n    RAG processing          :s4, after s3, 15s\n    Manager assessment      :s5, after s4, 5s\n\n    section Analysis Agent\n    Evidence extraction     :a1, after s5, 15s\n    Code generation        :a2, after a1, 20s\n    Code execution         :a3, after a2, 25s\n    Synthesis              :a4, after a3, 20s\n    Manager assessment     :a5, after a4, 5s\n\n    section Report Agent\n    Report assembly        :r1, after a5, 30s\n    Visualization          :r2, after r1, 15s\n    Formatting             :r3, after r2, 10s\n\n    section Manager Synthesis\n    Final synthesis        :f1, after r3, 10s
"},{"location":"architecture/workflow-diagrams/#key-differences-from-original-design","title":"Key Differences from Original Design","text":"Aspect Original (Judge-in-Loop) New (Magentic) Control Flow Fixed sequential phases Dynamic agent selection Quality Control Separate Judge Agent Manager assessment built-in Retry Logic Phase-level with feedback Agent-level with adaptation Flexibility Rigid 4-phase pipeline Adaptive workflow Complexity 5 agents (including Judge) 4 agents (no Judge) Progress Tracking Manual state management Built-in round/stall detection Agent Coordination Sequential handoff Manager-driven dynamic selection Error Recovery Retry same phase Try different agent or replan"},{"location":"architecture/workflow-diagrams/#simplified-design-principles","title":"Simplified Design Principles","text":"
  1. Manager is Intelligent: LLM-powered manager handles planning, selection, and quality assessment
  2. No Separate Judge: Manager's assessment phase replaces dedicated Judge Agent
  3. Dynamic Workflow: Agents can be called multiple times in any order based on need
  4. Built-in Safety: max_round_count (15) and max_stall_count (3) prevent infinite loops
  5. Event-Driven UI: Real-time streaming updates to Gradio interface
  6. MCP-Powered Tools: All external capabilities via Model Context Protocol
  7. Shared Context: Centralized state accessible to all agents
  8. Progress Awareness: Manager tracks what's been done and what's needed
"},{"location":"architecture/workflow-diagrams/#legend","title":"Legend","text":""},{"location":"architecture/workflow-diagrams/#implementation-highlights","title":"Implementation Highlights","text":"

Simple 4-Agent Setup: 

workflow = (\n    MagenticBuilder()\n    .participants(\n        hypothesis=HypothesisAgent(tools=[background_tool]),\n        search=SearchAgent(tools=[web_search, rag_tool]),\n        analysis=AnalysisAgent(tools=[code_execution]),\n        report=ReportAgent(tools=[code_execution, visualization])\n    )\n    .with_standard_manager(\n        chat_client=AnthropicClient(model=\"claude-sonnet-4\"),\n        max_round_count=15,    # Prevent infinite loops\n        max_stall_count=3      # Detect stuck workflows\n    )\n    .build()\n)\n

Manager handles quality assessment in its instructions: - Checks hypothesis quality (testable, novel, clear) - Validates search results (relevant, authoritative, recent) - Assesses analysis soundness (methodology, evidence, conclusions) - Ensures report completeness (all sections, proper citations)

No separate Judge Agent needed - manager does it all!

Document Version: 2.0 (Magentic Simplified) Last Updated: 2025-11-24 Architecture: Microsoft Magentic Orchestration Pattern Agents: 4 (Hypothesis, Search, Analysis, Report) + 1 Manager License: MIT

"},{"location":"architecture/workflow-diagrams/#see-also","title":"See Also","text":""},{"location":"architecture/workflows/","title":"DeepCritical Workflow - Simplified Magentic Architecture","text":"

Architecture Pattern: Microsoft Magentic Orchestration Design Philosophy: Simple, dynamic, manager-driven coordination Key Innovation: Intelligent manager replaces rigid sequential phases

"},{"location":"architecture/workflows/#1-high-level-magentic-workflow","title":"1. High-Level Magentic Workflow","text":"
flowchart TD\n    Start([User Query]) --> Manager[Magentic Manager<br/>Plan \u2022 Select \u2022 Assess \u2022 Adapt]\n\n    Manager -->|Plans| Task1[Task Decomposition]\n    Task1 --> Manager\n\n    Manager -->|Selects & Executes| HypAgent[Hypothesis Agent]\n    Manager -->|Selects & Executes| SearchAgent[Search Agent]\n    Manager -->|Selects & Executes| AnalysisAgent[Analysis Agent]\n    Manager -->|Selects & Executes| ReportAgent[Report Agent]\n\n    HypAgent -->|Results| Manager\n    SearchAgent -->|Results| Manager\n    AnalysisAgent -->|Results| Manager\n    ReportAgent -->|Results| Manager\n\n    Manager -->|Assesses Quality| Decision{Good Enough?}\n    Decision -->|No - Refine| Manager\n    Decision -->|No - Different Agent| Manager\n    Decision -->|No - Stalled| Replan[Reset Plan]\n    Replan --> Manager\n\n    Decision -->|Yes| Synthesis[Synthesize Final Result]\n    Synthesis --> Output([Research Report])\n\n    style Start fill:#e1f5e1\n    style Manager fill:#ffe6e6\n    style HypAgent fill:#fff4e6\n    style SearchAgent fill:#fff4e6\n    style AnalysisAgent fill:#fff4e6\n    style ReportAgent fill:#fff4e6\n    style Decision fill:#ffd6d6\n    style Synthesis fill:#d4edda\n    style Output fill:#e1f5e1
"},{"location":"architecture/workflows/#2-magentic-manager-the-6-phase-cycle","title":"2. Magentic Manager: The 6-Phase Cycle","text":"
flowchart LR\n    P1[1. Planning<br/>Analyze task<br/>Create strategy] --> P2[2. Agent Selection<br/>Pick best agent<br/>for subtask]\n    P2 --> P3[3. Execution<br/>Run selected<br/>agent with tools]\n    P3 --> P4[4. Assessment<br/>Evaluate quality<br/>Check progress]\n    P4 --> Decision{Quality OK?<br/>Progress made?}\n    Decision -->|Yes| P6[6. Synthesis<br/>Combine results<br/>Generate report]\n    Decision -->|No| P5[5. Iteration<br/>Adjust plan<br/>Try again]\n    P5 --> P2\n    P6 --> Done([Complete])\n\n    style P1 fill:#fff4e6\n    style P2 fill:#ffe6e6\n    style P3 fill:#e6f3ff\n    style P4 fill:#ffd6d6\n    style P5 fill:#fff3cd\n    style P6 fill:#d4edda\n    style Done fill:#e1f5e1
"},{"location":"architecture/workflows/#3-simplified-agent-architecture","title":"3. Simplified Agent Architecture","text":"
graph TB\n    subgraph \"Orchestration Layer\"\n        Manager[Magentic Manager<br/>\u2022 Plans workflow<br/>\u2022 Selects agents<br/>\u2022 Assesses quality<br/>\u2022 Adapts strategy]\n        SharedContext[(Shared Context<br/>\u2022 Hypotheses<br/>\u2022 Search Results<br/>\u2022 Analysis<br/>\u2022 Progress)]\n        Manager <--> SharedContext\n    end\n\n    subgraph \"Specialist Agents\"\n        HypAgent[Hypothesis Agent<br/>\u2022 Domain understanding<br/>\u2022 Hypothesis generation<br/>\u2022 Testability refinement]\n        SearchAgent[Search Agent<br/>\u2022 Multi-source search<br/>\u2022 RAG retrieval<br/>\u2022 Result ranking]\n        AnalysisAgent[Analysis Agent<br/>\u2022 Evidence extraction<br/>\u2022 Statistical analysis<br/>\u2022 Code execution]\n        ReportAgent[Report Agent<br/>\u2022 Report assembly<br/>\u2022 Visualization<br/>\u2022 Citation formatting]\n    end\n\n    subgraph \"MCP Tools\"\n        WebSearch[Web Search<br/>PubMed \u2022 arXiv \u2022 bioRxiv]\n        CodeExec[Code Execution<br/>Sandboxed Python]\n        RAG[RAG Retrieval<br/>Vector DB \u2022 Embeddings]\n        Viz[Visualization<br/>Charts \u2022 Graphs]\n    end\n\n    Manager -->|Selects & Directs| HypAgent\n    Manager -->|Selects & Directs| SearchAgent\n    Manager -->|Selects & Directs| AnalysisAgent\n    Manager -->|Selects & Directs| ReportAgent\n\n    HypAgent --> SharedContext\n    SearchAgent --> SharedContext\n    AnalysisAgent --> SharedContext\n    ReportAgent --> SharedContext\n\n    SearchAgent --> WebSearch\n    SearchAgent --> RAG\n    AnalysisAgent --> CodeExec\n    ReportAgent --> CodeExec\n    ReportAgent --> Viz\n\n    style Manager fill:#ffe6e6\n    style SharedContext fill:#ffe6f0\n    style HypAgent fill:#fff4e6\n    style SearchAgent fill:#fff4e6\n    style AnalysisAgent fill:#fff4e6\n    style ReportAgent fill:#fff4e6\n    style WebSearch fill:#e6f3ff\n    style CodeExec fill:#e6f3ff\n    style RAG fill:#e6f3ff\n    style Viz fill:#e6f3ff
"},{"location":"architecture/workflows/#4-dynamic-workflow-example","title":"4. Dynamic Workflow Example","text":"
sequenceDiagram\n    participant User\n    participant Manager\n    participant HypAgent\n    participant SearchAgent\n    participant AnalysisAgent\n    participant ReportAgent\n\n    User->>Manager: \"Research protein folding in Alzheimer's\"\n\n    Note over Manager: PLAN: Generate hypotheses \u2192 Search \u2192 Analyze \u2192 Report\n\n    Manager->>HypAgent: Generate 3 hypotheses\n    HypAgent-->>Manager: Returns 3 hypotheses\n    Note over Manager: ASSESS: Good quality, proceed\n\n    Manager->>SearchAgent: Search literature for hypothesis 1\n    SearchAgent-->>Manager: Returns 15 papers\n    Note over Manager: ASSESS: Good results, continue\n\n    Manager->>SearchAgent: Search for hypothesis 2\n    SearchAgent-->>Manager: Only 2 papers found\n    Note over Manager: ASSESS: Insufficient, refine search\n\n    Manager->>SearchAgent: Refined query for hypothesis 2\n    SearchAgent-->>Manager: Returns 12 papers\n    Note over Manager: ASSESS: Better, proceed\n\n    Manager->>AnalysisAgent: Analyze evidence for all hypotheses\n    AnalysisAgent-->>Manager: Returns analysis with code\n    Note over Manager: ASSESS: Complete, generate report\n\n    Manager->>ReportAgent: Create comprehensive report\n    ReportAgent-->>Manager: Returns formatted report\n    Note over Manager: SYNTHESIZE: Combine all results\n\n    Manager->>User: Final Research Report
"},{"location":"architecture/workflows/#5-manager-decision-logic","title":"5. Manager Decision Logic","text":"
flowchart TD\n    Start([Manager Receives Task]) --> Plan[Create Initial Plan]\n\n    Plan --> Select[Select Agent for Next Subtask]\n    Select --> Execute[Execute Agent]\n    Execute --> Collect[Collect Results]\n\n    Collect --> Assess[Assess Quality & Progress]\n\n    Assess --> Q1{Quality Sufficient?}\n    Q1 -->|No| Q2{Same Agent Can Fix?}\n    Q2 -->|Yes| Feedback[Provide Specific Feedback]\n    Feedback --> Execute\n    Q2 -->|No| Different[Try Different Agent]\n    Different --> Select\n\n    Q1 -->|Yes| Q3{Task Complete?}\n    Q3 -->|No| Q4{Making Progress?}\n    Q4 -->|Yes| Select\n    Q4 -->|No - Stalled| Replan[Reset Plan & Approach]\n    Replan --> Plan\n\n    Q3 -->|Yes| Synth[Synthesize Final Result]\n    Synth --> Done([Return Report])\n\n    style Start fill:#e1f5e1\n    style Plan fill:#fff4e6\n    style Select fill:#ffe6e6\n    style Execute fill:#e6f3ff\n    style Assess fill:#ffd6d6\n    style Q1 fill:#ffe6e6\n    style Q2 fill:#ffe6e6\n    style Q3 fill:#ffe6e6\n    style Q4 fill:#ffe6e6\n    style Synth fill:#d4edda\n    style Done fill:#e1f5e1
"},{"location":"architecture/workflows/#6-hypothesis-agent-workflow","title":"6. Hypothesis Agent Workflow","text":"
flowchart LR\n    Input[Research Query] --> Domain[Identify Domain<br/>& Key Concepts]\n    Domain --> Context[Retrieve Background<br/>Knowledge]\n    Context --> Generate[Generate 3-5<br/>Initial Hypotheses]\n    Generate --> Refine[Refine for<br/>Testability]\n    Refine --> Rank[Rank by<br/>Quality Score]\n    Rank --> Output[Return Top<br/>Hypotheses]\n\n    Output --> Struct[Hypothesis Structure:<br/>\u2022 Statement<br/>\u2022 Rationale<br/>\u2022 Testability Score<br/>\u2022 Data Requirements<br/>\u2022 Expected Outcomes]\n\n    style Input fill:#e1f5e1\n    style Output fill:#fff4e6\n    style Struct fill:#e6f3ff
"},{"location":"architecture/workflows/#7-search-agent-workflow","title":"7. Search Agent Workflow","text":"
flowchart TD\n    Input[Hypotheses] --> Strategy[Formulate Search<br/>Strategy per Hypothesis]\n\n    Strategy --> Multi[Multi-Source Search]\n\n    Multi --> PubMed[PubMed Search<br/>via MCP]\n    Multi --> ArXiv[arXiv Search<br/>via MCP]\n    Multi --> BioRxiv[bioRxiv Search<br/>via MCP]\n\n    PubMed --> Aggregate[Aggregate Results]\n    ArXiv --> Aggregate\n    BioRxiv --> Aggregate\n\n    Aggregate --> Filter[Filter & Rank<br/>by Relevance]\n    Filter --> Dedup[Deduplicate<br/>Cross-Reference]\n    Dedup --> Embed[Embed Documents<br/>via MCP]\n    Embed --> Vector[(Vector DB)]\n    Vector --> RAGRetrieval[RAG Retrieval<br/>Top-K per Hypothesis]\n    RAGRetrieval --> Output[Return Contextualized<br/>Search Results]\n\n    style Input fill:#fff4e6\n    style Multi fill:#ffe6e6\n    style Vector fill:#ffe6f0\n    style Output fill:#e6f3ff
"},{"location":"architecture/workflows/#8-analysis-agent-workflow","title":"8. Analysis Agent Workflow","text":"
flowchart TD\n    Input1[Hypotheses] --> Extract\n    Input2[Search Results] --> Extract[Extract Evidence<br/>per Hypothesis]\n\n    Extract --> Methods[Determine Analysis<br/>Methods Needed]\n\n    Methods --> Branch{Requires<br/>Computation?}\n    Branch -->|Yes| GenCode[Generate Python<br/>Analysis Code]\n    Branch -->|No| Qual[Qualitative<br/>Synthesis]\n\n    GenCode --> Execute[Execute Code<br/>via MCP Sandbox]\n    Execute --> Interpret1[Interpret<br/>Results]\n    Qual --> Interpret2[Interpret<br/>Findings]\n\n    Interpret1 --> Synthesize[Synthesize Evidence<br/>Across Sources]\n    Interpret2 --> Synthesize\n\n    Synthesize --> Verdict[Determine Verdict<br/>per Hypothesis]\n    Verdict --> Support[\u2022 Supported<br/>\u2022 Refuted<br/>\u2022 Inconclusive]\n    Support --> Gaps[Identify Knowledge<br/>Gaps & Limitations]\n    Gaps --> Output[Return Analysis<br/>Report]\n\n    style Input1 fill:#fff4e6\n    style Input2 fill:#e6f3ff\n    style Execute fill:#ffe6e6\n    style Output fill:#e6ffe6
"},{"location":"architecture/workflows/#9-report-agent-workflow","title":"9. Report Agent Workflow","text":"
flowchart TD\n    Input1[Query] --> Assemble\n    Input2[Hypotheses] --> Assemble\n    Input3[Search Results] --> Assemble\n    Input4[Analysis] --> Assemble[Assemble Report<br/>Sections]\n\n    Assemble --> Exec[Executive Summary]\n    Assemble --> Intro[Introduction]\n    Assemble --> Methods[Methods]\n    Assemble --> Results[Results per<br/>Hypothesis]\n    Assemble --> Discussion[Discussion]\n    Assemble --> Future[Future Directions]\n    Assemble --> Refs[References]\n\n    Results --> VizCheck{Needs<br/>Visualization?}\n    VizCheck -->|Yes| GenViz[Generate Viz Code]\n    GenViz --> ExecViz[Execute via MCP<br/>Create Charts]\n    ExecViz --> Combine\n    VizCheck -->|No| Combine[Combine All<br/>Sections]\n\n    Exec --> Combine\n    Intro --> Combine\n    Methods --> Combine\n    Discussion --> Combine\n    Future --> Combine\n    Refs --> Combine\n\n    Combine --> Format[Format Output]\n    Format --> MD[Markdown]\n    Format --> PDF[PDF]\n    Format --> JSON[JSON]\n\n    MD --> Output[Return Final<br/>Report]\n    PDF --> Output\n    JSON --> Output\n\n    style Input1 fill:#e1f5e1\n    style Input2 fill:#fff4e6\n    style Input3 fill:#e6f3ff\n    style Input4 fill:#e6ffe6\n    style Output fill:#d4edda
"},{"location":"architecture/workflows/#10-data-flow-event-streaming","title":"10. Data Flow & Event Streaming","text":"
flowchart TD\n    User[\ud83d\udc64 User] -->|Research Query| UI[Gradio UI]\n    UI -->|Submit| Manager[Magentic Manager]\n\n    Manager -->|Event: Planning| UI\n    Manager -->|Select Agent| HypAgent[Hypothesis Agent]\n    HypAgent -->|Event: Delta/Message| UI\n    HypAgent -->|Hypotheses| Context[(Shared Context)]\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| SearchAgent[Search Agent]\n    SearchAgent -->|MCP Request| WebSearch[Web Search Tool]\n    WebSearch -->|Results| SearchAgent\n    SearchAgent -->|Event: Delta/Message| UI\n    SearchAgent -->|Documents| Context\n    SearchAgent -->|Embeddings| VectorDB[(Vector DB)]\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| AnalysisAgent[Analysis Agent]\n    AnalysisAgent -->|MCP Request| CodeExec[Code Execution Tool]\n    CodeExec -->|Results| AnalysisAgent\n    AnalysisAgent -->|Event: Delta/Message| UI\n    AnalysisAgent -->|Analysis| Context\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| ReportAgent[Report Agent]\n    ReportAgent -->|MCP Request| CodeExec\n    ReportAgent -->|Event: Delta/Message| UI\n    ReportAgent -->|Report| Context\n\n    Manager -->|Event: Final Result| UI\n    UI -->|Display| User\n\n    style User fill:#e1f5e1\n    style UI fill:#e6f3ff\n    style Manager fill:#ffe6e6\n    style Context fill:#ffe6f0\n    style VectorDB fill:#ffe6f0\n    style WebSearch fill:#f0f0f0\n    style CodeExec fill:#f0f0f0
"},{"location":"architecture/workflows/#11-mcp-tool-architecture","title":"11. MCP Tool Architecture","text":"
graph TB\n    subgraph \"Agent Layer\"\n        Manager[Magentic Manager]\n        HypAgent[Hypothesis Agent]\n        SearchAgent[Search Agent]\n        AnalysisAgent[Analysis Agent]\n        ReportAgent[Report Agent]\n    end\n\n    subgraph \"MCP Protocol Layer\"\n        Registry[MCP Tool Registry<br/>\u2022 Discovers tools<br/>\u2022 Routes requests<br/>\u2022 Manages connections]\n    end\n\n    subgraph \"MCP Servers\"\n        Server1[Web Search Server<br/>localhost:8001<br/>\u2022 PubMed<br/>\u2022 arXiv<br/>\u2022 bioRxiv]\n        Server2[Code Execution Server<br/>localhost:8002<br/>\u2022 Sandboxed Python<br/>\u2022 Package management]\n        Server3[RAG Server<br/>localhost:8003<br/>\u2022 Vector embeddings<br/>\u2022 Similarity search]\n        Server4[Visualization Server<br/>localhost:8004<br/>\u2022 Chart generation<br/>\u2022 Plot rendering]\n    end\n\n    subgraph \"External Services\"\n        PubMed[PubMed API]\n        ArXiv[arXiv API]\n        BioRxiv[bioRxiv API]\n        Modal[Modal Sandbox]\n        ChromaDB[(ChromaDB)]\n    end\n\n    SearchAgent -->|Request| Registry\n    AnalysisAgent -->|Request| Registry\n    ReportAgent -->|Request| Registry\n\n    Registry --> Server1\n    Registry --> Server2\n    Registry --> Server3\n    Registry --> Server4\n\n    Server1 --> PubMed\n    Server1 --> ArXiv\n    Server1 --> BioRxiv\n    Server2 --> Modal\n    Server3 --> ChromaDB\n\n    style Manager fill:#ffe6e6\n    style Registry fill:#fff4e6\n    style Server1 fill:#e6f3ff\n    style Server2 fill:#e6f3ff\n    style Server3 fill:#e6f3ff\n    style Server4 fill:#e6f3ff
"},{"location":"architecture/workflows/#12-progress-tracking-stall-detection","title":"12. Progress Tracking & Stall Detection","text":"
stateDiagram-v2\n    [*] --> Initialization: User Query\n\n    Initialization --> Planning: Manager starts\n\n    Planning --> AgentExecution: Select agent\n\n    AgentExecution --> Assessment: Collect results\n\n    Assessment --> QualityCheck: Evaluate output\n\n    QualityCheck --> AgentExecution: Poor quality<br/>(retry < max_rounds)\n    QualityCheck --> Planning: Poor quality<br/>(try different agent)\n    QualityCheck --> NextAgent: Good quality<br/>(task incomplete)\n    QualityCheck --> Synthesis: Good quality<br/>(task complete)\n\n    NextAgent --> AgentExecution: Select next agent\n\n    state StallDetection <<choice>>\n    Assessment --> StallDetection: Check progress\n    StallDetection --> Planning: No progress<br/>(stall count < max)\n    StallDetection --> ErrorRecovery: No progress<br/>(max stalls reached)\n\n    ErrorRecovery --> PartialReport: Generate partial results\n    PartialReport --> [*]\n\n    Synthesis --> FinalReport: Combine all outputs\n    FinalReport --> [*]\n\n    note right of QualityCheck\n        Manager assesses:\n        \u2022 Output completeness\n        \u2022 Quality metrics\n        \u2022 Progress made\n    end note\n\n    note right of StallDetection\n        Stall = no new progress\n        after agent execution\n        Triggers plan reset\n    end note
"},{"location":"architecture/workflows/#13-gradio-ui-integration","title":"13. Gradio UI Integration","text":"
graph TD\n    App[Gradio App<br/>DeepCritical Research Agent]\n\n    App --> Input[Input Section]\n    App --> Status[Status Section]\n    App --> Output[Output Section]\n\n    Input --> Query[Research Question<br/>Text Area]\n    Input --> Controls[Controls]\n    Controls --> MaxHyp[Max Hypotheses: 1-10]\n    Controls --> MaxRounds[Max Rounds: 5-20]\n    Controls --> Submit[Start Research Button]\n\n    Status --> Log[Real-time Event Log<br/>\u2022 Manager planning<br/>\u2022 Agent selection<br/>\u2022 Execution updates<br/>\u2022 Quality assessment]\n    Status --> Progress[Progress Tracker<br/>\u2022 Current agent<br/>\u2022 Round count<br/>\u2022 Stall count]\n\n    Output --> Tabs[Tabbed Results]\n    Tabs --> Tab1[Hypotheses Tab<br/>Generated hypotheses with scores]\n    Tabs --> Tab2[Search Results Tab<br/>Papers & sources found]\n    Tabs --> Tab3[Analysis Tab<br/>Evidence & verdicts]\n    Tabs --> Tab4[Report Tab<br/>Final research report]\n    Tab4 --> Download[Download Report<br/>MD / PDF / JSON]\n\n    Submit -.->|Triggers| Workflow[Magentic Workflow]\n    Workflow -.->|MagenticOrchestratorMessageEvent| Log\n    Workflow -.->|MagenticAgentDeltaEvent| Log\n    Workflow -.->|MagenticAgentMessageEvent| Log\n    Workflow -.->|MagenticFinalResultEvent| Tab4\n\n    style App fill:#e1f5e1\n    style Input fill:#fff4e6\n    style Status fill:#e6f3ff\n    style Output fill:#e6ffe6\n    style Workflow fill:#ffe6e6
"},{"location":"architecture/workflows/#14-complete-system-context","title":"14. Complete System Context","text":"
graph LR\n    User[\ud83d\udc64 Researcher<br/>Asks research questions] -->|Submits query| DC[DeepCritical<br/>Magentic Workflow]\n\n    DC -->|Literature search| PubMed[PubMed API<br/>Medical papers]\n    DC -->|Preprint search| ArXiv[arXiv API<br/>Scientific preprints]\n    DC -->|Biology search| BioRxiv[bioRxiv API<br/>Biology preprints]\n    DC -->|Agent reasoning| Claude[Claude API<br/>Sonnet 4 / Opus]\n    DC -->|Code execution| Modal[Modal Sandbox<br/>Safe Python env]\n    DC -->|Vector storage| Chroma[ChromaDB<br/>Embeddings & RAG]\n\n    DC -->|Deployed on| HF[HuggingFace Spaces<br/>Gradio 6.0]\n\n    PubMed -->|Results| DC\n    ArXiv -->|Results| DC\n    BioRxiv -->|Results| DC\n    Claude -->|Responses| DC\n    Modal -->|Output| DC\n    Chroma -->|Context| DC\n\n    DC -->|Research report| User\n\n    style User fill:#e1f5e1\n    style DC fill:#ffe6e6\n    style PubMed fill:#e6f3ff\n    style ArXiv fill:#e6f3ff\n    style BioRxiv fill:#e6f3ff\n    style Claude fill:#ffd6d6\n    style Modal fill:#f0f0f0\n    style Chroma fill:#ffe6f0\n    style HF fill:#d4edda
"},{"location":"architecture/workflows/#15-workflow-timeline-simplified","title":"15. Workflow Timeline (Simplified)","text":"
gantt\n    title DeepCritical Magentic Workflow - Typical Execution\n    dateFormat mm:ss\n    axisFormat %M:%S\n\n    section Manager Planning\n    Initial planning         :p1, 00:00, 10s\n\n    section Hypothesis Agent\n    Generate hypotheses      :h1, after p1, 30s\n    Manager assessment       :h2, after h1, 5s\n\n    section Search Agent\n    Search hypothesis 1      :s1, after h2, 20s\n    Search hypothesis 2      :s2, after s1, 20s\n    Search hypothesis 3      :s3, after s2, 20s\n    RAG processing          :s4, after s3, 15s\n    Manager assessment      :s5, after s4, 5s\n\n    section Analysis Agent\n    Evidence extraction     :a1, after s5, 15s\n    Code generation        :a2, after a1, 20s\n    Code execution         :a3, after a2, 25s\n    Synthesis              :a4, after a3, 20s\n    Manager assessment     :a5, after a4, 5s\n\n    section Report Agent\n    Report assembly        :r1, after a5, 30s\n    Visualization          :r2, after r1, 15s\n    Formatting             :r3, after r2, 10s\n\n    section Manager Synthesis\n    Final synthesis        :f1, after r3, 10s
"},{"location":"architecture/workflows/#key-differences-from-original-design","title":"Key Differences from Original Design","text":"Aspect Original (Judge-in-Loop) New (Magentic) Control Flow Fixed sequential phases Dynamic agent selection Quality Control Separate Judge Agent Manager assessment built-in Retry Logic Phase-level with feedback Agent-level with adaptation Flexibility Rigid 4-phase pipeline Adaptive workflow Complexity 5 agents (including Judge) 4 agents (no Judge) Progress Tracking Manual state management Built-in round/stall detection Agent Coordination Sequential handoff Manager-driven dynamic selection Error Recovery Retry same phase Try different agent or replan"},{"location":"architecture/workflows/#simplified-design-principles","title":"Simplified Design Principles","text":"
  1. Manager is Intelligent: LLM-powered manager handles planning, selection, and quality assessment
  2. No Separate Judge: Manager's assessment phase replaces dedicated Judge Agent
  3. Dynamic Workflow: Agents can be called multiple times in any order based on need
  4. Built-in Safety: max_round_count (15) and max_stall_count (3) prevent infinite loops
  5. Event-Driven UI: Real-time streaming updates to Gradio interface
  6. MCP-Powered Tools: All external capabilities via Model Context Protocol
  7. Shared Context: Centralized state accessible to all agents
  8. Progress Awareness: Manager tracks what's been done and what's needed
"},{"location":"architecture/workflows/#legend","title":"Legend","text":""},{"location":"architecture/workflows/#implementation-highlights","title":"Implementation Highlights","text":"

Simple 4-Agent Setup: 

workflow = (\n    MagenticBuilder()\n    .participants(\n        hypothesis=HypothesisAgent(tools=[background_tool]),\n        search=SearchAgent(tools=[web_search, rag_tool]),\n        analysis=AnalysisAgent(tools=[code_execution]),\n        report=ReportAgent(tools=[code_execution, visualization])\n    )\n    .with_standard_manager(\n        chat_client=AnthropicClient(model=\"claude-sonnet-4\"),\n        max_round_count=15,    # Prevent infinite loops\n        max_stall_count=3      # Detect stuck workflows\n    )\n    .build()\n)\n

Manager handles quality assessment in its instructions: - Checks hypothesis quality (testable, novel, clear) - Validates search results (relevant, authoritative, recent) - Assesses analysis soundness (methodology, evidence, conclusions) - Ensures report completeness (all sections, proper citations)

No separate Judge Agent needed - manager does it all!

Document Version: 2.0 (Magentic Simplified) Last Updated: 2025-11-24 Architecture: Microsoft Magentic Orchestration Pattern Agents: 4 (Hypothesis, Search, Analysis, Report) + 1 Manager License: MIT

"},{"location":"configuration/","title":"Configuration Guide","text":""},{"location":"configuration/#overview","title":"Overview","text":"

DeepCritical uses Pydantic Settings for centralized configuration management. All settings are defined in the Settings class in src/utils/config.py and can be configured via environment variables or a .env file.

The configuration system provides:

"},{"location":"configuration/#quick-start","title":"Quick Start","text":"
  1. Create a .env file in the project root
  2. Set at least one LLM API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, or HF_TOKEN)
  3. Optionally configure other services as needed
  4. The application will automatically load and validate your configuration
"},{"location":"configuration/#configuration-system-architecture","title":"Configuration System Architecture","text":""},{"location":"configuration/#settings-class","title":"Settings Class","text":"

The [Settings][settings-class] class extends BaseSettings from pydantic_settings and defines all application configuration:

View source

"},{"location":"configuration/#singleton-instance","title":"Singleton Instance","text":"

A global settings instance is available for import:

View source

"},{"location":"configuration/#usage-pattern","title":"Usage Pattern","text":"

Access configuration throughout the codebase:

from src.utils.config import settings\n\n# Check if API keys are available\nif settings.has_openai_key:\n    # Use OpenAI\n    pass\n\n# Access configuration values\nmax_iterations = settings.max_iterations\nweb_search_provider = settings.web_search_provider\n
"},{"location":"configuration/#required-configuration","title":"Required Configuration","text":""},{"location":"configuration/#llm-provider","title":"LLM Provider","text":"

You must configure at least one LLM provider. The system supports:

"},{"location":"configuration/#openai-configuration","title":"OpenAI Configuration","text":"
LLM_PROVIDER=openai\nOPENAI_API_KEY=your_openai_api_key_here\nOPENAI_MODEL=gpt-5.1\n

The default model is defined in the Settings class:

"},{"location":"configuration/#anthropic-configuration","title":"Anthropic Configuration","text":"
LLM_PROVIDER=anthropic\nANTHROPIC_API_KEY=your_anthropic_api_key_here\nANTHROPIC_MODEL=claude-sonnet-4-5-20250929\n

The default model is defined in the Settings class:

"},{"location":"configuration/#huggingface-configuration","title":"HuggingFace Configuration","text":"

HuggingFace can work without an API key for public models, but an API key provides higher rate limits:

# Option 1: Using HF_TOKEN (preferred)\nHF_TOKEN=your_huggingface_token_here\n\n# Option 2: Using HUGGINGFACE_API_KEY (alternative)\nHUGGINGFACE_API_KEY=your_huggingface_api_key_here\n\n# Default model\nHUGGINGFACE_MODEL=meta-llama/Llama-3.1-8B-Instruct\n

The HuggingFace token can be set via either environment variable:

"},{"location":"configuration/#optional-configuration","title":"Optional Configuration","text":""},{"location":"configuration/#embedding-configuration","title":"Embedding Configuration","text":"

DeepCritical supports multiple embedding providers for semantic search and RAG:

# Embedding Provider: \"openai\", \"local\", or \"huggingface\"\nEMBEDDING_PROVIDER=local\n\n# OpenAI Embedding Model (used by LlamaIndex RAG)\nOPENAI_EMBEDDING_MODEL=text-embedding-3-small\n\n# Local Embedding Model (sentence-transformers, used by EmbeddingService)\nLOCAL_EMBEDDING_MODEL=all-MiniLM-L6-v2\n\n# HuggingFace Embedding Model\nHUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2\n

The embedding provider configuration:

Note: OpenAI embeddings require OPENAI_API_KEY. The local provider (default) uses sentence-transformers and requires no API key.

"},{"location":"configuration/#web-search-configuration","title":"Web Search Configuration","text":"

DeepCritical supports multiple web search providers:

# Web Search Provider: \"serper\", \"searchxng\", \"brave\", \"tavily\", or \"duckduckgo\"\n# Default: \"duckduckgo\" (no API key required)\nWEB_SEARCH_PROVIDER=duckduckgo\n\n# Serper API Key (for Google search via Serper)\nSERPER_API_KEY=your_serper_api_key_here\n\n# SearchXNG Host URL (for self-hosted search)\nSEARCHXNG_HOST=http://localhost:8080\n\n# Brave Search API Key\nBRAVE_API_KEY=your_brave_api_key_here\n\n# Tavily API Key\nTAVILY_API_KEY=your_tavily_api_key_here\n

The web search provider configuration:

Note: DuckDuckGo is the default and requires no API key, making it ideal for development and testing.

"},{"location":"configuration/#pubmed-configuration","title":"PubMed Configuration","text":"

PubMed search supports optional NCBI API key for higher rate limits:

# NCBI API Key (optional, for higher rate limits: 10 req/sec vs 3 req/sec)\nNCBI_API_KEY=your_ncbi_api_key_here\n

The PubMed tool uses this configuration:

"},{"location":"configuration/#agent-configuration","title":"Agent Configuration","text":"

Control agent behavior and research loop execution:

# Maximum iterations per research loop (1-50, default: 10)\nMAX_ITERATIONS=10\n\n# Search timeout in seconds\nSEARCH_TIMEOUT=30\n\n# Use graph-based execution for research flows\nUSE_GRAPH_EXECUTION=false\n

The agent configuration fields:

"},{"location":"configuration/#budget-rate-limiting-configuration","title":"Budget & Rate Limiting Configuration","text":"

Control resource limits for research loops:

# Default token budget per research loop (1000-1000000, default: 100000)\nDEFAULT_TOKEN_LIMIT=100000\n\n# Default time limit per research loop in minutes (1-120, default: 10)\nDEFAULT_TIME_LIMIT_MINUTES=10\n\n# Default iterations limit per research loop (1-50, default: 10)\nDEFAULT_ITERATIONS_LIMIT=10\n

The budget configuration with validation:

"},{"location":"configuration/#rag-service-configuration","title":"RAG Service Configuration","text":"

Configure the Retrieval-Augmented Generation service:

# ChromaDB collection name for RAG\nRAG_COLLECTION_NAME=deepcritical_evidence\n\n# Number of top results to retrieve from RAG (1-50, default: 5)\nRAG_SIMILARITY_TOP_K=5\n\n# Automatically ingest evidence into RAG\nRAG_AUTO_INGEST=true\n

The RAG configuration:

"},{"location":"configuration/#chromadb-configuration","title":"ChromaDB Configuration","text":"

Configure the vector database for embeddings and RAG:

# ChromaDB storage path\nCHROMA_DB_PATH=./chroma_db\n\n# Whether to persist ChromaDB to disk\nCHROMA_DB_PERSIST=true\n\n# ChromaDB server host (for remote ChromaDB, optional)\nCHROMA_DB_HOST=localhost\n\n# ChromaDB server port (for remote ChromaDB, optional)\nCHROMA_DB_PORT=8000\n

The ChromaDB configuration:

"},{"location":"configuration/#external-services","title":"External Services","text":""},{"location":"configuration/#modal-configuration","title":"Modal Configuration","text":"

Modal is used for secure sandbox execution of statistical analysis:

# Modal Token ID (for Modal sandbox execution)\nMODAL_TOKEN_ID=your_modal_token_id_here\n\n# Modal Token Secret\nMODAL_TOKEN_SECRET=your_modal_token_secret_here\n

The Modal configuration:

"},{"location":"configuration/#logging-configuration","title":"Logging Configuration","text":"

Configure structured logging:

# Log Level: \"DEBUG\", \"INFO\", \"WARNING\", or \"ERROR\"\nLOG_LEVEL=INFO\n

The logging configuration:

Logging is configured via the configure_logging() function:

"},{"location":"configuration/#configuration-properties","title":"Configuration Properties","text":"

The Settings class provides helpful properties for checking configuration state:

"},{"location":"configuration/#api-key-availability","title":"API Key Availability","text":"

Check which API keys are available:

Usage:

from src.utils.config import settings\n\n# Check API key availability\nif settings.has_openai_key:\n    # Use OpenAI\n    pass\n\nif settings.has_anthropic_key:\n    # Use Anthropic\n    pass\n\nif settings.has_huggingface_key:\n    # Use HuggingFace\n    pass\n\nif settings.has_any_llm_key:\n    # At least one LLM is available\n    pass\n
"},{"location":"configuration/#service-availability","title":"Service Availability","text":"

Check if external services are configured:

Usage:

from src.utils.config import settings\n\n# Check service availability\nif settings.modal_available:\n    # Use Modal sandbox\n    pass\n\nif settings.web_search_available:\n    # Web search is configured\n    pass\n
"},{"location":"configuration/#api-key-retrieval","title":"API Key Retrieval","text":"

Get the API key for the configured provider:

For OpenAI-specific operations (e.g., Magentic mode):

"},{"location":"configuration/#configuration-usage-in-codebase","title":"Configuration Usage in Codebase","text":"

The configuration system is used throughout the codebase:

"},{"location":"configuration/#llm-factory","title":"LLM Factory","text":"

The LLM factory uses settings to create appropriate models:

"},{"location":"configuration/#embedding-service","title":"Embedding Service","text":"

The embedding service uses local embedding model configuration:

"},{"location":"configuration/#orchestrator-factory","title":"Orchestrator Factory","text":"

The orchestrator factory uses settings to determine mode:

"},{"location":"configuration/#environment-variables-reference","title":"Environment Variables Reference","text":""},{"location":"configuration/#required-at-least-one-llm","title":"Required (at least one LLM)","text":""},{"location":"configuration/#llm-configuration-variables","title":"LLM Configuration Variables","text":""},{"location":"configuration/#embedding-configuration-variables","title":"Embedding Configuration Variables","text":""},{"location":"configuration/#web-search-configuration-variables","title":"Web Search Configuration Variables","text":""},{"location":"configuration/#pubmed-configuration-variables","title":"PubMed Configuration Variables","text":""},{"location":"configuration/#agent-configuration-variables","title":"Agent Configuration Variables","text":""},{"location":"configuration/#budget-configuration-variables","title":"Budget Configuration Variables","text":""},{"location":"configuration/#rag-configuration-variables","title":"RAG Configuration Variables","text":""},{"location":"configuration/#chromadb-configuration-variables","title":"ChromaDB Configuration Variables","text":""},{"location":"configuration/#external-services-variables","title":"External Services Variables","text":""},{"location":"configuration/#logging-configuration-variables","title":"Logging Configuration Variables","text":""},{"location":"configuration/#validation","title":"Validation","text":"

Settings are validated on load using Pydantic validation:

"},{"location":"configuration/#validation-examples","title":"Validation Examples","text":"

The max_iterations field has range validation:

The llm_provider field has literal validation:

"},{"location":"configuration/#error-handling","title":"Error Handling","text":"

Configuration errors raise ConfigurationError from src/utils/exceptions.py:

```22:25:src/utils/exceptions.py class ConfigurationError(DeepCriticalError): \"\"\"Raised when configuration is invalid.\"\"\"

pass\n

```

"},{"location":"configuration/#error-handling-example","title":"Error Handling Example","text":"

python from src.utils.config import settings from src.utils.exceptions import ConfigurationError try: api_key = settings.get_api_key() except ConfigurationError as e: print(f\"Configuration error: {e}\")

"},{"location":"configuration/#common-configuration-errors","title":"Common Configuration Errors","text":"
  1. Missing API Key: When get_api_key() is called but the required API key is not set
  2. Invalid Provider: When llm_provider is set to an unsupported value
  3. Out of Range: When numeric values exceed their min/max constraints
  4. Invalid Literal: When enum fields receive unsupported values
"},{"location":"configuration/#configuration-best-practices","title":"Configuration Best Practices","text":"
  1. Use .env File: Store sensitive keys in .env file (add to .gitignore)
  2. Check Availability: Use properties like has_openai_key before accessing API keys
  3. Handle Errors: Always catch ConfigurationError when calling get_api_key()
  4. Validate Early: Configuration is validated on import, so errors surface immediately
  5. Use Defaults: Leverage sensible defaults for optional configuration
"},{"location":"configuration/#future-enhancements","title":"Future Enhancements","text":"

The following configurations are planned for future phases:

  1. Additional LLM Providers: DeepSeek, OpenRouter, Gemini, Perplexity, Azure OpenAI, Local models
  2. Model Selection: Reasoning/main/fast model configuration
  3. Service Integration: Additional service integrations and configurations
"},{"location":"configuration/CONFIGURATION/","title":"Configuration Guide","text":""},{"location":"configuration/CONFIGURATION/#overview","title":"Overview","text":"

DeepCritical uses Pydantic Settings for centralized configuration management. All settings are defined in the Settings class in src/utils/config.py and can be configured via environment variables or a .env file.

The configuration system provides:

"},{"location":"configuration/CONFIGURATION/#quick-start","title":"Quick Start","text":"
  1. Create a .env file in the project root
  2. Set at least one LLM API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, or HF_TOKEN)
  3. Optionally configure other services as needed
  4. The application will automatically load and validate your configuration
"},{"location":"configuration/CONFIGURATION/#configuration-system-architecture","title":"Configuration System Architecture","text":""},{"location":"configuration/CONFIGURATION/#settings-class","title":"Settings Class","text":"

The Settings class extends BaseSettings from pydantic_settings and defines all application configuration:

"},{"location":"configuration/CONFIGURATION/#singleton-instance","title":"Singleton Instance","text":"

A global settings instance is available for import:

"},{"location":"configuration/CONFIGURATION/#usage-pattern","title":"Usage Pattern","text":"

Access configuration throughout the codebase:

from src.utils.config import settings\n\n# Check if API keys are available\nif settings.has_openai_key:\n    # Use OpenAI\n    pass\n\n# Access configuration values\nmax_iterations = settings.max_iterations\nweb_search_provider = settings.web_search_provider\n
"},{"location":"configuration/CONFIGURATION/#required-configuration","title":"Required Configuration","text":""},{"location":"configuration/CONFIGURATION/#llm-provider","title":"LLM Provider","text":"

You must configure at least one LLM provider. The system supports:

"},{"location":"configuration/CONFIGURATION/#openai-configuration","title":"OpenAI Configuration","text":"
LLM_PROVIDER=openai\nOPENAI_API_KEY=your_openai_api_key_here\nOPENAI_MODEL=gpt-5.1\n

The default model is defined in the Settings class:

"},{"location":"configuration/CONFIGURATION/#anthropic-configuration","title":"Anthropic Configuration","text":"
LLM_PROVIDER=anthropic\nANTHROPIC_API_KEY=your_anthropic_api_key_here\nANTHROPIC_MODEL=claude-sonnet-4-5-20250929\n

The default model is defined in the Settings class:

"},{"location":"configuration/CONFIGURATION/#huggingface-configuration","title":"HuggingFace Configuration","text":"

HuggingFace can work without an API key for public models, but an API key provides higher rate limits:

# Option 1: Using HF_TOKEN (preferred)\nHF_TOKEN=your_huggingface_token_here\n\n# Option 2: Using HUGGINGFACE_API_KEY (alternative)\nHUGGINGFACE_API_KEY=your_huggingface_api_key_here\n\n# Default model\nHUGGINGFACE_MODEL=meta-llama/Llama-3.1-8B-Instruct\n

The HuggingFace token can be set via either environment variable:

"},{"location":"configuration/CONFIGURATION/#optional-configuration","title":"Optional Configuration","text":""},{"location":"configuration/CONFIGURATION/#embedding-configuration","title":"Embedding Configuration","text":"

DeepCritical supports multiple embedding providers for semantic search and RAG:

# Embedding Provider: \"openai\", \"local\", or \"huggingface\"\nEMBEDDING_PROVIDER=local\n\n# OpenAI Embedding Model (used by LlamaIndex RAG)\nOPENAI_EMBEDDING_MODEL=text-embedding-3-small\n\n# Local Embedding Model (sentence-transformers, used by EmbeddingService)\nLOCAL_EMBEDDING_MODEL=all-MiniLM-L6-v2\n\n# HuggingFace Embedding Model\nHUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2\n

The embedding provider configuration:

Note: OpenAI embeddings require OPENAI_API_KEY. The local provider (default) uses sentence-transformers and requires no API key.

"},{"location":"configuration/CONFIGURATION/#web-search-configuration","title":"Web Search Configuration","text":"

DeepCritical supports multiple web search providers:

# Web Search Provider: \"serper\", \"searchxng\", \"brave\", \"tavily\", or \"duckduckgo\"\n# Default: \"duckduckgo\" (no API key required)\nWEB_SEARCH_PROVIDER=duckduckgo\n\n# Serper API Key (for Google search via Serper)\nSERPER_API_KEY=your_serper_api_key_here\n\n# SearchXNG Host URL (for self-hosted search)\nSEARCHXNG_HOST=http://localhost:8080\n\n# Brave Search API Key\nBRAVE_API_KEY=your_brave_api_key_here\n\n# Tavily API Key\nTAVILY_API_KEY=your_tavily_api_key_here\n

The web search provider configuration:

Note: DuckDuckGo is the default and requires no API key, making it ideal for development and testing.

"},{"location":"configuration/CONFIGURATION/#pubmed-configuration","title":"PubMed Configuration","text":"

PubMed search supports optional NCBI API key for higher rate limits:

# NCBI API Key (optional, for higher rate limits: 10 req/sec vs 3 req/sec)\nNCBI_API_KEY=your_ncbi_api_key_here\n

The PubMed tool uses this configuration:

"},{"location":"configuration/CONFIGURATION/#agent-configuration","title":"Agent Configuration","text":"

Control agent behavior and research loop execution:

# Maximum iterations per research loop (1-50, default: 10)\nMAX_ITERATIONS=10\n\n# Search timeout in seconds\nSEARCH_TIMEOUT=30\n\n# Use graph-based execution for research flows\nUSE_GRAPH_EXECUTION=false\n

The agent configuration fields:

"},{"location":"configuration/CONFIGURATION/#budget-rate-limiting-configuration","title":"Budget & Rate Limiting Configuration","text":"

Control resource limits for research loops:

# Default token budget per research loop (1000-1000000, default: 100000)\nDEFAULT_TOKEN_LIMIT=100000\n\n# Default time limit per research loop in minutes (1-120, default: 10)\nDEFAULT_TIME_LIMIT_MINUTES=10\n\n# Default iterations limit per research loop (1-50, default: 10)\nDEFAULT_ITERATIONS_LIMIT=10\n

The budget configuration with validation:

"},{"location":"configuration/CONFIGURATION/#rag-service-configuration","title":"RAG Service Configuration","text":"

Configure the Retrieval-Augmented Generation service:

# ChromaDB collection name for RAG\nRAG_COLLECTION_NAME=deepcritical_evidence\n\n# Number of top results to retrieve from RAG (1-50, default: 5)\nRAG_SIMILARITY_TOP_K=5\n\n# Automatically ingest evidence into RAG\nRAG_AUTO_INGEST=true\n

The RAG configuration:

"},{"location":"configuration/CONFIGURATION/#chromadb-configuration","title":"ChromaDB Configuration","text":"

Configure the vector database for embeddings and RAG:

# ChromaDB storage path\nCHROMA_DB_PATH=./chroma_db\n\n# Whether to persist ChromaDB to disk\nCHROMA_DB_PERSIST=true\n\n# ChromaDB server host (for remote ChromaDB, optional)\nCHROMA_DB_HOST=localhost\n\n# ChromaDB server port (for remote ChromaDB, optional)\nCHROMA_DB_PORT=8000\n

The ChromaDB configuration:

"},{"location":"configuration/CONFIGURATION/#external-services","title":"External Services","text":""},{"location":"configuration/CONFIGURATION/#modal-configuration","title":"Modal Configuration","text":"

Modal is used for secure sandbox execution of statistical analysis:

# Modal Token ID (for Modal sandbox execution)\nMODAL_TOKEN_ID=your_modal_token_id_here\n\n# Modal Token Secret\nMODAL_TOKEN_SECRET=your_modal_token_secret_here\n

The Modal configuration:

"},{"location":"configuration/CONFIGURATION/#logging-configuration","title":"Logging Configuration","text":"

Configure structured logging:

# Log Level: \"DEBUG\", \"INFO\", \"WARNING\", or \"ERROR\"\nLOG_LEVEL=INFO\n

The logging configuration:

Logging is configured via the configure_logging() function:

"},{"location":"configuration/CONFIGURATION/#configuration-properties","title":"Configuration Properties","text":"

The Settings class provides helpful properties for checking configuration state:

"},{"location":"configuration/CONFIGURATION/#api-key-availability","title":"API Key Availability","text":"

Check which API keys are available:

Usage:

from src.utils.config import settings\n\n# Check API key availability\nif settings.has_openai_key:\n    # Use OpenAI\n    pass\n\nif settings.has_anthropic_key:\n    # Use Anthropic\n    pass\n\nif settings.has_huggingface_key:\n    # Use HuggingFace\n    pass\n\nif settings.has_any_llm_key:\n    # At least one LLM is available\n    pass\n
"},{"location":"configuration/CONFIGURATION/#service-availability","title":"Service Availability","text":"

Check if external services are configured:

Usage:

from src.utils.config import settings\n\n# Check service availability\nif settings.modal_available:\n    # Use Modal sandbox\n    pass\n\nif settings.web_search_available:\n    # Web search is configured\n    pass\n
"},{"location":"configuration/CONFIGURATION/#api-key-retrieval","title":"API Key Retrieval","text":"

Get the API key for the configured provider:

For OpenAI-specific operations (e.g., Magentic mode):

"},{"location":"configuration/CONFIGURATION/#configuration-usage-in-codebase","title":"Configuration Usage in Codebase","text":"

The configuration system is used throughout the codebase:

"},{"location":"configuration/CONFIGURATION/#llm-factory","title":"LLM Factory","text":"

The LLM factory uses settings to create appropriate models:

"},{"location":"configuration/CONFIGURATION/#embedding-service","title":"Embedding Service","text":"

The embedding service uses local embedding model configuration:

"},{"location":"configuration/CONFIGURATION/#orchestrator-factory","title":"Orchestrator Factory","text":"

The orchestrator factory uses settings to determine mode:

"},{"location":"configuration/CONFIGURATION/#environment-variables-reference","title":"Environment Variables Reference","text":""},{"location":"configuration/CONFIGURATION/#required-at-least-one-llm","title":"Required (at least one LLM)","text":""},{"location":"configuration/CONFIGURATION/#llm-configuration-variables","title":"LLM Configuration Variables","text":""},{"location":"configuration/CONFIGURATION/#embedding-configuration-variables","title":"Embedding Configuration Variables","text":""},{"location":"configuration/CONFIGURATION/#web-search-configuration-variables","title":"Web Search Configuration Variables","text":""},{"location":"configuration/CONFIGURATION/#pubmed-configuration-variables","title":"PubMed Configuration Variables","text":""},{"location":"configuration/CONFIGURATION/#agent-configuration-variables","title":"Agent Configuration Variables","text":""},{"location":"configuration/CONFIGURATION/#budget-configuration-variables","title":"Budget Configuration Variables","text":""},{"location":"configuration/CONFIGURATION/#rag-configuration-variables","title":"RAG Configuration Variables","text":""},{"location":"configuration/CONFIGURATION/#chromadb-configuration-variables","title":"ChromaDB Configuration Variables","text":""},{"location":"configuration/CONFIGURATION/#external-services-variables","title":"External Services Variables","text":""},{"location":"configuration/CONFIGURATION/#logging-configuration-variables","title":"Logging Configuration Variables","text":""},{"location":"configuration/CONFIGURATION/#validation","title":"Validation","text":"

Settings are validated on load using Pydantic validation:

"},{"location":"configuration/CONFIGURATION/#validation-examples","title":"Validation Examples","text":"

The max_iterations field has range validation:

The llm_provider field has literal validation:

"},{"location":"configuration/CONFIGURATION/#error-handling","title":"Error Handling","text":"

Configuration errors raise ConfigurationError from src/utils/exceptions.py:

"},{"location":"configuration/CONFIGURATION/#error-handling-example","title":"Error Handling Example","text":"
from src.utils.config import settings\nfrom src.utils.exceptions import ConfigurationError\n\ntry:\n    api_key = settings.get_api_key()\nexcept ConfigurationError as e:\n    print(f\"Configuration error: {e}\")\n
"},{"location":"configuration/CONFIGURATION/#common-configuration-errors","title":"Common Configuration Errors","text":"
  1. Missing API Key: When get_api_key() is called but the required API key is not set
  2. Invalid Provider: When llm_provider is set to an unsupported value
  3. Out of Range: When numeric values exceed their min/max constraints
  4. Invalid Literal: When enum fields receive unsupported values
"},{"location":"configuration/CONFIGURATION/#configuration-best-practices","title":"Configuration Best Practices","text":"
  1. Use .env File: Store sensitive keys in .env file (add to .gitignore)
  2. Check Availability: Use properties like has_openai_key before accessing API keys
  3. Handle Errors: Always catch ConfigurationError when calling get_api_key()
  4. Validate Early: Configuration is validated on import, so errors surface immediately
  5. Use Defaults: Leverage sensible defaults for optional configuration
"},{"location":"configuration/CONFIGURATION/#future-enhancements","title":"Future Enhancements","text":"

The following configurations are planned for future phases:

  1. Additional LLM Providers: DeepSeek, OpenRouter, Gemini, Perplexity, Azure OpenAI, Local models
  2. Model Selection: Reasoning/main/fast model configuration
  3. Service Integration: Additional service integrations and configurations
"},{"location":"contributing/","title":"Contributing to DeepCritical","text":"

Thank you for your interest in contributing to DeepCritical! This guide will help you get started.

"},{"location":"contributing/#git-workflow","title":"Git Workflow","text":""},{"location":"contributing/#development-commands","title":"Development Commands","text":"
make install      # Install dependencies + pre-commit\nmake check        # Lint + typecheck + test (MUST PASS)\nmake test         # Run unit tests\nmake lint         # Run ruff\nmake format       # Format with ruff\nmake typecheck    # Run mypy\nmake test-cov     # Test with coverage\n
"},{"location":"contributing/#getting-started","title":"Getting Started","text":"
  1. Fork the repository on GitHub
  2. Clone your fork: 
    git clone https://github.com/yourusername/GradioDemo.git\ncd GradioDemo\n
  3. Install dependencies: 
    make install\n
  4. Create a feature branch: 
    git checkout -b yourname-feature-name\n
  5. Make your changes following the guidelines below
  6. Run checks: 
    make check\n
  7. Commit and push: 
    git commit -m \"Description of changes\"\ngit push origin yourname-feature-name\n
  8. Create a pull request on GitHub
"},{"location":"contributing/#development-guidelines","title":"Development Guidelines","text":""},{"location":"contributing/#code-style","title":"Code Style","text":""},{"location":"contributing/#error-handling","title":"Error Handling","text":""},{"location":"contributing/#testing","title":"Testing","text":""},{"location":"contributing/#implementation-patterns","title":"Implementation Patterns","text":""},{"location":"contributing/#prompt-engineering","title":"Prompt Engineering","text":""},{"location":"contributing/#code-quality","title":"Code Quality","text":""},{"location":"contributing/#mcp-integration","title":"MCP Integration","text":""},{"location":"contributing/#mcp-tools","title":"MCP Tools","text":""},{"location":"contributing/#gradio-mcp-server","title":"Gradio MCP Server","text":""},{"location":"contributing/#common-pitfalls","title":"Common Pitfalls","text":"
  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Hallucinated citations: Always validate references
  4. Global mutable state: Use ContextVar or pass via parameters
  5. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
  6. Rate limiting: Always implement for external APIs
  7. Error chaining: Always use from e when raising exceptions
"},{"location":"contributing/#key-principles","title":"Key Principles","text":"
  1. Type Safety First: All code must pass mypy --strict
  2. Async Everything: All I/O must be async
  3. Test-Driven: Write tests before implementation
  4. No Hallucinations: Validate all citations
  5. Graceful Degradation: Support free tier (HF Inference) when no API keys
  6. Lazy Loading: Don't require optional dependencies at import time
  7. Structured Logging: Use structlog, never print()
  8. Error Chaining: Always preserve exception context
"},{"location":"contributing/#pull-request-process","title":"Pull Request Process","text":"
  1. Ensure all checks pass: make check
  2. Update documentation if needed
  3. Add tests for new features
  4. Update CHANGELOG if applicable
  5. Request review from maintainers
  6. Address review feedback
  7. Wait for approval before merging
"},{"location":"contributing/#questions","title":"Questions?","text":"

Thank you for contributing to DeepCritical!

"},{"location":"contributing/code-quality/","title":"Code Quality & Documentation","text":"

This document outlines code quality standards and documentation requirements.

"},{"location":"contributing/code-quality/#linting","title":"Linting","text":""},{"location":"contributing/code-quality/#type-checking","title":"Type Checking","text":""},{"location":"contributing/code-quality/#pre-commit","title":"Pre-commit","text":""},{"location":"contributing/code-quality/#documentation","title":"Documentation","text":""},{"location":"contributing/code-quality/#docstrings","title":"Docstrings","text":"

Example:

"},{"location":"contributing/code-quality/#code-comments","title":"Code Comments","text":""},{"location":"contributing/code-quality/#see-also","title":"See Also","text":""},{"location":"contributing/code-style/","title":"Code Style & Conventions","text":"

This document outlines the code style and conventions for DeepCritical.

"},{"location":"contributing/code-style/#type-safety","title":"Type Safety","text":""},{"location":"contributing/code-style/#pydantic-models","title":"Pydantic Models","text":""},{"location":"contributing/code-style/#async-patterns","title":"Async Patterns","text":" 
loop = asyncio.get_running_loop()\nresult = await loop.run_in_executor(None, cpu_bound_function, args)\n
"},{"location":"contributing/code-style/#common-pitfalls","title":"Common Pitfalls","text":"
  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Global mutable state: Use ContextVar or pass via parameters
  4. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
"},{"location":"contributing/code-style/#see-also","title":"See Also","text":""},{"location":"contributing/error-handling/","title":"Error Handling & Logging","text":"

This document outlines error handling and logging conventions for DeepCritical.

"},{"location":"contributing/error-handling/#exception-hierarchy","title":"Exception Hierarchy","text":"

Use custom exception hierarchy (src/utils/exceptions.py):

"},{"location":"contributing/error-handling/#error-handling-rules","title":"Error Handling Rules","text":" 
logger.error(\"Operation failed\", error=str(e), context=value)\n
"},{"location":"contributing/error-handling/#logging","title":"Logging","text":""},{"location":"contributing/error-handling/#logging-examples","title":"Logging Examples","text":"
logger.info(\"Starting search\", query=query, tools=[t.name for t in tools])\nlogger.warning(\"Search tool failed\", tool=tool.name, error=str(result))\nlogger.error(\"Assessment failed\", error=str(e))\n
"},{"location":"contributing/error-handling/#error-chaining","title":"Error Chaining","text":"

Always preserve exception context:

try:\n    result = await api_call()\nexcept httpx.HTTPError as e:\n    raise SearchError(f\"API call failed: {e}\") from e\n
"},{"location":"contributing/error-handling/#see-also","title":"See Also","text":""},{"location":"contributing/implementation-patterns/","title":"Implementation Patterns","text":"

This document outlines common implementation patterns used in DeepCritical.

"},{"location":"contributing/implementation-patterns/#search-tools","title":"Search Tools","text":"

All tools implement SearchTool protocol (src/tools/base.py):

Example pattern:

class MySearchTool:\n    @property\n    def name(self) -> str:\n        return \"mytool\"\n    \n    @retry(stop=stop_after_attempt(3), wait=wait_exponential(...))\n    async def search(self, query: str, max_results: int = 10) -> list[Evidence]:\n        # Implementation\n        return evidence_list\n
"},{"location":"contributing/implementation-patterns/#judge-handlers","title":"Judge Handlers","text":""},{"location":"contributing/implementation-patterns/#agent-factory-pattern","title":"Agent Factory Pattern","text":""},{"location":"contributing/implementation-patterns/#state-management","title":"State Management","text":""},{"location":"contributing/implementation-patterns/#singleton-pattern","title":"Singleton Pattern","text":"

Use @lru_cache(maxsize=1) for singletons:

"},{"location":"contributing/implementation-patterns/#see-also","title":"See Also","text":""},{"location":"contributing/prompt-engineering/","title":"Prompt Engineering & Citation Validation","text":"

This document outlines prompt engineering guidelines and citation validation rules.

"},{"location":"contributing/prompt-engineering/#judge-prompts","title":"Judge Prompts","text":""},{"location":"contributing/prompt-engineering/#hypothesis-prompts","title":"Hypothesis Prompts","text":""},{"location":"contributing/prompt-engineering/#report-prompts","title":"Report Prompts","text":""},{"location":"contributing/prompt-engineering/#citation-validation","title":"Citation Validation","text":""},{"location":"contributing/prompt-engineering/#citation-validation-rules","title":"Citation Validation Rules","text":"
  1. Every reference URL must EXACTLY match a provided evidence URL
  2. Do NOT invent, fabricate, or hallucinate any references
  3. Do NOT modify paper titles, authors, dates, or URLs
  4. If unsure about a citation, OMIT it rather than guess
  5. Copy URLs exactly as provided - do not create similar-looking URLs
"},{"location":"contributing/prompt-engineering/#evidence-selection","title":"Evidence Selection","text":""},{"location":"contributing/prompt-engineering/#see-also","title":"See Also","text":""},{"location":"contributing/testing/","title":"Testing Requirements","text":"

This document outlines testing requirements and guidelines for DeepCritical.

"},{"location":"contributing/testing/#test-structure","title":"Test Structure","text":""},{"location":"contributing/testing/#mocking","title":"Mocking","text":""},{"location":"contributing/testing/#tdd-workflow","title":"TDD Workflow","text":"
  1. Write failing test in tests/unit/
  2. Implement in src/
  3. Ensure test passes
  4. Run make check (lint + typecheck + test)
"},{"location":"contributing/testing/#test-examples","title":"Test Examples","text":"
@pytest.mark.unit\nasync def test_pubmed_search(mock_httpx_client):\n    tool = PubMedTool()\n    results = await tool.search(\"metformin\", max_results=5)\n    assert len(results) > 0\n    assert all(isinstance(r, Evidence) for r in results)\n\n@pytest.mark.integration\nasync def test_real_pubmed_search():\n    tool = PubMedTool()\n    results = await tool.search(\"metformin\", max_results=3)\n    assert len(results) <= 3\n
"},{"location":"contributing/testing/#test-coverage","title":"Test Coverage","text":""},{"location":"contributing/testing/#see-also","title":"See Also","text":""},{"location":"getting-started/examples/","title":"Examples","text":"

This page provides examples of using DeepCritical for various research tasks.

"},{"location":"getting-started/examples/#basic-research-query","title":"Basic Research Query","text":""},{"location":"getting-started/examples/#example-1-drug-information","title":"Example 1: Drug Information","text":"

Query: 

What are the latest treatments for Alzheimer's disease?\n

What DeepCritical Does: 1. Searches PubMed for recent papers 2. Searches ClinicalTrials.gov for active trials 3. Evaluates evidence quality 4. Synthesizes findings into a comprehensive report

"},{"location":"getting-started/examples/#example-2-clinical-trial-search","title":"Example 2: Clinical Trial Search","text":"

Query: 

What clinical trials are investigating metformin for cancer prevention?\n

What DeepCritical Does: 1. Searches ClinicalTrials.gov for relevant trials 2. Searches PubMed for supporting literature 3. Provides trial details and status 4. Summarizes findings

"},{"location":"getting-started/examples/#advanced-research-queries","title":"Advanced Research Queries","text":""},{"location":"getting-started/examples/#example-3-comprehensive-review","title":"Example 3: Comprehensive Review","text":"

Query: 

Review the evidence for using metformin as an anti-aging intervention, \nincluding clinical trials, mechanisms of action, and safety profile.\n

What DeepCritical Does: 1. Uses deep research mode (multi-section) 2. Searches multiple sources in parallel 3. Generates sections on: - Clinical trials - Mechanisms of action - Safety profile 4. Synthesizes comprehensive report

"},{"location":"getting-started/examples/#example-4-hypothesis-testing","title":"Example 4: Hypothesis Testing","text":"

Query: 

Test the hypothesis that regular exercise reduces Alzheimer's disease risk.\n

What DeepCritical Does: 1. Generates testable hypotheses 2. Searches for supporting/contradicting evidence 3. Performs statistical analysis (if Modal configured) 4. Provides verdict: SUPPORTED, REFUTED, or INCONCLUSIVE

"},{"location":"getting-started/examples/#mcp-tool-examples","title":"MCP Tool Examples","text":""},{"location":"getting-started/examples/#using-search_pubmed","title":"Using search_pubmed","text":"
Search PubMed for \"CRISPR gene editing cancer therapy\"\n
"},{"location":"getting-started/examples/#using-search_clinical_trials","title":"Using search_clinical_trials","text":"
Find active clinical trials for \"diabetes type 2 treatment\"\n
"},{"location":"getting-started/examples/#using-search_all","title":"Using search_all","text":"
Search all sources for \"COVID-19 vaccine side effects\"\n
"},{"location":"getting-started/examples/#using-analyze_hypothesis","title":"Using analyze_hypothesis","text":"
Analyze whether vitamin D supplementation reduces COVID-19 severity\n
"},{"location":"getting-started/examples/#code-examples","title":"Code Examples","text":""},{"location":"getting-started/examples/#python-api-usage","title":"Python API Usage","text":"
from src.orchestrator_factory import create_orchestrator\nfrom src.tools.search_handler import SearchHandler\nfrom src.agent_factory.judges import create_judge_handler\n\n# Create orchestrator\nsearch_handler = SearchHandler()\njudge_handler = create_judge_handler()\n
# Run research query\nquery = \"What are the latest treatments for Alzheimer's disease?\"\nasync for event in orchestrator.run(query):\n    print(f\"Event: {event.type} - {event.data}\")\n
"},{"location":"getting-started/examples/#gradio-ui-integration","title":"Gradio UI Integration","text":"
import gradio as gr\nfrom src.app import create_research_interface\n\n# Create interface\ninterface = create_research_interface()\n\n# Launch\ninterface.launch(server_name=\"0.0.0.0\", server_port=7860)\n
"},{"location":"getting-started/examples/#research-patterns","title":"Research Patterns","text":""},{"location":"getting-started/examples/#iterative-research","title":"Iterative Research","text":"

Single-loop research with search-judge-synthesize cycles:

from src.orchestrator.research_flow import IterativeResearchFlow\n
async for event in flow.run(query):\n    # Handle events\n    pass\n
"},{"location":"getting-started/examples/#deep-research","title":"Deep Research","text":"

Multi-section parallel research:

from src.orchestrator.research_flow import DeepResearchFlow\n
async for event in flow.run(query):\n    # Handle events\n    pass\n
"},{"location":"getting-started/examples/#configuration-examples","title":"Configuration Examples","text":""},{"location":"getting-started/examples/#basic-configuration","title":"Basic Configuration","text":"
# .env file\nLLM_PROVIDER=openai\nOPENAI_API_KEY=your_key_here\nMAX_ITERATIONS=10\n
"},{"location":"getting-started/examples/#advanced-configuration","title":"Advanced Configuration","text":"
# .env file\nLLM_PROVIDER=anthropic\nANTHROPIC_API_KEY=your_key_here\nEMBEDDING_PROVIDER=local\nWEB_SEARCH_PROVIDER=duckduckgo\nMAX_ITERATIONS=20\nDEFAULT_TOKEN_LIMIT=200000\nUSE_GRAPH_EXECUTION=true\n
"},{"location":"getting-started/examples/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/installation/","title":"Installation","text":"

This guide will help you install and set up DeepCritical on your system.

"},{"location":"getting-started/installation/#prerequisites","title":"Prerequisites","text":""},{"location":"getting-started/installation/#installation-steps","title":"Installation Steps","text":""},{"location":"getting-started/installation/#1-install-uv-recommended","title":"1. Install uv (Recommended)","text":"

uv is a fast Python package installer and resolver. Install it with:

pip install uv\n
"},{"location":"getting-started/installation/#2-clone-the-repository","title":"2. Clone the Repository","text":"
git clone https://github.com/DeepCritical/GradioDemo.git\ncd GradioDemo\n
"},{"location":"getting-started/installation/#3-install-dependencies","title":"3. Install Dependencies","text":"

Using uv (recommended):

uv sync\n

Using pip:

pip install -e .\n
"},{"location":"getting-started/installation/#4-install-optional-dependencies","title":"4. Install Optional Dependencies","text":"

For embeddings support (local sentence-transformers):

uv sync --extra embeddings\n

For Modal sandbox execution:

uv sync --extra modal\n

For Magentic orchestration:

uv sync --extra magentic\n

Install all extras:

uv sync --all-extras\n
"},{"location":"getting-started/installation/#5-configure-environment-variables","title":"5. Configure Environment Variables","text":"

Create a .env file in the project root:

# Required: At least one LLM provider\nLLM_PROVIDER=openai  # or \"anthropic\" or \"huggingface\"\nOPENAI_API_KEY=your_openai_api_key_here\n\n# Optional: Other services\nNCBI_API_KEY=your_ncbi_api_key_here  # For higher PubMed rate limits\nMODAL_TOKEN_ID=your_modal_token_id\nMODAL_TOKEN_SECRET=your_modal_token_secret\n

See the Configuration Guide for all available options.

"},{"location":"getting-started/installation/#6-verify-installation","title":"6. Verify Installation","text":"

Run the application:

uv run gradio run src/app.py\n

Open your browser to http://localhost:7860 to verify the installation.

"},{"location":"getting-started/installation/#development-setup","title":"Development Setup","text":"

For development, install dev dependencies:

uv sync --all-extras --dev\n

Install pre-commit hooks:

uv run pre-commit install\n
"},{"location":"getting-started/installation/#troubleshooting","title":"Troubleshooting","text":""},{"location":"getting-started/installation/#common-issues","title":"Common Issues","text":"

Import Errors: - Ensure you've installed all required dependencies - Check that Python 3.11+ is being used

API Key Errors: - Verify your .env file is in the project root - Check that API keys are correctly formatted - Ensure at least one LLM provider is configured

Module Not Found: - Run uv sync or pip install -e . again - Check that you're in the correct virtual environment

Port Already in Use: - Change the port in src/app.py or use environment variable - Kill the process using port 7860

"},{"location":"getting-started/installation/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/installation/#updated-upstream","title":"<<<<<<< Updated upstream","text":"

<<<<<<< Updated upstream

=======

Stashed changes

Stashed changes

"},{"location":"getting-started/mcp-integration/","title":"MCP Integration","text":"

DeepCritical exposes a Model Context Protocol (MCP) server, allowing you to use its search tools directly from Claude Desktop or other MCP clients.

"},{"location":"getting-started/mcp-integration/#what-is-mcp","title":"What is MCP?","text":"

The Model Context Protocol (MCP) is a standard for connecting AI assistants to external tools and data sources. DeepCritical implements an MCP server that exposes its search capabilities as MCP tools.

"},{"location":"getting-started/mcp-integration/#mcp-server-url","title":"MCP Server URL","text":"

When running locally:

http://localhost:7860/gradio_api/mcp/\n
"},{"location":"getting-started/mcp-integration/#claude-desktop-configuration","title":"Claude Desktop Configuration","text":""},{"location":"getting-started/mcp-integration/#1-locate-configuration-file","title":"1. Locate Configuration File","text":"

macOS: 

~/Library/Application Support/Claude/claude_desktop_config.json\n

Windows: 

%APPDATA%\\Claude\\claude_desktop_config.json\n

Linux: 

~/.config/Claude/claude_desktop_config.json\n

"},{"location":"getting-started/mcp-integration/#2-add-deepcritical-server","title":"2. Add DeepCritical Server","text":"

Edit claude_desktop_config.json and add:

{\n  \"mcpServers\": {\n    \"deepcritical\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#3-restart-claude-desktop","title":"3. Restart Claude Desktop","text":"

Close and restart Claude Desktop for changes to take effect.

"},{"location":"getting-started/mcp-integration/#4-verify-connection","title":"4. Verify Connection","text":"

In Claude Desktop, you should see DeepCritical tools available: - search_pubmed - search_clinical_trials - search_biorxiv - search_all - analyze_hypothesis

"},{"location":"getting-started/mcp-integration/#available-tools","title":"Available Tools","text":""},{"location":"getting-started/mcp-integration/#search_pubmed","title":"search_pubmed","text":"

Search peer-reviewed biomedical literature from PubMed.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search PubMed for \"metformin diabetes\"\n

"},{"location":"getting-started/mcp-integration/#search_clinical_trials","title":"search_clinical_trials","text":"

Search ClinicalTrials.gov for interventional studies.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search clinical trials for \"Alzheimer's disease treatment\"\n

"},{"location":"getting-started/mcp-integration/#search_biorxiv","title":"search_biorxiv","text":"

Search bioRxiv/medRxiv preprints via Europe PMC.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search bioRxiv for \"CRISPR gene editing\"\n

"},{"location":"getting-started/mcp-integration/#search_all","title":"search_all","text":"

Search all sources simultaneously (PubMed, ClinicalTrials.gov, Europe PMC).

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results per source (default: 10)

Example: 

Search all sources for \"COVID-19 vaccine efficacy\"\n

"},{"location":"getting-started/mcp-integration/#analyze_hypothesis","title":"analyze_hypothesis","text":"

Perform secure statistical analysis using Modal sandboxes.

Parameters: - hypothesis (string): Hypothesis to analyze - data (string, optional): Data description or code

Example: 

Analyze the hypothesis that metformin reduces cancer risk\n

"},{"location":"getting-started/mcp-integration/#using-tools-in-claude-desktop","title":"Using Tools in Claude Desktop","text":"

Once configured, you can ask Claude to use DeepCritical tools:

Use DeepCritical to search PubMed for recent papers on Alzheimer's disease treatments.\n

Claude will automatically: 1. Call the appropriate DeepCritical tool 2. Retrieve results 3. Use the results in its response

"},{"location":"getting-started/mcp-integration/#troubleshooting","title":"Troubleshooting","text":""},{"location":"getting-started/mcp-integration/#connection-issues","title":"Connection Issues","text":"

Server Not Found: - Ensure DeepCritical is running (uv run gradio run src/app.py) - Verify the URL in claude_desktop_config.json is correct - Check that port 7860 is not blocked by firewall

Tools Not Appearing: - Restart Claude Desktop after configuration changes - Check Claude Desktop logs for errors - Verify MCP server is accessible at the configured URL

"},{"location":"getting-started/mcp-integration/#authentication","title":"Authentication","text":"

If DeepCritical requires authentication: - Configure API keys in DeepCritical settings - Use HuggingFace OAuth login - Ensure API keys are valid

"},{"location":"getting-started/mcp-integration/#advanced-configuration","title":"Advanced Configuration","text":""},{"location":"getting-started/mcp-integration/#custom-port","title":"Custom Port","text":"

If running on a different port, update the URL:

{\n  \"mcpServers\": {\n    \"deepcritical\": {\n      \"url\": \"http://localhost:8080/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#multiple-instances","title":"Multiple Instances","text":"

You can configure multiple DeepCritical instances:

{\n  \"mcpServers\": {\n    \"deepcritical-local\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    },\n    \"deepcritical-remote\": {\n      \"url\": \"https://your-server.com/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/quick-start/","title":"Quick Start Guide","text":"

Get up and running with DeepCritical in minutes.

"},{"location":"getting-started/quick-start/#start-the-application","title":"Start the Application","text":"
uv run gradio run src/app.py\n

Open your browser to http://localhost:7860.

"},{"location":"getting-started/quick-start/#first-research-query","title":"First Research Query","text":"
  1. Enter a Research Question

Type your research question in the chat interface, for example: - \"What are the latest treatments for Alzheimer's disease?\" - \"Review the evidence for metformin in cancer prevention\" - \"What clinical trials are investigating COVID-19 vaccines?\"

  1. Submit the Query

Click \"Submit\" or press Enter. The system will: - Generate observations about your query - Identify knowledge gaps - Search multiple sources (PubMed, ClinicalTrials.gov, Europe PMC) - Evaluate evidence quality - Synthesize findings into a report

  1. Review Results

Watch the real-time progress in the chat interface: - Search operations and results - Evidence evaluation - Report generation - Final research report with citations

"},{"location":"getting-started/quick-start/#authentication","title":"Authentication","text":""},{"location":"getting-started/quick-start/#huggingface-oauth-recommended","title":"HuggingFace OAuth (Recommended)","text":"
  1. Click \"Sign in with HuggingFace\" at the top of the app
  2. Authorize the application
  3. Your HuggingFace API token will be automatically used
  4. No need to manually enter API keys
"},{"location":"getting-started/quick-start/#manual-api-key","title":"Manual API Key","text":"
  1. Open the Settings accordion
  2. Enter your API key:
  3. OpenAI API key
  4. Anthropic API key
  5. HuggingFace API key
  6. Click \"Save Settings\"
  7. Manual keys take priority over OAuth tokens
"},{"location":"getting-started/quick-start/#understanding-the-interface","title":"Understanding the Interface","text":""},{"location":"getting-started/quick-start/#chat-interface","title":"Chat Interface","text":""},{"location":"getting-started/quick-start/#status-indicators","title":"Status Indicators","text":""},{"location":"getting-started/quick-start/#settings","title":"Settings","text":""},{"location":"getting-started/quick-start/#example-queries","title":"Example Queries","text":""},{"location":"getting-started/quick-start/#simple-query","title":"Simple Query","text":"
What are the side effects of metformin?\n
"},{"location":"getting-started/quick-start/#complex-query","title":"Complex Query","text":"
Review the evidence for using metformin as an anti-aging intervention, \nincluding clinical trials, mechanisms of action, and safety profile.\n
"},{"location":"getting-started/quick-start/#clinical-trial-query","title":"Clinical Trial Query","text":"
What are the active clinical trials investigating Alzheimer's disease treatments?\n
"},{"location":"getting-started/quick-start/#next-steps","title":"Next Steps","text":""},{"location":"overview/architecture/","title":"Architecture Overview","text":"

DeepCritical is a deep research agent system that uses iterative search-and-judge loops to comprehensively answer research questions. The system supports multiple orchestration patterns, graph-based execution, parallel research workflows, and long-running task management with real-time streaming.

"},{"location":"overview/architecture/#core-architecture","title":"Core Architecture","text":""},{"location":"overview/architecture/#orchestration-patterns","title":"Orchestration Patterns","text":"
  1. Graph Orchestrator (src/orchestrator/graph_orchestrator.py):
  2. Graph-based execution using Pydantic AI agents as nodes
  3. Supports both iterative and deep research patterns
  4. Node types: Agent, State, Decision, Parallel
  5. Edge types: Sequential, Conditional, Parallel
  6. Conditional routing based on knowledge gaps, budget, and iterations
  7. Parallel execution for concurrent research loops
  8. Event streaming via AsyncGenerator[AgentEvent] for real-time UI updates

  9. Fallback to agent chains when graph execution is disabled

  10. Deep Research Flow (src/orchestrator/research_flow.py):

  11. Pattern: Planner \u2192 Parallel Iterative Loops (one per section) \u2192 Synthesis
  12. Uses PlannerAgent to break query into report sections
  13. Runs IterativeResearchFlow instances in parallel per section via WorkflowManager
  14. Synthesizes results using LongWriterAgent or ProofreaderAgent
  15. Supports both graph execution (use_graph=True) and agent chains (use_graph=False)
  16. Budget tracking per section and globally

  17. State synchronization across parallel loops

  18. Iterative Research Flow (src/orchestrator/research_flow.py):

  19. Pattern: Generate observations \u2192 Evaluate gaps \u2192 Select tools \u2192 Execute \u2192 Judge \u2192 Continue/Complete
  20. Uses KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent, WriterAgent
  21. JudgeHandler assesses evidence sufficiency
  22. Iterates until research complete or constraints met (iterations, time, tokens)

  23. Supports graph execution and agent chains

  24. Magentic Orchestrator (src/orchestrator_magentic.py):

  25. Multi-agent coordination using agent-framework-core
  26. ChatAgent pattern with internal LLMs per agent
  27. Uses MagenticBuilder with participants: searcher, hypothesizer, judge, reporter
  28. Manager orchestrates agents via OpenAIChatClient
  29. Requires OpenAI API key (function calling support)
  30. Event-driven: converts Magentic events to AgentEvent for UI streaming

  31. Supports long-running workflows with max rounds and stall/reset handling

  32. Hierarchical Orchestrator (src/orchestrator_hierarchical.py):

  33. Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge
  34. Adapts Magentic ChatAgent to SubIterationTeam protocol
  35. Event-driven via asyncio.Queue for coordination

  36. Supports sub-iteration patterns for complex research tasks

  37. Legacy Simple Mode (src/legacy_orchestrator.py):

  38. Linear search-judge-synthesize loop
  39. Uses SearchHandlerProtocol and JudgeHandlerProtocol
  40. Generator-based design yielding AgentEvent objects
  41. Backward compatibility for simple use cases
"},{"location":"overview/architecture/#long-running-task-support","title":"Long-Running Task Support","text":"

The system is designed for long-running research tasks with comprehensive state management and streaming:

  1. Event Streaming:
  2. All orchestrators yield AgentEvent objects via AsyncGenerator
  3. Real-time UI updates through Gradio chat interface
  4. Event types: started, searching, search_complete, judging, judge_complete, looping, synthesizing, hypothesizing, complete, error

  5. Metadata includes iteration numbers, tool names, result counts, durations

  6. Budget Tracking (src/middleware/budget_tracker.py):

  7. Per-loop and global budget management
  8. Tracks: tokens, time (seconds), iterations
  9. Budget enforcement at decision nodes
  10. Token estimation (~4 chars per token)
  11. Early termination when budgets exceeded

  12. Budget summaries for monitoring

  13. Workflow Manager (src/middleware/workflow_manager.py):

  14. Coordinates parallel research loops
  15. Tracks loop status: pending, running, completed, failed, cancelled
  16. Synchronizes evidence between loops and global state
  17. Handles errors per loop (doesn't fail all if one fails)
  18. Supports loop cancellation and timeout handling

  19. Evidence deduplication across parallel loops

  20. State Management (src/middleware/state_machine.py):

  21. Thread-safe isolation using ContextVar for concurrent requests
  22. WorkflowState tracks: evidence, conversation history, embedding service
  23. Evidence deduplication by URL
  24. Semantic search via embedding service
  25. State persistence across long-running workflows

  26. Supports both iterative and deep research patterns

  27. Gradio UI (src/app.py):

  28. Real-time streaming of research progress
  29. Accordion-based UI for pending/done operations
  30. OAuth integration (HuggingFace)
  31. Multiple backend support (API keys, free tier)
  32. Handles long-running tasks with progress indicators
  33. Event accumulation for pending operations
"},{"location":"overview/architecture/#graph-architecture","title":"Graph Architecture","text":"

The graph orchestrator (src/orchestrator/graph_orchestrator.py) implements a flexible graph-based execution model:

Node Types:

Edge Types:

Graph Patterns:

Execution Flow:

  1. Graph construction from nodes and edges
  2. Graph validation (no cycles, all nodes reachable)
  3. Graph execution from entry node
  4. Node execution based on type
  5. Edge evaluation for next node(s)
  6. Parallel execution via asyncio.gather()
  7. State updates at state nodes
  8. Event streaming for UI
"},{"location":"overview/architecture/#key-components","title":"Key Components","text":""},{"location":"overview/architecture/#research-team-parallel-execution","title":"Research Team & Parallel Execution","text":"

The system supports complex research workflows through:

  1. WorkflowManager: Coordinates multiple parallel research loops
  2. Creates and tracks ResearchLoop instances
  3. Runs loops in parallel via asyncio.gather()
  4. Synchronizes evidence to global state

  5. Handles loop failures gracefully

  6. Deep Research Pattern: Breaks complex queries into sections

  7. Planner creates report outline with sections
  8. Each section runs as independent iterative research loop
  9. Loops execute in parallel
  10. Evidence shared across loops via global state

  11. Final synthesis combines all section results

  12. State Synchronization: Thread-safe evidence sharing

  13. Evidence deduplication by URL
  14. Global state accessible to all loops
  15. Semantic search across all collected evidence
  16. Conversation history tracking per iteration
"},{"location":"overview/architecture/#configuration-modes","title":"Configuration & Modes","text":""},{"location":"overview/features/","title":"Features","text":"

DeepCritical provides a comprehensive set of features for AI-assisted research:

"},{"location":"overview/features/#core-features","title":"Core Features","text":""},{"location":"overview/features/#multi-source-search","title":"Multi-Source Search","text":""},{"location":"overview/features/#mcp-integration","title":"MCP Integration","text":""},{"location":"overview/features/#authentication","title":"Authentication","text":""},{"location":"overview/features/#secure-code-execution","title":"Secure Code Execution","text":""},{"location":"overview/features/#semantic-search-rag","title":"Semantic Search & RAG","text":""},{"location":"overview/features/#orchestration-patterns","title":"Orchestration Patterns","text":""},{"location":"overview/features/#real-time-streaming","title":"Real-Time Streaming","text":""},{"location":"overview/features/#budget-management","title":"Budget Management","text":""},{"location":"overview/features/#state-management","title":"State Management","text":""},{"location":"overview/features/#advanced-features","title":"Advanced Features","text":""},{"location":"overview/features/#agent-system","title":"Agent System","text":""},{"location":"overview/features/#search-tools","title":"Search Tools","text":""},{"location":"overview/features/#error-handling","title":"Error Handling","text":""},{"location":"overview/features/#configuration","title":"Configuration","text":""},{"location":"overview/features/#testing","title":"Testing","text":""},{"location":"overview/features/#ui-features","title":"UI Features","text":""},{"location":"overview/features/#gradio-interface","title":"Gradio Interface","text":""},{"location":"overview/features/#mcp-server","title":"MCP Server","text":""},{"location":"overview/features/#development-features","title":"Development Features","text":""},{"location":"overview/features/#code-quality","title":"Code Quality","text":""},{"location":"overview/features/#documentation","title":"Documentation","text":""},{"location":"overview/quick-start/","title":"Quick Start","text":"

Get started with DeepCritical in minutes.

"},{"location":"overview/quick-start/#installation","title":"Installation","text":"
# Install uv if you haven't already\npip install uv\n\n# Sync dependencies\nuv sync\n
"},{"location":"overview/quick-start/#run-the-ui","title":"Run the UI","text":"
# Start the Gradio app\nuv run gradio run src/app.py\n

Open your browser to http://localhost:7860.

"},{"location":"overview/quick-start/#basic-usage","title":"Basic Usage","text":""},{"location":"overview/quick-start/#1-authentication-optional","title":"1. Authentication (Optional)","text":"

HuggingFace OAuth Login: - Click the \"Sign in with HuggingFace\" button at the top of the app - Your HuggingFace API token will be automatically used for AI inference - No need to manually enter API keys when logged in

Manual API Key (BYOK): - Provide your own API key in the Settings accordion - Supports HuggingFace, OpenAI, or Anthropic API keys - Manual keys take priority over OAuth tokens

"},{"location":"overview/quick-start/#2-start-a-research-query","title":"2. Start a Research Query","text":"
  1. Enter your research question in the chat interface
  2. Click \"Submit\" or press Enter
  3. Watch the real-time progress as the system:
  4. Generates observations
  5. Identifies knowledge gaps
  6. Searches multiple sources
  7. Evaluates evidence
  8. Synthesizes findings
  9. Review the final research report
"},{"location":"overview/quick-start/#3-mcp-integration-optional","title":"3. MCP Integration (Optional)","text":"

Connect DeepCritical to Claude Desktop:

  1. Add to your claude_desktop_config.json: 

    {\n  \"mcpServers\": {\n    \"deepcritical\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    }\n  }\n}\n

  2. Restart Claude Desktop

  3. Use DeepCritical tools directly from Claude Desktop
"},{"location":"overview/quick-start/#available-tools","title":"Available Tools","text":""},{"location":"overview/quick-start/#next-steps","title":"Next Steps","text":""}]} \ No newline at end of file +{"config":{"lang":["en"],"separator":"[\\s\\-,:!=\\[\\]()\"`/]+|\\.(?!\\d)|&[lg]t;|&","pipeline":["stopWordFilter"],"fields":{"title":{"boost":1000.0},"text":{"boost":1.0},"tags":{"boost":1000000.0}}},"docs":[{"location":"","title":"The DETERMINATOR","text":"

Generalist Deep Research Agent - Stops at Nothing Until Finding Precise Answers

The DETERMINATOR is a powerful generalist deep research agent system that uses iterative search-and-judge loops to comprehensively investigate any research question. It stops at nothing until finding precise answers, only stopping at configured limits (budget, time, iterations).

Key Features: - Generalist: Handles queries from any domain (medical, technical, business, scientific, etc.) - Automatic Source Selection: Automatically determines if medical knowledge sources (PubMed, ClinicalTrials.gov) are needed - Multi-Source Search: Web search, PubMed, ClinicalTrials.gov, Europe PMC, RAG - Iterative Refinement: Continues searching and refining until precise answers are found - Evidence Synthesis: Comprehensive reports with proper citations

Important: The DETERMINATOR is a research tool that synthesizes evidence. It cannot provide medical advice or answer medical questions directly.

"},{"location":"#features","title":"Features","text":""},{"location":"#quick-start","title":"Quick Start","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync dependencies\nuv sync\n\n# Start the Gradio app\nuv run gradio run src/app.py\n

Open your browser to http://localhost:7860.

For detailed installation and setup instructions, see the Getting Started Guide.

"},{"location":"#architecture","title":"Architecture","text":"

The DETERMINATOR uses a Vertical Slice Architecture:

  1. Search Slice: Retrieving evidence from multiple sources (web, PubMed, ClinicalTrials.gov, Europe PMC, RAG) based on query analysis
  2. Judge Slice: Evaluating evidence quality using LLMs
  3. Orchestrator Slice: Managing the research loop and UI

The system supports three main research patterns:

Learn more about the Architecture.

"},{"location":"#documentation","title":"Documentation","text":""},{"location":"#links","title":"Links","text":""},{"location":"LICENSE/","title":"License","text":"

DeepCritical is licensed under the MIT License.

"},{"location":"LICENSE/#mit-license","title":"MIT License","text":"

Copyright (c) 2024 DeepCritical Team

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/","title":"MkDocs & Material UI Improvement Assessment","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#current-configuration-analysis","title":"Current Configuration Analysis","text":"

Your current mkdocs.yml already includes many excellent features: - \u2705 Material theme with light/dark mode toggle - \u2705 Navigation tabs, sections, expand, and top navigation - \u2705 Search with suggestions and highlighting - \u2705 Code annotation and copy buttons - \u2705 Mermaid diagram support - \u2705 Code include plugin - \u2705 Minification for performance - \u2705 Comprehensive markdown extensions

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#recommended-improvements","title":"Recommended Improvements","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#1-versioning--multi-version-documentation--high-priority","title":"1. Versioning & Multi-Version Documentation \u2b50 High Priority","text":"

If you plan to maintain multiple versions or branches:

plugins:\n  - search\n  - mermaid2\n  - codeinclude\n  - minify:\n      minify_html: true\n      minify_js: true\n      minify_css: true\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n  # Optional: For versioning\n  # - versioning:\n  #     version: ['dev', 'main']\n

Benefits: Shows when pages were last updated, helps users understand document freshness.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#2-git-integration--revision-information--high-priority","title":"2. Git Integration & Revision Information \u2b50 High Priority","text":"

Add revision dates and authors to pages:

plugins:\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n      fallback_to_build_date: true\n  - git-committers:\n      repository: DeepCritical/GradioDemo\n      branch: dev\n

Benefits: Users see when content was last updated, builds trust in documentation freshness.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#3-enhanced-navigation-features--high-priority","title":"3. Enhanced Navigation Features \u2b50 High Priority","text":"

Add breadcrumbs and improve navigation:

theme:\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.indexes  # Add index pages\n    - navigation.instant  # Instant page loads\n    - navigation.tracking  # Track scroll position\n    - navigation.smooth  # Smooth scrolling\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link  # Link to specific tabs\n    - content.tooltips  # Tooltips for abbreviations\n

Benefits: Better UX, easier navigation, professional feel.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#4-content-tabs-for-code-examples--high-priority","title":"4. Content Tabs for Code Examples \u2b50 High Priority","text":"

Perfect for showing multiple code examples (Python, TypeScript, etc.):

markdown_extensions:\n  - pymdownx.tabbed:\n      alternate_style: true\n      combine_header_slug: true  # Add this\n

Usage in docs: 

=== \"Python\"\n    ```python\n    def example():\n        pass\n    ```\n\n=== \"TypeScript\"\n    ```typescript\n    function example() {}\n    ```\n

Benefits: Clean way to show multiple implementations without cluttering pages.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#5-enhanced-admonitions--medium-priority","title":"5. Enhanced Admonitions \u2b50 Medium Priority","text":"

Add more admonition types and better styling:

markdown_extensions:\n  - admonition\n  - pymdownx.details\n  - pymdownx.superfences:\n      custom_fences:\n        - name: mermaid\n          class: mermaid\n          format: !!python/name:pymdownx.superfences.fence_code_format\n        # Add custom admonition fences\n        - name: danger\n          class: danger\n          format: !!python/name:pymdownx.superfences.fence_code_format\n

Usage: 

!!! danger \"Important\"\n    This is a critical warning.\n

Benefits: Better visual hierarchy for warnings, tips, and important information.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#6-math-formula-support--medium-priority-if-needed","title":"6. Math Formula Support \u2b50 Medium Priority (if needed)","text":"

If your documentation includes mathematical formulas:

markdown_extensions:\n  - pymdownx.arithmatex:\n      generic: true\n  - pymdownx.superfences:\n      custom_fences:\n        - name: math\n          class: arithmetic\n          format: !!python/name:pymdownx.superfences.fence_code_format\n\nextra_javascript:\n  - javascripts/mathjax.js\n  - https://polyfill.io/v3/polyfill.min.js?features=es6\n  - https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js\n

Benefits: Essential for scientific/technical documentation with formulas.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#7-better-code-highlighting--medium-priority","title":"7. Better Code Highlighting \u2b50 Medium Priority","text":"

Add more language support and better themes:

markdown_extensions:\n  - pymdownx.highlight:\n      anchor_linenums: true\n      line_spans: __span\n      pygments_lang_class: true\n      use_pygments: true\n      noclasses: false  # Use CSS classes instead of inline styles\n

Benefits: Better syntax highlighting, more language support.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#8-social-links-enhancement--low-priority","title":"8. Social Links Enhancement \u2b50 Low Priority","text":"

Add more social platforms and better icons:

extra:\n  social:\n    - icon: fontawesome/brands/github\n      link: https://github.com/DeepCritical/GradioDemo\n      name: GitHub\n    - icon: fontawesome/brands/twitter\n      link: https://twitter.com/yourhandle\n      name: Twitter\n    - icon: material/web\n      link: https://huggingface.co/spaces/DataQuests/DeepCritical\n      name: HuggingFace Space\n    - icon: fontawesome/brands/discord\n      link: https://discord.gg/yourserver\n      name: Discord\n

Benefits: Better community engagement, more ways to connect.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#9-analytics-integration--medium-priority","title":"9. Analytics Integration \u2b50 Medium Priority","text":"

Add privacy-respecting analytics:

extra:\n  analytics:\n    provider: google\n    property: G-XXXXXXXXXX\n  # Or use privacy-focused alternative:\n  # analytics:\n  #   provider: plausible\n  #   domain: yourdomain.com\n

Benefits: Understand how users interact with your documentation.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#10-custom-cssjs-for-branding--low-priority","title":"10. Custom CSS/JS for Branding \u2b50 Low Priority","text":"

Add custom styling:

extra_css:\n  - stylesheets/extra.css\n\nextra_javascript:\n  - javascripts/extra.js\n

Benefits: Customize appearance, add interactive features.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#11-better-table-of-contents--medium-priority","title":"11. Better Table of Contents \u2b50 Medium Priority","text":"

Enhance TOC with more options:

markdown_extensions:\n  - toc:\n      permalink: true\n      permalink_title: \"Anchor link to this section\"\n      baselevel: 1\n      toc_depth: 3\n      slugify: !!python/object/apply:pymdownx.slugs.slugify\n        kwds:\n          case: lower\n

Benefits: Better navigation within long pages, SEO-friendly anchor links.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#12-image-optimization--medium-priority","title":"12. Image Optimization \u2b50 Medium Priority","text":"

Add image handling plugin:

plugins:\n  - search\n  - mermaid2\n  - codeinclude\n  - minify:\n      minify_html: true\n      minify_js: true\n      minify_css: true\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n  # Optional: Image optimization\n  # - awesome-pages  # For better page organization\n

Benefits: Faster page loads, better mobile experience.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#13-keyboard-shortcuts--low-priority","title":"13. Keyboard Shortcuts \u2b50 Low Priority","text":"

Enable keyboard navigation:

theme:\n  keyboard_shortcuts:\n    search: true\n    previous: true\n    next: true\n

Benefits: Power users can navigate faster.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#14-print-styles--low-priority","title":"14. Print Styles \u2b50 Low Priority","text":"

Better printing experience:

theme:\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.indexes\n    - navigation.instant\n    - navigation.tracking\n    - navigation.smooth\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link\n    - content.tooltips\n    - content.action.edit  # Edit button\n    - content.action.view  # View source\n

Benefits: Users can print documentation cleanly.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#15-better-search-configuration--medium-priority","title":"15. Better Search Configuration \u2b50 Medium Priority","text":"

Enhance search capabilities:

plugins:\n  - search:\n      lang:\n        - en\n      separator: '[\\s\\-,:!=\\[\\]()\"`/]+|\\.(?!\\d)|&[lg]t;|&amp;'\n      prebuild_index: true  # For faster search\n      indexing: full  # Full-text indexing\n

Benefits: Faster, more accurate search results.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#16-api-documentation-enhancements--high-priority-for-your-api-docs","title":"16. API Documentation Enhancements \u2b50 High Priority (for your API docs)","text":"

Since you have extensive API documentation, consider:

markdown_extensions:\n  - pymdownx.superfences:\n      custom_fences:\n        - name: mermaid\n          class: mermaid\n          format: !!python/name:pymdownx.superfences.fence_code_format\n      preserve_tabs: true\n  # Add API-specific features\n  - attr_list\n  - md_in_html\n  - pymdownx.caret\n  - pymdownx.tilde\n

Benefits: Better formatting for API endpoints, parameters, responses.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#17-blognews-section--low-priority-if-needed","title":"17. Blog/News Section \u2b50 Low Priority (if needed)","text":"

If you want to add a blog:

plugins:\n  - blog:\n      blog_dir: blog\n      blog_description: \"News and updates\"\n      post_date_format: full\n      post_url_format: '{slug}'\n      archive: true\n

Benefits: Keep users updated with changelog, announcements.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#18-tags-and-categories--low-priority","title":"18. Tags and Categories \u2b50 Low Priority","text":"

Organize content with tags:

markdown_extensions:\n  - meta\n

Then in frontmatter: 

---\ntags:\n  - api\n  - agents\n  - getting-started\n---\n

Benefits: Better content organization, related content discovery.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#19-better-mobile-experience--high-priority","title":"19. Better Mobile Experience \u2b50 High Priority","text":"

Ensure mobile optimization:

theme:\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.instant  # Helps on mobile\n    - navigation.tracking\n    - navigation.smooth\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link\n    - content.tooltips\n    - toc.integrate  # Better mobile TOC\n

Benefits: Better experience for mobile users (growing segment).

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#20-feedback-mechanism--medium-priority","title":"20. Feedback Mechanism \u2b50 Medium Priority","text":"

Add feedback buttons:

extra:\n  feedback:\n    title: \"Was this page helpful?\"\n    ratings:\n      - icon: material/thumb-up-outline\n        name: \"This page was helpful\"\n      - icon: material/thumb-down-outline\n        name: \"This page could be improved\"\n

Benefits: Understand what content needs improvement.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#priority-recommendations","title":"Priority Recommendations","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#immediate-high-impact-easy-implementation","title":"Immediate (High Impact, Easy Implementation)","text":"
  1. \u2705 Git revision dates - Shows content freshness
  2. \u2705 Enhanced navigation features - Better UX
  3. \u2705 Content tabs - Perfect for code examples
  4. \u2705 Better search configuration - Faster search
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#short-term-high-impact-medium-effort","title":"Short-term (High Impact, Medium Effort)","text":"
  1. \u2705 API documentation enhancements - Better API docs
  2. \u2705 Enhanced admonitions - Better visual hierarchy
  3. \u2705 Mobile optimization - Better mobile experience
  4. \u2705 Analytics - Understand user behavior
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#long-term-nice-to-have","title":"Long-term (Nice to Have)","text":"
  1. \u26a0\ufe0f Versioning - If you need multiple versions
  2. \u26a0\ufe0f Math formulas - If you have mathematical content
  3. \u26a0\ufe0f Blog section - If you want to publish updates
  4. \u26a0\ufe0f Custom CSS/JS - For advanced customization
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#implementation-example","title":"Implementation Example","text":"

Here's an enhanced mkdocs.yml with the high-priority improvements:

site_name: The DETERMINATOR\nsite_description: Generalist Deep Research Agent that Stops at Nothing\nsite_author: The DETERMINATOR Team\nsite_url: https://deepcritical.github.io/GradioDemo/\n\nrepo_name: DeepCritical/GradioDemo\nrepo_url: https://github.com/DeepCritical/GradioDemo\nedit_uri: edit/dev/docs/\n\nstrict: false\n\ntheme:\n  name: material\n  palette:\n    - scheme: default\n      primary: orange\n      accent: red\n      toggle:\n        icon: material/brightness-7\n        name: Switch to dark mode\n    - scheme: slate\n      primary: orange\n      accent: red\n      toggle:\n        icon: material/brightness-4\n        name: Switch to light mode\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.indexes\n    - navigation.instant\n    - navigation.tracking\n    - navigation.smooth\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link\n    - content.tooltips\n    - toc.integrate\n  icon:\n    repo: fontawesome/brands/github\n  language: en\n\nplugins:\n  - search:\n      lang:\n        - en\n      separator: '[\\s\\-,:!=\\[\\]()\"`/]+|\\.(?!\\d)|&[lg]t;|&amp;'\n      prebuild_index: true\n      indexing: full\n  - mermaid2\n  - codeinclude\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n      fallback_to_build_date: true\n  - minify:\n      minify_html: true\n      minify_js: true\n      minify_css: true\n\nmarkdown_extensions:\n  - dev.docs_plugins:\n      base_path: \".\"\n  - pymdownx.highlight:\n      anchor_linenums: true\n      line_spans: __span\n      pygments_lang_class: true\n      use_pygments: true\n      noclasses: false\n  - pymdownx.inlinehilite\n  - pymdownx.superfences:\n      custom_fences:\n        - name: mermaid\n          class: mermaid\n          format: !!python/name:pymdownx.superfences.fence_code_format\n      preserve_tabs: true\n  - pymdownx.tabbed:\n      alternate_style: true\n      combine_header_slug: true\n  - pymdownx.tasklist:\n      custom_checkbox: true\n  - pymdownx.emoji:\n      emoji_generator: !!python/name:pymdownx.emoji.to_svg\n      emoji_index: !!python/name:pymdownx.emoji.twemoji\n  - pymdownx.snippets\n  - admonition\n  - pymdownx.details\n  - attr_list\n  - md_in_html\n  - tables\n  - meta\n  - toc:\n      permalink: true\n      permalink_title: \"Anchor link to this section\"\n      baselevel: 1\n      toc_depth: 3\n      slugify: !!python/object/apply:pymdownx.slugs.slugify\n        kwds:\n          case: lower\n\nnav:\n  - Home: index.md\n  - Overview:\n    - overview/architecture.md\n    - overview/features.md\n  - Getting Started:\n    - getting-started/installation.md\n    - getting-started/quick-start.md\n    - getting-started/mcp-integration.md\n    - getting-started/examples.md\n  - Configuration:\n    - configuration/index.md\n  - Architecture:\n    - \"Graph Orchestration\": architecture/graph_orchestration.md\n    - \"Workflow Diagrams\": architecture/workflow-diagrams.md\n    - \"Agents\": architecture/agents.md\n    - \"Orchestrators\": architecture/orchestrators.md\n    - \"Tools\": architecture/tools.md\n    - \"Middleware\": architecture/middleware.md\n    - \"Services\": architecture/services.md\n  - API Reference:\n    - api/agents.md\n    - api/tools.md\n    - api/orchestrators.md\n    - api/services.md\n    - api/models.md\n  - Contributing:\n    - contributing/index.md\n    - contributing/code-quality.md\n    - contributing/code-style.md\n    - contributing/error-handling.md\n    - contributing/implementation-patterns.md\n    - contributing/prompt-engineering.md\n    - contributing/testing.md\n  - License: LICENSE.md\n  - Team: team.md\n\nextra:\n  social:\n    - icon: fontawesome/brands/github\n      link: https://github.com/DeepCritical/GradioDemo\n      name: GitHub\n    - icon: material/web\n      link: https://huggingface.co/spaces/DataQuests/DeepCritical\n      name: HuggingFace Space\n  version:\n    provider: mike\n  generator:\n    enabled: false\n\ncopyright: Copyright &copy; 2024 DeepCritical Team\n
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#additional-documentation-improvements","title":"Additional Documentation Improvements","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#content-structure","title":"Content Structure","text":"
  1. Add a changelog page - Keep users informed of updates
  2. Add a FAQ section - Address common questions
  3. Add a glossary - Define technical terms
  4. Add a troubleshooting guide - Help users solve common issues
  5. Add video tutorials - Embed videos for complex topics
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#visual-enhancements","title":"Visual Enhancements","text":"
  1. Add diagrams - Use more Mermaid diagrams for complex flows
  2. Add screenshots - Visual guides for UI features
  3. Add code examples - More practical examples
  4. Add comparison tables - Compare different approaches/options
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#seo--discoverability","title":"SEO & Discoverability","text":"
  1. Add meta descriptions - Better search engine results
  2. Add Open Graph tags - Better social media sharing
  3. Add sitemap - Help search engines index your docs
  4. Add robots.txt - Control search engine crawling
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#next-steps","title":"Next Steps","text":"
  1. Review this assessment
  2. Prioritize features based on your needs
  3. Test changes in a branch
  4. Gather user feedback
  5. Iterate and improve
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#resources","title":"Resources","text":""},{"location":"team/","title":"Team","text":"

DeepCritical is developed by a team of researchers and developers working on AI-assisted research.

"},{"location":"team/#team-members","title":"Team Members","text":""},{"location":"team/#zj","title":"ZJ","text":""},{"location":"team/#mario-aderman","title":"Mario Aderman","text":""},{"location":"team/#joseph-pollack","title":"Joseph Pollack","text":""},{"location":"team/#virat-chauran","title":"Virat Chauran","text":""},{"location":"team/#anna-bossler","title":"Anna Bossler","text":""},{"location":"team/#about","title":"About","text":"

The DeepCritical team met online in the Alzheimer's Critical Literature Review Group in the Hugging Science initiative. We're building the agent framework we want to use for AI-assisted research to turn the vast amounts of clinical data into cures.

"},{"location":"team/#contributing","title":"Contributing","text":"

We welcome contributions! See the Contributing Guide for details.

"},{"location":"team/#links","title":"Links","text":""},{"location":"api/agents/","title":"Agents API Reference","text":"

This page documents the API for DeepCritical agents.

"},{"location":"api/agents/#knowledgegapagent","title":"KnowledgeGapAgent","text":"

Module: src.agents.knowledge_gap

Purpose: Evaluates research state and identifies knowledge gaps.

"},{"location":"api/agents/#methods","title":"Methods","text":""},{"location":"api/agents/#evaluate","title":"evaluate","text":"

Evaluates research completeness and identifies outstanding knowledge gaps.

Parameters: - query: Research query string - background_context: Background context for the query (default: \"\") - conversation_history: History of actions, findings, and thoughts as string (default: \"\") - iteration: Current iteration number (default: 0) - time_elapsed_minutes: Elapsed time in minutes (default: 0.0) - max_time_minutes: Maximum time limit in minutes (default: 10)

Returns: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

"},{"location":"api/agents/#toolselectoragent","title":"ToolSelectorAgent","text":"

Module: src.agents.tool_selector

Purpose: Selects appropriate tools for addressing knowledge gaps.

"},{"location":"api/agents/#methods_1","title":"Methods","text":""},{"location":"api/agents/#select_tools","title":"select_tools","text":"

Selects tools for addressing a knowledge gap.

Parameters: - gap: The knowledge gap to address - query: Research query string - background_context: Optional background context (default: \"\") - conversation_history: History of actions, findings, and thoughts as string (default: \"\")

Returns: AgentSelectionPlan with list of AgentTask objects.

"},{"location":"api/agents/#writeragent","title":"WriterAgent","text":"

Module: src.agents.writer

Purpose: Generates final reports from research findings.

"},{"location":"api/agents/#methods_2","title":"Methods","text":""},{"location":"api/agents/#write_report","title":"write_report","text":"

Generates a markdown report from research findings.

Parameters: - query: Research query string - findings: Research findings to include in report - output_length: Optional description of desired output length (default: \"\") - output_instructions: Optional additional instructions for report generation (default: \"\")

Returns: Markdown string with numbered citations.

"},{"location":"api/agents/#longwriteragent","title":"LongWriterAgent","text":"

Module: src.agents.long_writer

Purpose: Long-form report generation with section-by-section writing.

"},{"location":"api/agents/#methods_3","title":"Methods","text":""},{"location":"api/agents/#write_next_section","title":"write_next_section","text":"

Writes the next section of a long-form report.

Parameters: - original_query: The original research query - report_draft: Current report draft as string (all sections written so far) - next_section_title: Title of the section to write - next_section_draft: Draft content for the next section

Returns: LongWriterOutput with formatted section and references.

"},{"location":"api/agents/#write_report_1","title":"write_report","text":"

Generates final report from draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Complete report draft

Returns: Final markdown report string.

"},{"location":"api/agents/#proofreaderagent","title":"ProofreaderAgent","text":"

Module: src.agents.proofreader

Purpose: Proofreads and polishes report drafts.

"},{"location":"api/agents/#methods_4","title":"Methods","text":""},{"location":"api/agents/#proofread","title":"proofread","text":"

Proofreads and polishes a report draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Report draft to proofread

Returns: Polished markdown string.

"},{"location":"api/agents/#thinkingagent","title":"ThinkingAgent","text":"

Module: src.agents.thinking

Purpose: Generates observations from conversation history.

"},{"location":"api/agents/#methods_5","title":"Methods","text":""},{"location":"api/agents/#generate_observations","title":"generate_observations","text":"

Generates observations from conversation history.

Parameters: - query: Research query string - background_context: Optional background context (default: \"\") - conversation_history: History of actions, findings, and thoughts as string (default: \"\") - iteration: Current iteration number (default: 1)

Returns: Observation string.

"},{"location":"api/agents/#inputparseragent","title":"InputParserAgent","text":"

Module: src.agents.input_parser

Purpose: Parses and improves user queries, detects research mode.

"},{"location":"api/agents/#methods_6","title":"Methods","text":""},{"location":"api/agents/#parse","title":"parse","text":"

Parses and improves a user query.

Parameters: - query: Original query string

Returns: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: \"iterative\" or \"deep\" - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"api/agents/#factory-functions","title":"Factory Functions","text":"

All agents have factory functions in src.agent_factory.agents:

Parameters: - model: Optional Pydantic AI model. If None, uses get_model() from settings. - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars)

Returns: Agent instance.

"},{"location":"api/agents/#see-also","title":"See Also","text":""},{"location":"api/models/","title":"Models API Reference","text":"

This page documents the Pydantic models used throughout DeepCritical.

"},{"location":"api/models/#evidence","title":"Evidence","text":"

Module: src.utils.models

Purpose: Represents evidence from search results.

Fields: - citation: Citation information (title, URL, date, authors) - content: Evidence text content - relevance: Relevance score (0.0-1.0) - metadata: Additional metadata dictionary

"},{"location":"api/models/#citation","title":"Citation","text":"

Module: src.utils.models

Purpose: Citation information for evidence.

Fields: - source: Source name (e.g., \"pubmed\", \"clinicaltrials\", \"europepmc\", \"web\", \"rag\") - title: Article/trial title - url: Source URL - date: Publication date (YYYY-MM-DD or \"Unknown\") - authors: List of authors (optional)

"},{"location":"api/models/#knowledgegapoutput","title":"KnowledgeGapOutput","text":"

Module: src.utils.models

Purpose: Output from knowledge gap evaluation.

Fields: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

"},{"location":"api/models/#agentselectionplan","title":"AgentSelectionPlan","text":"

Module: src.utils.models

Purpose: Plan for tool/agent selection.

Fields: - tasks: List of agent tasks to execute

"},{"location":"api/models/#agenttask","title":"AgentTask","text":"

Module: src.utils.models

Purpose: Individual agent task.

Fields: - gap: The knowledge gap being addressed (optional) - agent: Name of agent to use - query: The specific query for the agent - entity_website: The website of the entity being researched, if known (optional)

"},{"location":"api/models/#reportdraft","title":"ReportDraft","text":"

Module: src.utils.models

Purpose: Draft structure for long-form reports.

Fields: - sections: List of report sections

"},{"location":"api/models/#reportsection","title":"ReportSection","text":"

Module: src.utils.models

Purpose: Individual section in a report draft.

Fields: - section_title: The title of the section - section_content: The content of the section

"},{"location":"api/models/#parsedquery","title":"ParsedQuery","text":"

Module: src.utils.models

Purpose: Parsed and improved query.

Fields: - original_query: Original query string - improved_query: Refined query string - research_mode: Research mode (\"iterative\" or \"deep\") - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"api/models/#conversation","title":"Conversation","text":"

Module: src.utils.models

Purpose: Conversation history with iterations.

Fields: - history: List of iteration data

"},{"location":"api/models/#iterationdata","title":"IterationData","text":"

Module: src.utils.models

Purpose: Data for a single iteration.

Fields: - gap: The gap addressed in the iteration - tool_calls: The tool calls made - findings: The findings collected from tool calls - thought: The thinking done to reflect on the success of the iteration and next steps

"},{"location":"api/models/#agentevent","title":"AgentEvent","text":"

Module: src.utils.models

Purpose: Event emitted during research execution.

Fields: - type: Event type (e.g., \"started\", \"search_complete\", \"complete\") - iteration: Iteration number (optional) - data: Event data dictionary

"},{"location":"api/models/#budgetstatus","title":"BudgetStatus","text":"

Module: src.utils.models

Purpose: Current budget status.

Fields: - tokens_used: Total tokens used - tokens_limit: Token budget limit - time_elapsed_seconds: Time elapsed in seconds - time_limit_seconds: Time budget limit (default: 600.0 seconds / 10 minutes) - iterations: Number of iterations completed - iterations_limit: Maximum iterations (default: 10) - iteration_tokens: Tokens used per iteration (iteration number -> token count)

"},{"location":"api/models/#see-also","title":"See Also","text":""},{"location":"api/orchestrators/","title":"Orchestrators API Reference","text":"

This page documents the API for DeepCritical orchestrators.

"},{"location":"api/orchestrators/#iterativeresearchflow","title":"IterativeResearchFlow","text":"

Module: src.orchestrator.research_flow

Purpose: Single-loop research with search-judge-synthesize cycles.

"},{"location":"api/orchestrators/#methods","title":"Methods","text":""},{"location":"api/orchestrators/#run","title":"run","text":"

Runs iterative research flow.

Parameters: - query: Research query string - background_context: Background context (default: \"\") - output_length: Optional description of desired output length (default: \"\") - output_instructions: Optional additional instructions for report generation (default: \"\")

Returns: Final report string.

Note: max_iterations, max_time_minutes, and token_budget are constructor parameters, not run() parameters.

"},{"location":"api/orchestrators/#deepresearchflow","title":"DeepResearchFlow","text":"

Module: src.orchestrator.research_flow

Purpose: Multi-section parallel research with planning and synthesis.

"},{"location":"api/orchestrators/#methods_1","title":"Methods","text":""},{"location":"api/orchestrators/#run_1","title":"run","text":"

Runs deep research flow.

Parameters: - query: Research query string

Returns: Final report string.

Note: max_iterations_per_section, max_time_minutes, and token_budget are constructor parameters, not run() parameters.

"},{"location":"api/orchestrators/#graphorchestrator","title":"GraphOrchestrator","text":"

Module: src.orchestrator.graph_orchestrator

Purpose: Graph-based execution using Pydantic AI agents as nodes.

"},{"location":"api/orchestrators/#methods_2","title":"Methods","text":""},{"location":"api/orchestrators/#run_2","title":"run","text":"

Runs graph-based research orchestration.

Parameters: - query: Research query string

Yields: AgentEvent objects during graph execution.

Note: research_mode and use_graph are constructor parameters, not run() parameters.

"},{"location":"api/orchestrators/#orchestrator-factory","title":"Orchestrator Factory","text":"

Module: src.orchestrator_factory

Purpose: Factory for creating orchestrators.

"},{"location":"api/orchestrators/#functions","title":"Functions","text":""},{"location":"api/orchestrators/#create_orchestrator","title":"create_orchestrator","text":"

Creates an orchestrator instance.

Parameters: - search_handler: Search handler protocol implementation (optional, required for simple mode) - judge_handler: Judge handler protocol implementation (optional, required for simple mode) - config: Configuration object (optional) - mode: Orchestrator mode (\"simple\", \"advanced\", \"magentic\", \"iterative\", \"deep\", \"auto\", or None for auto-detect) - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars)

Returns: Orchestrator instance.

Raises: - ValueError: If requirements not met

Modes: - \"simple\": Legacy orchestrator - \"advanced\" or \"magentic\": Magentic orchestrator (requires OpenAI API key) - None: Auto-detect based on API key availability

"},{"location":"api/orchestrators/#magenticorchestrator","title":"MagenticOrchestrator","text":"

Module: src.orchestrator_magentic

Purpose: Multi-agent coordination using Microsoft Agent Framework.

"},{"location":"api/orchestrators/#methods_3","title":"Methods","text":""},{"location":"api/orchestrators/#run_3","title":"run","text":"

Runs Magentic orchestration.

Parameters: - query: Research query string

Yields: AgentEvent objects converted from Magentic events.

Note: max_rounds and max_stalls are constructor parameters, not run() parameters.

Requirements: - agent-framework-core package - OpenAI API key

"},{"location":"api/orchestrators/#see-also","title":"See Also","text":""},{"location":"api/services/","title":"Services API Reference","text":"

This page documents the API for DeepCritical services.

"},{"location":"api/services/#embeddingservice","title":"EmbeddingService","text":"

Module: src.services.embeddings

Purpose: Local sentence-transformers for semantic search and deduplication.

"},{"location":"api/services/#methods","title":"Methods","text":""},{"location":"api/services/#embed","title":"embed","text":"

Generates embedding for a text string.

Parameters: - text: Text to embed

Returns: Embedding vector as list of floats.

"},{"location":"api/services/#embed_batch","title":"embed_batch","text":"
async def embed_batch(self, texts: list[str]) -> list[list[float]]\n

Generates embeddings for multiple texts.

Parameters: - texts: List of texts to embed

Returns: List of embedding vectors.

"},{"location":"api/services/#similarity","title":"similarity","text":"
async def similarity(self, text1: str, text2: str) -> float\n

Calculates similarity between two texts.

Parameters: - text1: First text - text2: Second text

Returns: Similarity score (0.0-1.0).

"},{"location":"api/services/#find_duplicates","title":"find_duplicates","text":"
async def find_duplicates(\n    self,\n    texts: list[str],\n    threshold: float = 0.85\n) -> list[tuple[int, int]]\n

Finds duplicate texts based on similarity threshold.

Parameters: - texts: List of texts to check - threshold: Similarity threshold (default: 0.85)

Returns: List of (index1, index2) tuples for duplicate pairs.

"},{"location":"api/services/#add_evidence","title":"add_evidence","text":"
async def add_evidence(\n    self,\n    evidence_id: str,\n    content: str,\n    metadata: dict[str, Any]\n) -> None\n

Adds evidence to vector store for semantic search.

Parameters: - evidence_id: Unique identifier for the evidence - content: Evidence text content - metadata: Additional metadata dictionary

"},{"location":"api/services/#search_similar","title":"search_similar","text":"
async def search_similar(\n    self,\n    query: str,\n    n_results: int = 5\n) -> list[dict[str, Any]]\n

Finds semantically similar evidence.

Parameters: - query: Search query string - n_results: Number of results to return (default: 5)

Returns: List of dictionaries with id, content, metadata, and distance keys.

"},{"location":"api/services/#deduplicate","title":"deduplicate","text":"
async def deduplicate(\n    self,\n    new_evidence: list[Evidence],\n    threshold: float = 0.9\n) -> list[Evidence]\n

Removes semantically duplicate evidence.

Parameters: - new_evidence: List of evidence items to deduplicate - threshold: Similarity threshold (default: 0.9, where 0.9 = 90% similar is duplicate)

Returns: List of unique evidence items (not already in vector store).

"},{"location":"api/services/#factory-function","title":"Factory Function","text":""},{"location":"api/services/#get_embedding_service","title":"get_embedding_service","text":"
@lru_cache(maxsize=1)\ndef get_embedding_service() -> EmbeddingService\n

Returns singleton EmbeddingService instance.

"},{"location":"api/services/#llamaindexragservice","title":"LlamaIndexRAGService","text":"

Module: src.services.rag

Purpose: Retrieval-Augmented Generation using LlamaIndex.

"},{"location":"api/services/#methods_1","title":"Methods","text":""},{"location":"api/services/#ingest_evidence","title":"ingest_evidence","text":"

Ingests evidence into RAG service.

Parameters: - evidence_list: List of Evidence objects to ingest

Note: Supports multiple embedding providers (OpenAI, local sentence-transformers, Hugging Face).

"},{"location":"api/services/#retrieve","title":"retrieve","text":"
def retrieve(\n    self,\n    query: str,\n    top_k: int | None = None\n) -> list[dict[str, Any]]\n

Retrieves relevant documents for a query.

Parameters: - query: Search query string - top_k: Number of top results to return (defaults to similarity_top_k from constructor)

Returns: List of dictionaries with text, score, and metadata keys.

"},{"location":"api/services/#query","title":"query","text":"
def query(\n    self,\n    query_str: str,\n    top_k: int | None = None\n) -> str\n

Queries RAG service and returns synthesized response.

Parameters: - query_str: Query string - top_k: Number of results to use (defaults to similarity_top_k from constructor)

Returns: Synthesized response string.

Raises: - ConfigurationError: If no LLM API key is available for query synthesis

"},{"location":"api/services/#ingest_documents","title":"ingest_documents","text":"
def ingest_documents(self, documents: list[Any]) -> None\n

Ingests raw LlamaIndex Documents.

Parameters: - documents: List of LlamaIndex Document objects

"},{"location":"api/services/#clear_collection","title":"clear_collection","text":"
def clear_collection(self) -> None\n

Clears all documents from the collection.

"},{"location":"api/services/#factory-function_1","title":"Factory Function","text":""},{"location":"api/services/#get_rag_service","title":"get_rag_service","text":"
def get_rag_service(\n    collection_name: str = \"deepcritical_evidence\",\n    oauth_token: str | None = None,\n    **kwargs: Any\n) -> LlamaIndexRAGService\n

Get or create a RAG service instance.

Parameters: - collection_name: Name of the ChromaDB collection (default: \"deepcritical_evidence\") - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars) - **kwargs: Additional arguments for LlamaIndexRAGService (e.g., use_openai_embeddings=False)

Returns: Configured LlamaIndexRAGService instance.

Note: By default, uses local embeddings (sentence-transformers) which require no API keys.

"},{"location":"api/services/#statisticalanalyzer","title":"StatisticalAnalyzer","text":"

Module: src.services.statistical_analyzer

Purpose: Secure execution of AI-generated statistical code.

"},{"location":"api/services/#methods_2","title":"Methods","text":""},{"location":"api/services/#analyze","title":"analyze","text":"
async def analyze(\n    self,\n    query: str,\n    evidence: list[Evidence],\n    hypothesis: dict[str, Any] | None = None\n) -> AnalysisResult\n

Analyzes a research question using statistical methods.

Parameters: - query: The research question - evidence: List of Evidence objects to analyze - hypothesis: Optional hypothesis dict with drug, target, pathway, effect, confidence keys

Returns: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - confidence: Confidence in verdict (0.0-1.0) - statistical_evidence: Summary of statistical findings - code_generated: Python code that was executed - execution_output: Output from code execution - key_takeaways: Key takeaways from analysis - limitations: List of limitations

Note: Requires Modal credentials for sandbox execution.

"},{"location":"api/services/#see-also","title":"See Also","text":""},{"location":"api/tools/","title":"Tools API Reference","text":"

This page documents the API for DeepCritical search tools.

"},{"location":"api/tools/#searchtool-protocol","title":"SearchTool Protocol","text":"

All tools implement the SearchTool protocol:

class SearchTool(Protocol):\n    @property\n    def name(self) -> str: ...\n    \n    async def search(\n        self, \n        query: str, \n        max_results: int = 10\n    ) -> list[Evidence]: ...\n
"},{"location":"api/tools/#pubmedtool","title":"PubMedTool","text":"

Module: src.tools.pubmed

Purpose: Search peer-reviewed biomedical literature from PubMed.

"},{"location":"api/tools/#properties","title":"Properties","text":""},{"location":"api/tools/#name","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"pubmed\"

"},{"location":"api/tools/#methods","title":"Methods","text":""},{"location":"api/tools/#search","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches PubMed for articles.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with PubMed articles.

Raises: - SearchError: If search fails (timeout, HTTP error, XML parsing error) - RateLimitError: If rate limit is exceeded (429 status code)

Note: Uses NCBI E-utilities (ESearch \u2192 EFetch). Rate limit: 0.34s between requests. Handles single vs. multiple articles.

"},{"location":"api/tools/#clinicaltrialstool","title":"ClinicalTrialsTool","text":"

Module: src.tools.clinicaltrials

Purpose: Search ClinicalTrials.gov for interventional studies.

"},{"location":"api/tools/#properties_1","title":"Properties","text":""},{"location":"api/tools/#name_1","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"clinicaltrials\"

"},{"location":"api/tools/#methods_1","title":"Methods","text":""},{"location":"api/tools/#search_1","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches ClinicalTrials.gov for trials.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with clinical trials.

Note: Only returns interventional studies with status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION. Uses requests library (NOT httpx - WAF blocks httpx). Runs in thread pool for async compatibility.

Raises: - SearchError: If search fails (HTTP error, request exception)

"},{"location":"api/tools/#europepmctool","title":"EuropePMCTool","text":"

Module: src.tools.europepmc

Purpose: Search Europe PMC for preprints and peer-reviewed articles.

"},{"location":"api/tools/#properties_2","title":"Properties","text":""},{"location":"api/tools/#name_2","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"europepmc\"

"},{"location":"api/tools/#methods_2","title":"Methods","text":""},{"location":"api/tools/#search_2","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches Europe PMC for articles and preprints.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with articles/preprints.

Note: Includes both preprints (marked with [PREPRINT - Not peer-reviewed]) and peer-reviewed articles. Handles preprint markers. Builds URLs from DOI or PMID.

Raises: - SearchError: If search fails (HTTP error, connection error)

"},{"location":"api/tools/#ragtool","title":"RAGTool","text":"

Module: src.tools.rag_tool

Purpose: Semantic search within collected evidence.

"},{"location":"api/tools/#initialization","title":"Initialization","text":"
def __init__(\n    self,\n    rag_service: LlamaIndexRAGService | None = None,\n    oauth_token: str | None = None\n) -> None\n

Parameters: - rag_service: Optional RAG service instance. If None, will be lazy-initialized. - oauth_token: Optional OAuth token from HuggingFace login (for RAG LLM)

"},{"location":"api/tools/#properties_3","title":"Properties","text":""},{"location":"api/tools/#name_3","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"rag\"

"},{"location":"api/tools/#methods_3","title":"Methods","text":""},{"location":"api/tools/#search_3","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches collected evidence using semantic similarity.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects from collected evidence.

Raises: - ConfigurationError: If RAG service is unavailable

Note: Requires evidence to be ingested into RAG service first. Wraps LlamaIndexRAGService. Returns Evidence from RAG results.

"},{"location":"api/tools/#searchhandler","title":"SearchHandler","text":"

Module: src.tools.search_handler

Purpose: Orchestrates parallel searches across multiple tools.

"},{"location":"api/tools/#initialization_1","title":"Initialization","text":"
def __init__(\n    self,\n    tools: list[SearchTool],\n    timeout: float = 30.0,\n    include_rag: bool = False,\n    auto_ingest_to_rag: bool = True,\n    oauth_token: str | None = None\n) -> None\n

Parameters: - tools: List of search tools to use - timeout: Timeout for each search in seconds (default: 30.0) - include_rag: Whether to include RAG tool in searches (default: False) - auto_ingest_to_rag: Whether to automatically ingest results into RAG (default: True) - oauth_token: Optional OAuth token from HuggingFace login (for RAG LLM)

"},{"location":"api/tools/#methods_4","title":"Methods","text":""},{"location":"api/tools/#execute","title":"execute","text":"

Searches multiple tools in parallel.

Parameters: - query: Search query string - max_results_per_tool: Maximum results per tool (default: 10)

Returns: SearchResult with: - query: The search query - evidence: Aggregated list of evidence - sources_searched: List of source names searched - total_found: Total number of results - errors: List of error messages from failed tools

Raises: - SearchError: If search times out

Note: Uses asyncio.gather() for parallel execution. Handles tool failures gracefully (returns errors in SearchResult.errors). Automatically ingests evidence into RAG if enabled.

"},{"location":"api/tools/#see-also","title":"See Also","text":""},{"location":"architecture/agents/","title":"Agents Architecture","text":"

DeepCritical uses Pydantic AI agents for all AI-powered operations. All agents follow a consistent pattern and use structured output types.

"},{"location":"architecture/agents/#agent-pattern","title":"Agent Pattern","text":""},{"location":"architecture/agents/#pydantic-ai-agents","title":"Pydantic AI Agents","text":"

Pydantic AI agents use the Agent class with the following structure:

Note: Factory functions accept an optional oauth_token parameter for HuggingFace authentication, which takes priority over environment variables.

"},{"location":"architecture/agents/#model-initialization","title":"Model Initialization","text":"

Agents use get_model() from src/agent_factory/judges.py if no model is provided. This supports:

The model selection is based on the configured LLM_PROVIDER in settings.

"},{"location":"architecture/agents/#error-handling","title":"Error Handling","text":"

Agents return fallback values on failure rather than raising exceptions:

All errors are logged with context using structlog.

"},{"location":"architecture/agents/#input-validation","title":"Input Validation","text":"

All agents validate inputs:

"},{"location":"architecture/agents/#output-types","title":"Output Types","text":"

Agents use structured output types from src/utils/models.py:

For text output (writer agents), agents return str directly.

"},{"location":"architecture/agents/#agent-types","title":"Agent Types","text":""},{"location":"architecture/agents/#knowledge-gap-agent","title":"Knowledge Gap Agent","text":"

File: src/agents/knowledge_gap.py

Purpose: Evaluates research state and identifies knowledge gaps.

Output: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

Methods: - async def evaluate(query, background_context, conversation_history, iteration, time_elapsed_minutes, max_time_minutes) -> KnowledgeGapOutput

"},{"location":"architecture/agents/#tool-selector-agent","title":"Tool Selector Agent","text":"

File: src/agents/tool_selector.py

Purpose: Selects appropriate tools for addressing knowledge gaps.

Output: AgentSelectionPlan with list of AgentTask objects.

Available Agents: - WebSearchAgent: General web search for fresh information - SiteCrawlerAgent: Research specific entities/companies - RAGAgent: Semantic search within collected evidence

"},{"location":"architecture/agents/#writer-agent","title":"Writer Agent","text":"

File: src/agents/writer.py

Purpose: Generates final reports from research findings.

Output: Markdown string with numbered citations.

Methods: - async def write_report(query, findings, output_length, output_instructions) -> str

Features: - Validates inputs - Truncates very long findings (max 50000 chars) with warning - Retry logic for transient failures (3 retries) - Citation validation before returning

"},{"location":"architecture/agents/#long-writer-agent","title":"Long Writer Agent","text":"

File: src/agents/long_writer.py

Purpose: Long-form report generation with section-by-section writing.

Input/Output: Uses ReportDraft models.

Methods: - async def write_next_section(query, draft, section_title, section_content) -> LongWriterOutput - async def write_report(query, report_title, report_draft) -> str

Features: - Writes sections iteratively - Aggregates references across sections - Reformats section headings and references - Deduplicates and renumbers references

"},{"location":"architecture/agents/#proofreader-agent","title":"Proofreader Agent","text":"

File: src/agents/proofreader.py

Purpose: Proofreads and polishes report drafts.

Input: ReportDraft Output: Polished markdown string

Methods: - async def proofread(query, report_title, report_draft) -> str

Features: - Removes duplicate content across sections - Adds executive summary if multiple sections - Preserves all references and citations - Improves flow and readability

"},{"location":"architecture/agents/#thinking-agent","title":"Thinking Agent","text":"

File: src/agents/thinking.py

Purpose: Generates observations from conversation history.

Output: Observation string

Methods: - async def generate_observations(query, background_context, conversation_history) -> str

"},{"location":"architecture/agents/#input-parser-agent","title":"Input Parser Agent","text":"

File: src/agents/input_parser.py

Purpose: Parses and improves user queries, detects research mode.

Output: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: \"iterative\" or \"deep\" - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"architecture/agents/#magentic-agents","title":"Magentic Agents","text":"

The following agents use the BaseAgent pattern from agent-framework and are used exclusively with MagenticOrchestrator:

"},{"location":"architecture/agents/#hypothesis-agent","title":"Hypothesis Agent","text":"

File: src/agents/hypothesis_agent.py

Purpose: Generates mechanistic hypotheses based on evidence.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Uses internal Pydantic AI Agent with HypothesisAssessment output type - Accesses shared evidence_store for evidence - Uses embedding service for diverse evidence selection (MMR algorithm) - Stores hypotheses in shared context

"},{"location":"architecture/agents/#search-agent","title":"Search Agent","text":"

File: src/agents/search_agent.py

Purpose: Wraps SearchHandler as an agent for Magentic orchestrator.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Executes searches via SearchHandlerProtocol - Deduplicates evidence using embedding service - Searches for semantically related evidence - Updates shared evidence store

"},{"location":"architecture/agents/#analysis-agent","title":"Analysis Agent","text":"

File: src/agents/analysis_agent.py

Purpose: Performs statistical analysis using Modal sandbox.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Wraps StatisticalAnalyzer service - Analyzes evidence and hypotheses - Returns verdict (SUPPORTED/REFUTED/INCONCLUSIVE) - Stores analysis results in shared context

"},{"location":"architecture/agents/#report-agent-magentic","title":"Report Agent (Magentic)","text":"

File: src/agents/report_agent.py

Purpose: Generates structured scientific reports from evidence and hypotheses.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Uses internal Pydantic AI Agent with ResearchReport output type - Accesses shared evidence store and hypotheses - Validates citations before returning - Formats report as markdown

"},{"location":"architecture/agents/#judge-agent","title":"Judge Agent","text":"

File: src/agents/judge_agent.py

Purpose: Evaluates evidence quality and determines if sufficient for synthesis.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse - async def run_stream(messages, thread, **kwargs) -> AsyncIterable[AgentRunResponseUpdate]

Features: - Wraps JudgeHandlerProtocol - Accesses shared evidence store - Returns JudgeAssessment with sufficient flag, confidence, and recommendation

"},{"location":"architecture/agents/#agent-patterns","title":"Agent Patterns","text":"

DeepCritical uses two distinct agent patterns:

"},{"location":"architecture/agents/#1-pydantic-ai-agents-traditional-pattern","title":"1. Pydantic AI Agents (Traditional Pattern)","text":"

These agents use the Pydantic AI Agent class directly and are used in iterative and deep research flows:

"},{"location":"architecture/agents/#2-magentic-agents-agent-framework-pattern","title":"2. Magentic Agents (Agent-Framework Pattern)","text":"

These agents use the BaseAgent class from agent-framework and are used in Magentic orchestrator:

Note: Magentic agents are used exclusively with the MagenticOrchestrator and follow the agent-framework protocol for multi-agent coordination.

"},{"location":"architecture/agents/#factory-functions","title":"Factory Functions","text":"

All agents have factory functions in src/agent_factory/agents.py:

Factory functions: - Use get_model() if no model provided - Accept oauth_token parameter for HuggingFace authentication - Raise ConfigurationError if creation fails - Log agent creation

"},{"location":"architecture/agents/#see-also","title":"See Also","text":""},{"location":"architecture/graph_orchestration/","title":"Graph Orchestration Architecture","text":""},{"location":"architecture/graph_orchestration/#overview","title":"Overview","text":"

DeepCritical implements a graph-based orchestration system for research workflows using Pydantic AI agents as nodes. This enables better parallel execution, conditional routing, and state management compared to simple agent chains.

"},{"location":"architecture/graph_orchestration/#graph-patterns","title":"Graph Patterns","text":""},{"location":"architecture/graph_orchestration/#iterative-research-graph","title":"Iterative Research Graph","text":"

The iterative research graph follows this pattern:

[Input] \u2192 [Thinking] \u2192 [Knowledge Gap] \u2192 [Decision: Complete?]\n                                              \u2193 No          \u2193 Yes\n                                    [Tool Selector]    [Writer]\n                                              \u2193\n                                    [Execute Tools] \u2192 [Loop Back]\n

Node IDs: thinking \u2192 knowledge_gap \u2192 continue_decision \u2192 tool_selector/writer \u2192 execute_tools \u2192 (loop back to thinking)

Special Node Handling: - execute_tools: State node that uses search_handler to execute searches and add evidence to workflow state - continue_decision: Decision node that routes based on research_complete flag from KnowledgeGapOutput

"},{"location":"architecture/graph_orchestration/#deep-research-graph","title":"Deep Research Graph","text":"

The deep research graph follows this pattern:

[Input] \u2192 [Planner] \u2192 [Store Plan] \u2192 [Parallel Loops] \u2192 [Collect Drafts] \u2192 [Synthesizer]\n                                        \u2193         \u2193         \u2193\n                                     [Loop1]  [Loop2]  [Loop3]\n

Node IDs: planner \u2192 store_plan \u2192 parallel_loops \u2192 collect_drafts \u2192 synthesizer

Special Node Handling: - planner: Agent node that creates ReportPlan with report outline - store_plan: State node that stores ReportPlan in context for parallel loops - parallel_loops: Parallel node that executes IterativeResearchFlow instances for each section - collect_drafts: State node that collects section drafts from parallel loops - synthesizer: Agent node that calls LongWriterAgent.write_report() directly with ReportDraft

"},{"location":"architecture/graph_orchestration/#deep-research","title":"Deep Research","text":"
\nsequenceDiagram\n    actor User\n    participant GraphOrchestrator\n    participant InputParser\n    participant GraphBuilder\n    participant GraphExecutor\n    participant Agent\n    participant BudgetTracker\n    participant WorkflowState\n\n    User->>GraphOrchestrator: run(query)\n    GraphOrchestrator->>InputParser: detect_research_mode(query)\n    InputParser-->>GraphOrchestrator: mode (iterative/deep)\n    GraphOrchestrator->>GraphBuilder: build_graph(mode)\n    GraphBuilder-->>GraphOrchestrator: ResearchGraph\n    GraphOrchestrator->>WorkflowState: init_workflow_state()\n    GraphOrchestrator->>BudgetTracker: create_budget()\n    GraphOrchestrator->>GraphExecutor: _execute_graph(graph)\n    \n    loop For each node in graph\n        GraphExecutor->>Agent: execute_node(agent_node)\n        Agent->>Agent: process_input\n        Agent-->>GraphExecutor: result\n        GraphExecutor->>WorkflowState: update_state(result)\n        GraphExecutor->>BudgetTracker: add_tokens(used)\n        GraphExecutor->>BudgetTracker: check_budget()\n        alt Budget exceeded\n            GraphExecutor->>GraphOrchestrator: emit(error_event)\n        else Continue\n            GraphExecutor->>GraphOrchestrator: emit(progress_event)\n        end\n    end\n    \n    GraphOrchestrator->>User: AsyncGenerator[AgentEvent]\n
"},{"location":"architecture/graph_orchestration/#iterative-research","title":"Iterative Research","text":"
sequenceDiagram\n    participant IterativeFlow\n    participant ThinkingAgent\n    participant KnowledgeGapAgent\n    participant ToolSelector\n    participant ToolExecutor\n    participant JudgeHandler\n    participant WriterAgent\n\n    IterativeFlow->>IterativeFlow: run(query)\n    \n    loop Until complete or max_iterations\n        IterativeFlow->>ThinkingAgent: generate_observations()\n        ThinkingAgent-->>IterativeFlow: observations\n        \n        IterativeFlow->>KnowledgeGapAgent: evaluate_gaps()\n        KnowledgeGapAgent-->>IterativeFlow: KnowledgeGapOutput\n        \n        alt Research complete\n            IterativeFlow->>WriterAgent: create_final_report()\n            WriterAgent-->>IterativeFlow: final_report\n        else Gaps remain\n            IterativeFlow->>ToolSelector: select_agents(gap)\n            ToolSelector-->>IterativeFlow: AgentSelectionPlan\n            \n            IterativeFlow->>ToolExecutor: execute_tool_tasks()\n            ToolExecutor-->>IterativeFlow: ToolAgentOutput[]\n            \n            IterativeFlow->>JudgeHandler: assess_evidence()\n            JudgeHandler-->>IterativeFlow: should_continue\n        end\n    end
"},{"location":"architecture/graph_orchestration/#graph-structure","title":"Graph Structure","text":""},{"location":"architecture/graph_orchestration/#nodes","title":"Nodes","text":"

Graph nodes represent different stages in the research workflow:

  1. Agent Nodes: Execute Pydantic AI agents
  2. Input: Prompt/query
  3. Output: Structured or unstructured response

  4. Examples: KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent

  5. State Nodes: Update or read workflow state

  6. Input: Current state
  7. Output: Updated state

  8. Examples: Update evidence, update conversation history

  9. Decision Nodes: Make routing decisions based on conditions

  10. Input: Current state/results
  11. Output: Next node ID

  12. Examples: Continue research vs. complete research

  13. Parallel Nodes: Execute multiple nodes concurrently

  14. Input: List of node IDs
  15. Output: Aggregated results
  16. Examples: Parallel iterative research loops
"},{"location":"architecture/graph_orchestration/#edges","title":"Edges","text":"

Edges define transitions between nodes:

  1. Sequential Edges: Always traversed (no condition)
  2. From: Source node
  3. To: Target node

  4. Condition: None (always True)

  5. Conditional Edges: Traversed based on condition

  6. From: Source node
  7. To: Target node
  8. Condition: Callable that returns bool

  9. Example: If research complete \u2192 go to writer, else \u2192 continue loop

  10. Parallel Edges: Used for parallel execution branches

  11. From: Parallel node
  12. To: Multiple target nodes
  13. Execution: All targets run concurrently
"},{"location":"architecture/graph_orchestration/#state-management","title":"State Management","text":"

State is managed via WorkflowState using ContextVar for thread-safe isolation:

State transitions occur at state nodes, which update the global workflow state.

"},{"location":"architecture/graph_orchestration/#execution-flow","title":"Execution Flow","text":"
  1. Graph Construction: Build graph from nodes and edges using create_iterative_graph() or create_deep_graph()
  2. Graph Validation: Ensure graph is valid (no cycles, all nodes reachable) via ResearchGraph.validate_structure()
  3. Graph Execution: Traverse graph from entry node using GraphOrchestrator._execute_graph()
  4. Node Execution: Execute each node based on type:
  5. Agent Nodes: Call agent.run() with transformed input
  6. State Nodes: Update workflow state via state_updater function
  7. Decision Nodes: Evaluate decision_function to get next node ID
  8. Parallel Nodes: Execute all parallel nodes concurrently via asyncio.gather()
  9. Edge Evaluation: Determine next node(s) based on edges and conditions
  10. Parallel Execution: Use asyncio.gather() for parallel nodes
  11. State Updates: Update state at state nodes via GraphExecutionContext.update_state()
  12. Event Streaming: Yield AgentEvent objects during execution for UI
"},{"location":"architecture/graph_orchestration/#graphexecutioncontext","title":"GraphExecutionContext","text":"

The GraphExecutionContext class manages execution state during graph traversal:

Methods: - set_node_result(node_id, result): Store result from node execution - get_node_result(node_id): Retrieve stored result - has_visited(node_id): Check if node was visited - mark_visited(node_id): Mark node as visited - update_state(updater, data): Update workflow state

"},{"location":"architecture/graph_orchestration/#conditional-routing","title":"Conditional Routing","text":"

Decision nodes evaluate conditions and return next node IDs:

"},{"location":"architecture/graph_orchestration/#parallel-execution","title":"Parallel Execution","text":"

Parallel nodes execute multiple nodes concurrently:

"},{"location":"architecture/graph_orchestration/#budget-enforcement","title":"Budget Enforcement","text":"

Budget constraints are enforced at decision nodes:

If any budget is exceeded, execution routes to exit node.

"},{"location":"architecture/graph_orchestration/#error-handling","title":"Error Handling","text":"

Errors are handled at multiple levels:

  1. Node Level: Catch errors in individual node execution
  2. Graph Level: Handle errors during graph traversal
  3. State Level: Rollback state changes on error

Errors are logged and yield error events for UI.

"},{"location":"architecture/graph_orchestration/#backward-compatibility","title":"Backward Compatibility","text":"

Graph execution is optional via feature flag:

This allows gradual migration and fallback if needed.

"},{"location":"architecture/graph_orchestration/#see-also","title":"See Also","text":""},{"location":"architecture/middleware/","title":"Middleware Architecture","text":"

DeepCritical uses middleware for state management, budget tracking, and workflow coordination.

"},{"location":"architecture/middleware/#state-management","title":"State Management","text":""},{"location":"architecture/middleware/#workflowstate","title":"WorkflowState","text":"

File: src/middleware/state_machine.py

Purpose: Thread-safe state management for research workflows

Implementation: Uses ContextVar for thread-safe isolation

State Components: - evidence: list[Evidence]: Collected evidence from searches - conversation: Conversation: Iteration history (gaps, tool calls, findings, thoughts) - embedding_service: Any: Embedding service for semantic search

Methods: - add_evidence(new_evidence: list[Evidence]) -> int: Adds evidence with URL-based deduplication. Returns the number of new items added (excluding duplicates). - async search_related(query: str, n_results: int = 5) -> list[Evidence]: Semantic search for related evidence using embedding service

Initialization:

Access:

"},{"location":"architecture/middleware/#workflow-manager","title":"Workflow Manager","text":"

File: src/middleware/workflow_manager.py

Purpose: Coordinates parallel research loops

Methods: - async add_loop(loop_id: str, query: str) -> ResearchLoop: Add a new research loop to manage - async run_loops_parallel(loop_configs: list[dict], loop_func: Callable, judge_handler: Any | None = None, budget_tracker: Any | None = None) -> list[Any]: Run multiple research loops in parallel. Takes configuration dicts and a loop function. - async update_loop_status(loop_id: str, status: LoopStatus, error: str | None = None): Update loop status - async sync_loop_evidence_to_state(loop_id: str): Synchronize evidence from a specific loop to global state

Features: - Uses asyncio.gather() for parallel execution - Handles errors per loop (doesn't fail all if one fails) - Tracks loop status: pending, running, completed, failed, cancelled - Evidence deduplication across parallel loops

Usage: 

from src.middleware.workflow_manager import WorkflowManager\n\nmanager = WorkflowManager()\nawait manager.add_loop(\"loop1\", \"Research query 1\")\nawait manager.add_loop(\"loop2\", \"Research query 2\")\n\nasync def run_research(config: dict) -> str:\n    loop_id = config[\"loop_id\"]\n    query = config[\"query\"]\n    # ... research logic ...\n    return \"report\"\n\nresults = await manager.run_loops_parallel(\n    loop_configs=[\n        {\"loop_id\": \"loop1\", \"query\": \"Research query 1\"},\n        {\"loop_id\": \"loop2\", \"query\": \"Research query 2\"},\n    ],\n    loop_func=run_research,\n)\n

"},{"location":"architecture/middleware/#budget-tracker","title":"Budget Tracker","text":"

File: src/middleware/budget_tracker.py

Purpose: Tracks and enforces resource limits

Budget Components: - Tokens: LLM token usage - Time: Elapsed time in seconds - Iterations: Number of iterations

Methods: - create_budget(loop_id: str, tokens_limit: int = 100000, time_limit_seconds: float = 600.0, iterations_limit: int = 10) -> BudgetStatus: Create a budget for a specific loop - add_tokens(loop_id: str, tokens: int): Add token usage to a loop's budget - start_timer(loop_id: str): Start time tracking for a loop - update_timer(loop_id: str): Update elapsed time for a loop - increment_iteration(loop_id: str): Increment iteration count for a loop - check_budget(loop_id: str) -> tuple[bool, str]: Check if a loop's budget has been exceeded. Returns (exceeded: bool, reason: str) - can_continue(loop_id: str) -> bool: Check if a loop can continue based on budget

Token Estimation: - estimate_tokens(text: str) -> int: ~4 chars per token - estimate_llm_call_tokens(prompt: str, response: str) -> int: Estimate LLM call tokens

Usage: 

from src.middleware.budget_tracker import BudgetTracker\n\ntracker = BudgetTracker()\nbudget = tracker.create_budget(\n    loop_id=\"research_loop\",\n    tokens_limit=100000,\n    time_limit_seconds=600,\n    iterations_limit=10\n)\ntracker.start_timer(\"research_loop\")\n# ... research operations ...\ntracker.add_tokens(\"research_loop\", 5000)\ntracker.update_timer(\"research_loop\")\nexceeded, reason = tracker.check_budget(\"research_loop\")\nif exceeded:\n    # Budget exceeded, stop research\n    pass\nif not tracker.can_continue(\"research_loop\"):\n    # Budget exceeded, stop research\n    pass\n

"},{"location":"architecture/middleware/#models","title":"Models","text":"

All middleware models are defined in src/utils/models.py:

"},{"location":"architecture/middleware/#thread-safety","title":"Thread Safety","text":"

All middleware components use ContextVar for thread-safe isolation:

"},{"location":"architecture/middleware/#see-also","title":"See Also","text":""},{"location":"architecture/orchestrators/","title":"Orchestrators Architecture","text":"

DeepCritical supports multiple orchestration patterns for research workflows.

"},{"location":"architecture/orchestrators/#research-flows","title":"Research Flows","text":""},{"location":"architecture/orchestrators/#iterativeresearchflow","title":"IterativeResearchFlow","text":"

File: src/orchestrator/research_flow.py

Pattern: Generate observations \u2192 Evaluate gaps \u2192 Select tools \u2192 Execute \u2192 Judge \u2192 Continue/Complete

Agents Used: - KnowledgeGapAgent: Evaluates research completeness - ToolSelectorAgent: Selects tools for addressing gaps - ThinkingAgent: Generates observations - WriterAgent: Creates final report - JudgeHandler: Assesses evidence sufficiency

Features: - Tracks iterations, time, budget - Supports graph execution (use_graph=True) and agent chains (use_graph=False) - Iterates until research complete or constraints met

Usage:

"},{"location":"architecture/orchestrators/#deepresearchflow","title":"DeepResearchFlow","text":"

File: src/orchestrator/research_flow.py

Pattern: Planner \u2192 Parallel iterative loops per section \u2192 Synthesizer

Agents Used: - PlannerAgent: Breaks query into report sections - IterativeResearchFlow: Per-section research (parallel) - LongWriterAgent or ProofreaderAgent: Final synthesis

Features: - Uses WorkflowManager for parallel execution - Budget tracking per section and globally - State synchronization across parallel loops - Supports graph execution and agent chains

Usage:

"},{"location":"architecture/orchestrators/#graph-orchestrator","title":"Graph Orchestrator","text":"

File: src/orchestrator/graph_orchestrator.py

Purpose: Graph-based execution using Pydantic AI agents as nodes

Features: - Uses graph execution (use_graph=True) or agent chains (use_graph=False) as fallback - Routes based on research mode (iterative/deep/auto) - Streams AgentEvent objects for UI - Uses GraphExecutionContext to manage execution state

Node Types: - Agent Nodes: Execute Pydantic AI agents - State Nodes: Update or read workflow state - Decision Nodes: Make routing decisions - Parallel Nodes: Execute multiple nodes concurrently

Edge Types: - Sequential Edges: Always traversed - Conditional Edges: Traversed based on condition - Parallel Edges: Used for parallel execution branches

Special Node Handling:

The GraphOrchestrator has special handling for certain nodes:

GraphExecutionContext:

The orchestrator uses GraphExecutionContext to manage execution state: - Tracks current node, visited nodes, and node results - Manages workflow state and budget tracker - Provides methods to store and retrieve node execution results

"},{"location":"architecture/orchestrators/#orchestrator-factory","title":"Orchestrator Factory","text":"

File: src/orchestrator_factory.py

Purpose: Factory for creating orchestrators

Modes: - Simple: Legacy orchestrator (backward compatible) - Advanced: Magentic orchestrator (requires OpenAI API key) - Auto-detect: Chooses based on API key availability

Usage:

"},{"location":"architecture/orchestrators/#magentic-orchestrator","title":"Magentic Orchestrator","text":"

File: src/orchestrator_magentic.py

Purpose: Multi-agent coordination using Microsoft Agent Framework

Features: - Uses agent-framework-core - ChatAgent pattern with internal LLMs per agent - MagenticBuilder with participants: - searcher: SearchAgent (wraps SearchHandler) - hypothesizer: HypothesisAgent (generates hypotheses) - judge: JudgeAgent (evaluates evidence) - reporter: ReportAgent (generates final report) - Manager orchestrates agents via chat client (OpenAI or HuggingFace) - Event-driven: converts Magentic events to AgentEvent for UI streaming via _process_event() method - Supports max rounds, stall detection, and reset handling

Event Processing:

The orchestrator processes Magentic events and converts them to AgentEvent: - MagenticOrchestratorMessageEvent \u2192 AgentEvent with type based on message content - MagenticAgentMessageEvent \u2192 AgentEvent with type based on agent name - MagenticAgentDeltaEvent \u2192 AgentEvent for streaming updates - MagenticFinalResultEvent \u2192 AgentEvent with type \"complete\"

Requirements: - agent-framework-core package - OpenAI API key or HuggingFace authentication

"},{"location":"architecture/orchestrators/#hierarchical-orchestrator","title":"Hierarchical Orchestrator","text":"

File: src/orchestrator_hierarchical.py

Purpose: Hierarchical orchestrator using middleware and sub-teams

Features: - Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge - Adapts Magentic ChatAgent to SubIterationTeam protocol - Event-driven via asyncio.Queue for coordination - Supports sub-iteration patterns for complex research tasks

"},{"location":"architecture/orchestrators/#legacy-simple-mode","title":"Legacy Simple Mode","text":"

File: src/legacy_orchestrator.py

Purpose: Linear search-judge-synthesize loop

Features: - Uses SearchHandlerProtocol and JudgeHandlerProtocol - Generator-based design yielding AgentEvent objects - Backward compatibility for simple use cases

"},{"location":"architecture/orchestrators/#state-initialization","title":"State Initialization","text":"

All orchestrators must initialize workflow state:

"},{"location":"architecture/orchestrators/#event-streaming","title":"Event Streaming","text":"

All orchestrators yield AgentEvent objects:

Event Types: - started: Research started - searching: Search in progress - search_complete: Search completed - judging: Evidence evaluation in progress - judge_complete: Evidence evaluation completed - looping: Iteration in progress - hypothesizing: Generating hypotheses - analyzing: Statistical analysis in progress - analysis_complete: Statistical analysis completed - synthesizing: Synthesizing results - complete: Research completed - error: Error occurred - streaming: Streaming update (delta events)

Event Structure:

"},{"location":"architecture/orchestrators/#see-also","title":"See Also","text":""},{"location":"architecture/services/","title":"Services Architecture","text":"

DeepCritical provides several services for embeddings, RAG, and statistical analysis.

"},{"location":"architecture/services/#embedding-service","title":"Embedding Service","text":"

File: src/services/embeddings.py

Purpose: Local sentence-transformers for semantic search and deduplication

Features: - No API Key Required: Uses local sentence-transformers models - Async-Safe: All operations use run_in_executor() to avoid blocking the event loop - ChromaDB Storage: In-memory vector storage for embeddings - Deduplication: 0.9 similarity threshold by default (90% similarity = duplicate, configurable)

Model: Configurable via settings.local_embedding_model (default: all-MiniLM-L6-v2)

Methods: - async def embed(text: str) -> list[float]: Generate embeddings (async-safe via run_in_executor()) - async def embed_batch(texts: list[str]) -> list[list[float]]: Batch embedding (more efficient) - async def add_evidence(evidence_id: str, content: str, metadata: dict[str, Any]) -> None: Add evidence to vector store - async def search_similar(query: str, n_results: int = 5) -> list[dict[str, Any]]: Find semantically similar evidence - async def deduplicate(new_evidence: list[Evidence], threshold: float = 0.9) -> list[Evidence]: Remove semantically duplicate evidence

Usage: 

from src.services.embeddings import get_embedding_service\n\nservice = get_embedding_service()\nembedding = await service.embed(\"text to embed\")\n

"},{"location":"architecture/services/#llamaindex-rag-service","title":"LlamaIndex RAG Service","text":"

File: src/services/llamaindex_rag.py

Purpose: Retrieval-Augmented Generation using LlamaIndex

Features: - Multiple Embedding Providers: OpenAI embeddings (requires OPENAI_API_KEY) or local sentence-transformers (no API key) - Multiple LLM Providers: HuggingFace LLM (preferred) or OpenAI LLM (fallback) for query synthesis - ChromaDB Storage: Vector database for document storage (supports in-memory mode) - Metadata Preservation: Preserves source, title, URL, date, authors - Lazy Initialization: Graceful fallback if dependencies not available

Initialization Parameters: - use_openai_embeddings: bool | None: Force OpenAI embeddings (None = auto-detect) - use_in_memory: bool: Use in-memory ChromaDB client (useful for tests) - oauth_token: str | None: Optional OAuth token from HuggingFace login (takes priority over env vars)

Methods: - async def ingest_evidence(evidence: list[Evidence]) -> None: Ingest evidence into RAG - async def retrieve(query: str, top_k: int = 5) -> list[Document]: Retrieve relevant documents - async def query(query: str, top_k: int = 5) -> str: Query with RAG

Usage: 

from src.services.llamaindex_rag import get_rag_service\n\nservice = get_rag_service(\n    use_openai_embeddings=False,  # Use local embeddings\n    use_in_memory=True,  # Use in-memory ChromaDB\n    oauth_token=token  # Optional HuggingFace token\n)\nif service:\n    documents = await service.retrieve(\"query\", top_k=5)\n

"},{"location":"architecture/services/#statistical-analyzer","title":"Statistical Analyzer","text":"

File: src/services/statistical_analyzer.py

Purpose: Secure execution of AI-generated statistical code

Features: - Modal Sandbox: Secure, isolated execution environment - Code Generation: Generates Python code via LLM - Library Pinning: Version-pinned libraries in SANDBOX_LIBRARIES - Network Isolation: block_network=True by default

Libraries Available: - pandas, numpy, scipy - matplotlib, scikit-learn - statsmodels

Output: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - code: Generated analysis code - output: Execution output - error: Error message if execution failed

Usage: 

from src.services.statistical_analyzer import StatisticalAnalyzer\n\nanalyzer = StatisticalAnalyzer()\nresult = await analyzer.analyze(\n    hypothesis=\"Metformin reduces cancer risk\",\n    evidence=evidence_list\n)\n

"},{"location":"architecture/services/#singleton-pattern","title":"Singleton Pattern","text":"

Services use singleton patterns for lazy initialization:

EmbeddingService: Uses a global variable pattern:

LlamaIndexRAGService: Direct instantiation (no caching):

This ensures: - Single instance per process - Lazy initialization - No dependencies required at import time

"},{"location":"architecture/services/#service-availability","title":"Service Availability","text":"

Services check availability before use:

from src.utils.config import settings\n\nif settings.modal_available:\n    # Use Modal sandbox\n    pass\n\nif settings.has_openai_key:\n    # Use OpenAI embeddings for RAG\n    pass\n
"},{"location":"architecture/services/#see-also","title":"See Also","text":""},{"location":"architecture/tools/","title":"Tools Architecture","text":"

DeepCritical implements a protocol-based search tool system for retrieving evidence from multiple sources.

"},{"location":"architecture/tools/#searchtool-protocol","title":"SearchTool Protocol","text":"

All tools implement the SearchTool protocol from src/tools/base.py:

"},{"location":"architecture/tools/#rate-limiting","title":"Rate Limiting","text":"

All tools use the @retry decorator from tenacity:

Tools with API rate limits implement _rate_limit() method and use shared rate limiters from src/tools/rate_limiter.py.

"},{"location":"architecture/tools/#error-handling","title":"Error Handling","text":"

Tools raise custom exceptions:

Tools handle HTTP errors (429, 500, timeout) and return empty lists on non-critical errors (with warning logs).

"},{"location":"architecture/tools/#query-preprocessing","title":"Query Preprocessing","text":"

Tools use preprocess_query() from src/tools/query_utils.py to:

"},{"location":"architecture/tools/#evidence-conversion","title":"Evidence Conversion","text":"

All tools convert API responses to Evidence objects with:

Missing fields are handled gracefully with defaults.

"},{"location":"architecture/tools/#tool-implementations","title":"Tool Implementations","text":""},{"location":"architecture/tools/#pubmed-tool","title":"PubMed Tool","text":"

File: src/tools/pubmed.py

API: NCBI E-utilities (ESearch \u2192 EFetch)

Rate Limiting: - 0.34s between requests (3 req/sec without API key) - 0.1s between requests (10 req/sec with NCBI API key)

Features: - XML parsing with xmltodict - Handles single vs. multiple articles - Query preprocessing - Evidence conversion with metadata extraction

"},{"location":"architecture/tools/#clinicaltrials-tool","title":"ClinicalTrials Tool","text":"

File: src/tools/clinicaltrials.py

API: ClinicalTrials.gov API v2

Important: Uses requests library (NOT httpx) because WAF blocks httpx TLS fingerprint.

Execution: Runs in thread pool: await asyncio.to_thread(requests.get, ...)

Filtering: - Only interventional studies - Status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION

Features: - Parses nested JSON structure - Extracts trial metadata - Evidence conversion

"},{"location":"architecture/tools/#europe-pmc-tool","title":"Europe PMC Tool","text":"

File: src/tools/europepmc.py

API: Europe PMC REST API

Features: - Handles preprint markers: [PREPRINT - Not peer-reviewed] - Builds URLs from DOI or PMID - Checks pubTypeList for preprint detection - Includes both preprints and peer-reviewed articles

"},{"location":"architecture/tools/#rag-tool","title":"RAG Tool","text":"

File: src/tools/rag_tool.py

Purpose: Semantic search within collected evidence

Implementation: Wraps LlamaIndexRAGService

Features: - Returns Evidence from RAG results - Handles evidence ingestion - Semantic similarity search - Metadata preservation

"},{"location":"architecture/tools/#search-handler","title":"Search Handler","text":"

File: src/tools/search_handler.py

Purpose: Orchestrates parallel searches across multiple tools

Initialization Parameters: - tools: list[SearchTool]: List of search tools to use - timeout: float = 30.0: Timeout for each search in seconds - include_rag: bool = False: Whether to include RAG tool in searches - auto_ingest_to_rag: bool = True: Whether to automatically ingest results into RAG - oauth_token: str | None = None: Optional OAuth token from HuggingFace login (for RAG LLM)

Methods: - async def execute(query: str, max_results_per_tool: int = 10) -> SearchResult: Execute search across all tools in parallel

Features: - Uses asyncio.gather() with return_exceptions=True for parallel execution - Aggregates results into SearchResult with evidence and metadata - Handles tool failures gracefully (continues with other tools) - Deduplicates results by URL - Automatically ingests results into RAG if auto_ingest_to_rag=True - Can add RAG tool dynamically via add_rag_tool() method

"},{"location":"architecture/tools/#tool-registration","title":"Tool Registration","text":"

Tools are registered in the search handler:

from src.tools.pubmed import PubMedTool\nfrom src.tools.clinicaltrials import ClinicalTrialsTool\nfrom src.tools.europepmc import EuropePMCTool\nfrom src.tools.search_handler import SearchHandler\n\nsearch_handler = SearchHandler(\n    tools=[\n        PubMedTool(),\n        ClinicalTrialsTool(),\n        EuropePMCTool(),\n    ],\n    include_rag=True,  # Include RAG tool for semantic search\n    auto_ingest_to_rag=True,  # Automatically ingest results into RAG\n    oauth_token=token  # Optional HuggingFace token for RAG LLM\n)\n\n# Execute search\nresult = await search_handler.execute(\"query\", max_results_per_tool=10)\n
"},{"location":"architecture/tools/#see-also","title":"See Also","text":""},{"location":"architecture/workflow-diagrams/","title":"DeepCritical Workflow - Simplified Magentic Architecture","text":"

Architecture Pattern: Microsoft Magentic Orchestration Design Philosophy: Simple, dynamic, manager-driven coordination Key Innovation: Intelligent manager replaces rigid sequential phases

"},{"location":"architecture/workflow-diagrams/#1-high-level-magentic-workflow","title":"1. High-Level Magentic Workflow","text":"
flowchart TD\n    Start([User Query]) --> Manager[Magentic Manager<br/>Plan \u2022 Select \u2022 Assess \u2022 Adapt]\n\n    Manager -->|Plans| Task1[Task Decomposition]\n    Task1 --> Manager\n\n    Manager -->|Selects & Executes| HypAgent[Hypothesis Agent]\n    Manager -->|Selects & Executes| SearchAgent[Search Agent]\n    Manager -->|Selects & Executes| AnalysisAgent[Analysis Agent]\n    Manager -->|Selects & Executes| ReportAgent[Report Agent]\n\n    HypAgent -->|Results| Manager\n    SearchAgent -->|Results| Manager\n    AnalysisAgent -->|Results| Manager\n    ReportAgent -->|Results| Manager\n\n    Manager -->|Assesses Quality| Decision{Good Enough?}\n    Decision -->|No - Refine| Manager\n    Decision -->|No - Different Agent| Manager\n    Decision -->|No - Stalled| Replan[Reset Plan]\n    Replan --> Manager\n\n    Decision -->|Yes| Synthesis[Synthesize Final Result]\n    Synthesis --> Output([Research Report])\n\n    style Start fill:#e1f5e1\n    style Manager fill:#ffe6e6\n    style HypAgent fill:#fff4e6\n    style SearchAgent fill:#fff4e6\n    style AnalysisAgent fill:#fff4e6\n    style ReportAgent fill:#fff4e6\n    style Decision fill:#ffd6d6\n    style Synthesis fill:#d4edda\n    style Output fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#2-magentic-manager-the-6-phase-cycle","title":"2. Magentic Manager: The 6-Phase Cycle","text":"
flowchart LR\n    P1[1. Planning<br/>Analyze task<br/>Create strategy] --> P2[2. Agent Selection<br/>Pick best agent<br/>for subtask]\n    P2 --> P3[3. Execution<br/>Run selected<br/>agent with tools]\n    P3 --> P4[4. Assessment<br/>Evaluate quality<br/>Check progress]\n    P4 --> Decision{Quality OK?<br/>Progress made?}\n    Decision -->|Yes| P6[6. Synthesis<br/>Combine results<br/>Generate report]\n    Decision -->|No| P5[5. Iteration<br/>Adjust plan<br/>Try again]\n    P5 --> P2\n    P6 --> Done([Complete])\n\n    style P1 fill:#fff4e6\n    style P2 fill:#ffe6e6\n    style P3 fill:#e6f3ff\n    style P4 fill:#ffd6d6\n    style P5 fill:#fff3cd\n    style P6 fill:#d4edda\n    style Done fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#3-simplified-agent-architecture","title":"3. Simplified Agent Architecture","text":"
graph TB\n    subgraph \"Orchestration Layer\"\n        Manager[Magentic Manager<br/>\u2022 Plans workflow<br/>\u2022 Selects agents<br/>\u2022 Assesses quality<br/>\u2022 Adapts strategy]\n        SharedContext[(Shared Context<br/>\u2022 Hypotheses<br/>\u2022 Search Results<br/>\u2022 Analysis<br/>\u2022 Progress)]\n        Manager <--> SharedContext\n    end\n\n    subgraph \"Specialist Agents\"\n        HypAgent[Hypothesis Agent<br/>\u2022 Domain understanding<br/>\u2022 Hypothesis generation<br/>\u2022 Testability refinement]\n        SearchAgent[Search Agent<br/>\u2022 Multi-source search<br/>\u2022 RAG retrieval<br/>\u2022 Result ranking]\n        AnalysisAgent[Analysis Agent<br/>\u2022 Evidence extraction<br/>\u2022 Statistical analysis<br/>\u2022 Code execution]\n        ReportAgent[Report Agent<br/>\u2022 Report assembly<br/>\u2022 Visualization<br/>\u2022 Citation formatting]\n    end\n\n    subgraph \"MCP Tools\"\n        WebSearch[Web Search<br/>PubMed \u2022 arXiv \u2022 bioRxiv]\n        CodeExec[Code Execution<br/>Sandboxed Python]\n        RAG[RAG Retrieval<br/>Vector DB \u2022 Embeddings]\n        Viz[Visualization<br/>Charts \u2022 Graphs]\n    end\n\n    Manager -->|Selects & Directs| HypAgent\n    Manager -->|Selects & Directs| SearchAgent\n    Manager -->|Selects & Directs| AnalysisAgent\n    Manager -->|Selects & Directs| ReportAgent\n\n    HypAgent --> SharedContext\n    SearchAgent --> SharedContext\n    AnalysisAgent --> SharedContext\n    ReportAgent --> SharedContext\n\n    SearchAgent --> WebSearch\n    SearchAgent --> RAG\n    AnalysisAgent --> CodeExec\n    ReportAgent --> CodeExec\n    ReportAgent --> Viz\n\n    style Manager fill:#ffe6e6\n    style SharedContext fill:#ffe6f0\n    style HypAgent fill:#fff4e6\n    style SearchAgent fill:#fff4e6\n    style AnalysisAgent fill:#fff4e6\n    style ReportAgent fill:#fff4e6\n    style WebSearch fill:#e6f3ff\n    style CodeExec fill:#e6f3ff\n    style RAG fill:#e6f3ff\n    style Viz fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#4-dynamic-workflow-example","title":"4. Dynamic Workflow Example","text":"
sequenceDiagram\n    participant User\n    participant Manager\n    participant HypAgent\n    participant SearchAgent\n    participant AnalysisAgent\n    participant ReportAgent\n\n    User->>Manager: \"Research protein folding in Alzheimer's\"\n\n    Note over Manager: PLAN: Generate hypotheses \u2192 Search \u2192 Analyze \u2192 Report\n\n    Manager->>HypAgent: Generate 3 hypotheses\n    HypAgent-->>Manager: Returns 3 hypotheses\n    Note over Manager: ASSESS: Good quality, proceed\n\n    Manager->>SearchAgent: Search literature for hypothesis 1\n    SearchAgent-->>Manager: Returns 15 papers\n    Note over Manager: ASSESS: Good results, continue\n\n    Manager->>SearchAgent: Search for hypothesis 2\n    SearchAgent-->>Manager: Only 2 papers found\n    Note over Manager: ASSESS: Insufficient, refine search\n\n    Manager->>SearchAgent: Refined query for hypothesis 2\n    SearchAgent-->>Manager: Returns 12 papers\n    Note over Manager: ASSESS: Better, proceed\n\n    Manager->>AnalysisAgent: Analyze evidence for all hypotheses\n    AnalysisAgent-->>Manager: Returns analysis with code\n    Note over Manager: ASSESS: Complete, generate report\n\n    Manager->>ReportAgent: Create comprehensive report\n    ReportAgent-->>Manager: Returns formatted report\n    Note over Manager: SYNTHESIZE: Combine all results\n\n    Manager->>User: Final Research Report
"},{"location":"architecture/workflow-diagrams/#5-manager-decision-logic","title":"5. Manager Decision Logic","text":"
flowchart TD\n    Start([Manager Receives Task]) --> Plan[Create Initial Plan]\n\n    Plan --> Select[Select Agent for Next Subtask]\n    Select --> Execute[Execute Agent]\n    Execute --> Collect[Collect Results]\n\n    Collect --> Assess[Assess Quality & Progress]\n\n    Assess --> Q1{Quality Sufficient?}\n    Q1 -->|No| Q2{Same Agent Can Fix?}\n    Q2 -->|Yes| Feedback[Provide Specific Feedback]\n    Feedback --> Execute\n    Q2 -->|No| Different[Try Different Agent]\n    Different --> Select\n\n    Q1 -->|Yes| Q3{Task Complete?}\n    Q3 -->|No| Q4{Making Progress?}\n    Q4 -->|Yes| Select\n    Q4 -->|No - Stalled| Replan[Reset Plan & Approach]\n    Replan --> Plan\n\n    Q3 -->|Yes| Synth[Synthesize Final Result]\n    Synth --> Done([Return Report])\n\n    style Start fill:#e1f5e1\n    style Plan fill:#fff4e6\n    style Select fill:#ffe6e6\n    style Execute fill:#e6f3ff\n    style Assess fill:#ffd6d6\n    style Q1 fill:#ffe6e6\n    style Q2 fill:#ffe6e6\n    style Q3 fill:#ffe6e6\n    style Q4 fill:#ffe6e6\n    style Synth fill:#d4edda\n    style Done fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#6-hypothesis-agent-workflow","title":"6. Hypothesis Agent Workflow","text":"
flowchart LR\n    Input[Research Query] --> Domain[Identify Domain<br/>& Key Concepts]\n    Domain --> Context[Retrieve Background<br/>Knowledge]\n    Context --> Generate[Generate 3-5<br/>Initial Hypotheses]\n    Generate --> Refine[Refine for<br/>Testability]\n    Refine --> Rank[Rank by<br/>Quality Score]\n    Rank --> Output[Return Top<br/>Hypotheses]\n\n    Output --> Struct[Hypothesis Structure:<br/>\u2022 Statement<br/>\u2022 Rationale<br/>\u2022 Testability Score<br/>\u2022 Data Requirements<br/>\u2022 Expected Outcomes]\n\n    style Input fill:#e1f5e1\n    style Output fill:#fff4e6\n    style Struct fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#7-search-agent-workflow","title":"7. Search Agent Workflow","text":"
flowchart TD\n    Input[Hypotheses] --> Strategy[Formulate Search<br/>Strategy per Hypothesis]\n\n    Strategy --> Multi[Multi-Source Search]\n\n    Multi --> PubMed[PubMed Search<br/>via MCP]\n    Multi --> ArXiv[arXiv Search<br/>via MCP]\n    Multi --> BioRxiv[bioRxiv Search<br/>via MCP]\n\n    PubMed --> Aggregate[Aggregate Results]\n    ArXiv --> Aggregate\n    BioRxiv --> Aggregate\n\n    Aggregate --> Filter[Filter & Rank<br/>by Relevance]\n    Filter --> Dedup[Deduplicate<br/>Cross-Reference]\n    Dedup --> Embed[Embed Documents<br/>via MCP]\n    Embed --> Vector[(Vector DB)]\n    Vector --> RAGRetrieval[RAG Retrieval<br/>Top-K per Hypothesis]\n    RAGRetrieval --> Output[Return Contextualized<br/>Search Results]\n\n    style Input fill:#fff4e6\n    style Multi fill:#ffe6e6\n    style Vector fill:#ffe6f0\n    style Output fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#8-analysis-agent-workflow","title":"8. Analysis Agent Workflow","text":"
flowchart TD\n    Input1[Hypotheses] --> Extract\n    Input2[Search Results] --> Extract[Extract Evidence<br/>per Hypothesis]\n\n    Extract --> Methods[Determine Analysis<br/>Methods Needed]\n\n    Methods --> Branch{Requires<br/>Computation?}\n    Branch -->|Yes| GenCode[Generate Python<br/>Analysis Code]\n    Branch -->|No| Qual[Qualitative<br/>Synthesis]\n\n    GenCode --> Execute[Execute Code<br/>via MCP Sandbox]\n    Execute --> Interpret1[Interpret<br/>Results]\n    Qual --> Interpret2[Interpret<br/>Findings]\n\n    Interpret1 --> Synthesize[Synthesize Evidence<br/>Across Sources]\n    Interpret2 --> Synthesize\n\n    Synthesize --> Verdict[Determine Verdict<br/>per Hypothesis]\n    Verdict --> Support[\u2022 Supported<br/>\u2022 Refuted<br/>\u2022 Inconclusive]\n    Support --> Gaps[Identify Knowledge<br/>Gaps & Limitations]\n    Gaps --> Output[Return Analysis<br/>Report]\n\n    style Input1 fill:#fff4e6\n    style Input2 fill:#e6f3ff\n    style Execute fill:#ffe6e6\n    style Output fill:#e6ffe6
"},{"location":"architecture/workflow-diagrams/#9-report-agent-workflow","title":"9. Report Agent Workflow","text":"
flowchart TD\n    Input1[Query] --> Assemble\n    Input2[Hypotheses] --> Assemble\n    Input3[Search Results] --> Assemble\n    Input4[Analysis] --> Assemble[Assemble Report<br/>Sections]\n\n    Assemble --> Exec[Executive Summary]\n    Assemble --> Intro[Introduction]\n    Assemble --> Methods[Methods]\n    Assemble --> Results[Results per<br/>Hypothesis]\n    Assemble --> Discussion[Discussion]\n    Assemble --> Future[Future Directions]\n    Assemble --> Refs[References]\n\n    Results --> VizCheck{Needs<br/>Visualization?}\n    VizCheck -->|Yes| GenViz[Generate Viz Code]\n    GenViz --> ExecViz[Execute via MCP<br/>Create Charts]\n    ExecViz --> Combine\n    VizCheck -->|No| Combine[Combine All<br/>Sections]\n\n    Exec --> Combine\n    Intro --> Combine\n    Methods --> Combine\n    Discussion --> Combine\n    Future --> Combine\n    Refs --> Combine\n\n    Combine --> Format[Format Output]\n    Format --> MD[Markdown]\n    Format --> PDF[PDF]\n    Format --> JSON[JSON]\n\n    MD --> Output[Return Final<br/>Report]\n    PDF --> Output\n    JSON --> Output\n\n    style Input1 fill:#e1f5e1\n    style Input2 fill:#fff4e6\n    style Input3 fill:#e6f3ff\n    style Input4 fill:#e6ffe6\n    style Output fill:#d4edda
"},{"location":"architecture/workflow-diagrams/#10-data-flow--event-streaming","title":"10. Data Flow & Event Streaming","text":"
flowchart TD\n    User[\ud83d\udc64 User] -->|Research Query| UI[Gradio UI]\n    UI -->|Submit| Manager[Magentic Manager]\n\n    Manager -->|Event: Planning| UI\n    Manager -->|Select Agent| HypAgent[Hypothesis Agent]\n    HypAgent -->|Event: Delta/Message| UI\n    HypAgent -->|Hypotheses| Context[(Shared Context)]\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| SearchAgent[Search Agent]\n    SearchAgent -->|MCP Request| WebSearch[Web Search Tool]\n    WebSearch -->|Results| SearchAgent\n    SearchAgent -->|Event: Delta/Message| UI\n    SearchAgent -->|Documents| Context\n    SearchAgent -->|Embeddings| VectorDB[(Vector DB)]\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| AnalysisAgent[Analysis Agent]\n    AnalysisAgent -->|MCP Request| CodeExec[Code Execution Tool]\n    CodeExec -->|Results| AnalysisAgent\n    AnalysisAgent -->|Event: Delta/Message| UI\n    AnalysisAgent -->|Analysis| Context\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| ReportAgent[Report Agent]\n    ReportAgent -->|MCP Request| CodeExec\n    ReportAgent -->|Event: Delta/Message| UI\n    ReportAgent -->|Report| Context\n\n    Manager -->|Event: Final Result| UI\n    UI -->|Display| User\n\n    style User fill:#e1f5e1\n    style UI fill:#e6f3ff\n    style Manager fill:#ffe6e6\n    style Context fill:#ffe6f0\n    style VectorDB fill:#ffe6f0\n    style WebSearch fill:#f0f0f0\n    style CodeExec fill:#f0f0f0
"},{"location":"architecture/workflow-diagrams/#11-mcp-tool-architecture","title":"11. MCP Tool Architecture","text":"
graph TB\n    subgraph \"Agent Layer\"\n        Manager[Magentic Manager]\n        HypAgent[Hypothesis Agent]\n        SearchAgent[Search Agent]\n        AnalysisAgent[Analysis Agent]\n        ReportAgent[Report Agent]\n    end\n\n    subgraph \"MCP Protocol Layer\"\n        Registry[MCP Tool Registry<br/>\u2022 Discovers tools<br/>\u2022 Routes requests<br/>\u2022 Manages connections]\n    end\n\n    subgraph \"MCP Servers\"\n        Server1[Web Search Server<br/>localhost:8001<br/>\u2022 PubMed<br/>\u2022 arXiv<br/>\u2022 bioRxiv]\n        Server2[Code Execution Server<br/>localhost:8002<br/>\u2022 Sandboxed Python<br/>\u2022 Package management]\n        Server3[RAG Server<br/>localhost:8003<br/>\u2022 Vector embeddings<br/>\u2022 Similarity search]\n        Server4[Visualization Server<br/>localhost:8004<br/>\u2022 Chart generation<br/>\u2022 Plot rendering]\n    end\n\n    subgraph \"External Services\"\n        PubMed[PubMed API]\n        ArXiv[arXiv API]\n        BioRxiv[bioRxiv API]\n        Modal[Modal Sandbox]\n        ChromaDB[(ChromaDB)]\n    end\n\n    SearchAgent -->|Request| Registry\n    AnalysisAgent -->|Request| Registry\n    ReportAgent -->|Request| Registry\n\n    Registry --> Server1\n    Registry --> Server2\n    Registry --> Server3\n    Registry --> Server4\n\n    Server1 --> PubMed\n    Server1 --> ArXiv\n    Server1 --> BioRxiv\n    Server2 --> Modal\n    Server3 --> ChromaDB\n\n    style Manager fill:#ffe6e6\n    style Registry fill:#fff4e6\n    style Server1 fill:#e6f3ff\n    style Server2 fill:#e6f3ff\n    style Server3 fill:#e6f3ff\n    style Server4 fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#12-progress-tracking--stall-detection","title":"12. Progress Tracking & Stall Detection","text":"
stateDiagram-v2\n    [*] --> Initialization: User Query\n\n    Initialization --> Planning: Manager starts\n\n    Planning --> AgentExecution: Select agent\n\n    AgentExecution --> Assessment: Collect results\n\n    Assessment --> QualityCheck: Evaluate output\n\n    QualityCheck --> AgentExecution: Poor quality<br/>(retry < max_rounds)\n    QualityCheck --> Planning: Poor quality<br/>(try different agent)\n    QualityCheck --> NextAgent: Good quality<br/>(task incomplete)\n    QualityCheck --> Synthesis: Good quality<br/>(task complete)\n\n    NextAgent --> AgentExecution: Select next agent\n\n    state StallDetection <<choice>>\n    Assessment --> StallDetection: Check progress\n    StallDetection --> Planning: No progress<br/>(stall count < max)\n    StallDetection --> ErrorRecovery: No progress<br/>(max stalls reached)\n\n    ErrorRecovery --> PartialReport: Generate partial results\n    PartialReport --> [*]\n\n    Synthesis --> FinalReport: Combine all outputs\n    FinalReport --> [*]\n\n    note right of QualityCheck\n        Manager assesses:\n        \u2022 Output completeness\n        \u2022 Quality metrics\n        \u2022 Progress made\n    end note\n\n    note right of StallDetection\n        Stall = no new progress\n        after agent execution\n        Triggers plan reset\n    end note
"},{"location":"architecture/workflow-diagrams/#13-gradio-ui-integration","title":"13. Gradio UI Integration","text":"
graph TD\n    App[Gradio App<br/>DeepCritical Research Agent]\n\n    App --> Input[Input Section]\n    App --> Status[Status Section]\n    App --> Output[Output Section]\n\n    Input --> Query[Research Question<br/>Text Area]\n    Input --> Controls[Controls]\n    Controls --> MaxHyp[Max Hypotheses: 1-10]\n    Controls --> MaxRounds[Max Rounds: 5-20]\n    Controls --> Submit[Start Research Button]\n\n    Status --> Log[Real-time Event Log<br/>\u2022 Manager planning<br/>\u2022 Agent selection<br/>\u2022 Execution updates<br/>\u2022 Quality assessment]\n    Status --> Progress[Progress Tracker<br/>\u2022 Current agent<br/>\u2022 Round count<br/>\u2022 Stall count]\n\n    Output --> Tabs[Tabbed Results]\n    Tabs --> Tab1[Hypotheses Tab<br/>Generated hypotheses with scores]\n    Tabs --> Tab2[Search Results Tab<br/>Papers & sources found]\n    Tabs --> Tab3[Analysis Tab<br/>Evidence & verdicts]\n    Tabs --> Tab4[Report Tab<br/>Final research report]\n    Tab4 --> Download[Download Report<br/>MD / PDF / JSON]\n\n    Submit -.->|Triggers| Workflow[Magentic Workflow]\n    Workflow -.->|MagenticOrchestratorMessageEvent| Log\n    Workflow -.->|MagenticAgentDeltaEvent| Log\n    Workflow -.->|MagenticAgentMessageEvent| Log\n    Workflow -.->|MagenticFinalResultEvent| Tab4\n\n    style App fill:#e1f5e1\n    style Input fill:#fff4e6\n    style Status fill:#e6f3ff\n    style Output fill:#e6ffe6\n    style Workflow fill:#ffe6e6
"},{"location":"architecture/workflow-diagrams/#14-complete-system-context","title":"14. Complete System Context","text":"
graph LR\n    User[\ud83d\udc64 Researcher<br/>Asks research questions] -->|Submits query| DC[DeepCritical<br/>Magentic Workflow]\n\n    DC -->|Literature search| PubMed[PubMed API<br/>Medical papers]\n    DC -->|Preprint search| ArXiv[arXiv API<br/>Scientific preprints]\n    DC -->|Biology search| BioRxiv[bioRxiv API<br/>Biology preprints]\n    DC -->|Agent reasoning| Claude[Claude API<br/>Sonnet 4 / Opus]\n    DC -->|Code execution| Modal[Modal Sandbox<br/>Safe Python env]\n    DC -->|Vector storage| Chroma[ChromaDB<br/>Embeddings & RAG]\n\n    DC -->|Deployed on| HF[HuggingFace Spaces<br/>Gradio 6.0]\n\n    PubMed -->|Results| DC\n    ArXiv -->|Results| DC\n    BioRxiv -->|Results| DC\n    Claude -->|Responses| DC\n    Modal -->|Output| DC\n    Chroma -->|Context| DC\n\n    DC -->|Research report| User\n\n    style User fill:#e1f5e1\n    style DC fill:#ffe6e6\n    style PubMed fill:#e6f3ff\n    style ArXiv fill:#e6f3ff\n    style BioRxiv fill:#e6f3ff\n    style Claude fill:#ffd6d6\n    style Modal fill:#f0f0f0\n    style Chroma fill:#ffe6f0\n    style HF fill:#d4edda
"},{"location":"architecture/workflow-diagrams/#15-workflow-timeline-simplified","title":"15. Workflow Timeline (Simplified)","text":"
gantt\n    title DeepCritical Magentic Workflow - Typical Execution\n    dateFormat mm:ss\n    axisFormat %M:%S\n\n    section Manager Planning\n    Initial planning         :p1, 00:00, 10s\n\n    section Hypothesis Agent\n    Generate hypotheses      :h1, after p1, 30s\n    Manager assessment       :h2, after h1, 5s\n\n    section Search Agent\n    Search hypothesis 1      :s1, after h2, 20s\n    Search hypothesis 2      :s2, after s1, 20s\n    Search hypothesis 3      :s3, after s2, 20s\n    RAG processing          :s4, after s3, 15s\n    Manager assessment      :s5, after s4, 5s\n\n    section Analysis Agent\n    Evidence extraction     :a1, after s5, 15s\n    Code generation        :a2, after a1, 20s\n    Code execution         :a3, after a2, 25s\n    Synthesis              :a4, after a3, 20s\n    Manager assessment     :a5, after a4, 5s\n\n    section Report Agent\n    Report assembly        :r1, after a5, 30s\n    Visualization          :r2, after r1, 15s\n    Formatting             :r3, after r2, 10s\n\n    section Manager Synthesis\n    Final synthesis        :f1, after r3, 10s
"},{"location":"architecture/workflow-diagrams/#key-differences-from-original-design","title":"Key Differences from Original Design","text":"Aspect Original (Judge-in-Loop) New (Magentic) Control Flow Fixed sequential phases Dynamic agent selection Quality Control Separate Judge Agent Manager assessment built-in Retry Logic Phase-level with feedback Agent-level with adaptation Flexibility Rigid 4-phase pipeline Adaptive workflow Complexity 5 agents (including Judge) 4 agents (no Judge) Progress Tracking Manual state management Built-in round/stall detection Agent Coordination Sequential handoff Manager-driven dynamic selection Error Recovery Retry same phase Try different agent or replan"},{"location":"architecture/workflow-diagrams/#simplified-design-principles","title":"Simplified Design Principles","text":"
  1. Manager is Intelligent: LLM-powered manager handles planning, selection, and quality assessment
  2. No Separate Judge: Manager's assessment phase replaces dedicated Judge Agent
  3. Dynamic Workflow: Agents can be called multiple times in any order based on need
  4. Built-in Safety: max_round_count (15) and max_stall_count (3) prevent infinite loops
  5. Event-Driven UI: Real-time streaming updates to Gradio interface
  6. MCP-Powered Tools: All external capabilities via Model Context Protocol
  7. Shared Context: Centralized state accessible to all agents
  8. Progress Awareness: Manager tracks what's been done and what's needed
"},{"location":"architecture/workflow-diagrams/#legend","title":"Legend","text":""},{"location":"architecture/workflow-diagrams/#implementation-highlights","title":"Implementation Highlights","text":"

Simple 4-Agent Setup:

Manager handles quality assessment in its instructions: - Checks hypothesis quality (testable, novel, clear) - Validates search results (relevant, authoritative, recent) - Assesses analysis soundness (methodology, evidence, conclusions) - Ensures report completeness (all sections, proper citations)

No separate Judge Agent needed - manager does it all!

Document Version: 2.0 (Magentic Simplified) Last Updated: 2025-11-24 Architecture: Microsoft Magentic Orchestration Pattern Agents: 4 (Hypothesis, Search, Analysis, Report) + 1 Manager License: MIT

"},{"location":"architecture/workflow-diagrams/#see-also","title":"See Also","text":""},{"location":"configuration/","title":"Configuration Guide","text":""},{"location":"configuration/#overview","title":"Overview","text":"

DeepCritical uses Pydantic Settings for centralized configuration management. All settings are defined in the Settings class in src/utils/config.py and can be configured via environment variables or a .env file.

The configuration system provides:

"},{"location":"configuration/#quick-start","title":"Quick Start","text":"
  1. Create a .env file in the project root
  2. Set at least one LLM API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, or HF_TOKEN)
  3. Optionally configure other services as needed
  4. The application will automatically load and validate your configuration
"},{"location":"configuration/#configuration-system-architecture","title":"Configuration System Architecture","text":""},{"location":"configuration/#settings-class","title":"Settings Class","text":"

The [Settings][settings-class] class extends BaseSettings from pydantic_settings and defines all application configuration:

View source

"},{"location":"configuration/#singleton-instance","title":"Singleton Instance","text":"

A global settings instance is available for import:

View source

"},{"location":"configuration/#usage-pattern","title":"Usage Pattern","text":"

Access configuration throughout the codebase:

from src.utils.config import settings\n\n# Check if API keys are available\nif settings.has_openai_key:\n    # Use OpenAI\n    pass\n\n# Access configuration values\nmax_iterations = settings.max_iterations\nweb_search_provider = settings.web_search_provider\n
"},{"location":"configuration/#required-configuration","title":"Required Configuration","text":""},{"location":"configuration/#llm-provider","title":"LLM Provider","text":"

You must configure at least one LLM provider. The system supports:

"},{"location":"configuration/#openai-configuration","title":"OpenAI Configuration","text":"
LLM_PROVIDER=openai\nOPENAI_API_KEY=your_openai_api_key_here\nOPENAI_MODEL=gpt-5.1\n

The default model is defined in the Settings class:

"},{"location":"configuration/#anthropic-configuration","title":"Anthropic Configuration","text":"
LLM_PROVIDER=anthropic\nANTHROPIC_API_KEY=your_anthropic_api_key_here\nANTHROPIC_MODEL=claude-sonnet-4-5-20250929\n

The default model is defined in the Settings class:

"},{"location":"configuration/#huggingface-configuration","title":"HuggingFace Configuration","text":"

HuggingFace can work without an API key for public models, but an API key provides higher rate limits:

# Option 1: Using HF_TOKEN (preferred)\nHF_TOKEN=your_huggingface_token_here\n\n# Option 2: Using HUGGINGFACE_API_KEY (alternative)\nHUGGINGFACE_API_KEY=your_huggingface_api_key_here\n\n# Default model\nHUGGINGFACE_MODEL=meta-llama/Llama-3.1-8B-Instruct\n

The HuggingFace token can be set via either environment variable:

"},{"location":"configuration/#optional-configuration","title":"Optional Configuration","text":""},{"location":"configuration/#embedding-configuration","title":"Embedding Configuration","text":"

DeepCritical supports multiple embedding providers for semantic search and RAG:

# Embedding Provider: \"openai\", \"local\", or \"huggingface\"\nEMBEDDING_PROVIDER=local\n\n# OpenAI Embedding Model (used by LlamaIndex RAG)\nOPENAI_EMBEDDING_MODEL=text-embedding-3-small\n\n# Local Embedding Model (sentence-transformers, used by EmbeddingService)\nLOCAL_EMBEDDING_MODEL=all-MiniLM-L6-v2\n\n# HuggingFace Embedding Model\nHUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2\n

The embedding provider configuration:

Note: OpenAI embeddings require OPENAI_API_KEY. The local provider (default) uses sentence-transformers and requires no API key.

"},{"location":"configuration/#web-search-configuration","title":"Web Search Configuration","text":"

DeepCritical supports multiple web search providers:

# Web Search Provider: \"serper\", \"searchxng\", \"brave\", \"tavily\", or \"duckduckgo\"\n# Default: \"duckduckgo\" (no API key required)\nWEB_SEARCH_PROVIDER=duckduckgo\n\n# Serper API Key (for Google search via Serper)\nSERPER_API_KEY=your_serper_api_key_here\n\n# SearchXNG Host URL (for self-hosted search)\nSEARCHXNG_HOST=http://localhost:8080\n\n# Brave Search API Key\nBRAVE_API_KEY=your_brave_api_key_here\n\n# Tavily API Key\nTAVILY_API_KEY=your_tavily_api_key_here\n

The web search provider configuration:

Note: DuckDuckGo is the default and requires no API key, making it ideal for development and testing.

"},{"location":"configuration/#pubmed-configuration","title":"PubMed Configuration","text":"

PubMed search supports optional NCBI API key for higher rate limits:

# NCBI API Key (optional, for higher rate limits: 10 req/sec vs 3 req/sec)\nNCBI_API_KEY=your_ncbi_api_key_here\n

The PubMed tool uses this configuration:

"},{"location":"configuration/#agent-configuration","title":"Agent Configuration","text":"

Control agent behavior and research loop execution:

# Maximum iterations per research loop (1-50, default: 10)\nMAX_ITERATIONS=10\n\n# Search timeout in seconds\nSEARCH_TIMEOUT=30\n\n# Use graph-based execution for research flows\nUSE_GRAPH_EXECUTION=false\n

The agent configuration fields:

"},{"location":"configuration/#budget--rate-limiting-configuration","title":"Budget & Rate Limiting Configuration","text":"

Control resource limits for research loops:

# Default token budget per research loop (1000-1000000, default: 100000)\nDEFAULT_TOKEN_LIMIT=100000\n\n# Default time limit per research loop in minutes (1-120, default: 10)\nDEFAULT_TIME_LIMIT_MINUTES=10\n\n# Default iterations limit per research loop (1-50, default: 10)\nDEFAULT_ITERATIONS_LIMIT=10\n

The budget configuration with validation:

"},{"location":"configuration/#rag-service-configuration","title":"RAG Service Configuration","text":"

Configure the Retrieval-Augmented Generation service:

# ChromaDB collection name for RAG\nRAG_COLLECTION_NAME=deepcritical_evidence\n\n# Number of top results to retrieve from RAG (1-50, default: 5)\nRAG_SIMILARITY_TOP_K=5\n\n# Automatically ingest evidence into RAG\nRAG_AUTO_INGEST=true\n

The RAG configuration:

"},{"location":"configuration/#chromadb-configuration","title":"ChromaDB Configuration","text":"

Configure the vector database for embeddings and RAG:

# ChromaDB storage path\nCHROMA_DB_PATH=./chroma_db\n\n# Whether to persist ChromaDB to disk\nCHROMA_DB_PERSIST=true\n\n# ChromaDB server host (for remote ChromaDB, optional)\nCHROMA_DB_HOST=localhost\n\n# ChromaDB server port (for remote ChromaDB, optional)\nCHROMA_DB_PORT=8000\n

The ChromaDB configuration:

"},{"location":"configuration/#external-services","title":"External Services","text":""},{"location":"configuration/#modal-configuration","title":"Modal Configuration","text":"

Modal is used for secure sandbox execution of statistical analysis:

# Modal Token ID (for Modal sandbox execution)\nMODAL_TOKEN_ID=your_modal_token_id_here\n\n# Modal Token Secret\nMODAL_TOKEN_SECRET=your_modal_token_secret_here\n

The Modal configuration:

"},{"location":"configuration/#logging-configuration","title":"Logging Configuration","text":"

Configure structured logging:

# Log Level: \"DEBUG\", \"INFO\", \"WARNING\", or \"ERROR\"\nLOG_LEVEL=INFO\n

The logging configuration:

Logging is configured via the configure_logging() function:

"},{"location":"configuration/#configuration-properties","title":"Configuration Properties","text":"

The Settings class provides helpful properties for checking configuration state:

"},{"location":"configuration/#api-key-availability","title":"API Key Availability","text":"

Check which API keys are available:

Usage:

from src.utils.config import settings\n\n# Check API key availability\nif settings.has_openai_key:\n    # Use OpenAI\n    pass\n\nif settings.has_anthropic_key:\n    # Use Anthropic\n    pass\n\nif settings.has_huggingface_key:\n    # Use HuggingFace\n    pass\n\nif settings.has_any_llm_key:\n    # At least one LLM is available\n    pass\n
"},{"location":"configuration/#service-availability","title":"Service Availability","text":"

Check if external services are configured:

Usage:

from src.utils.config import settings\n\n# Check service availability\nif settings.modal_available:\n    # Use Modal sandbox\n    pass\n\nif settings.web_search_available:\n    # Web search is configured\n    pass\n
"},{"location":"configuration/#api-key-retrieval","title":"API Key Retrieval","text":"

Get the API key for the configured provider:

For OpenAI-specific operations (e.g., Magentic mode):

"},{"location":"configuration/#configuration-usage-in-codebase","title":"Configuration Usage in Codebase","text":"

The configuration system is used throughout the codebase:

"},{"location":"configuration/#llm-factory","title":"LLM Factory","text":"

The LLM factory uses settings to create appropriate models:

"},{"location":"configuration/#embedding-service","title":"Embedding Service","text":"

The embedding service uses local embedding model configuration:

"},{"location":"configuration/#orchestrator-factory","title":"Orchestrator Factory","text":"

The orchestrator factory uses settings to determine mode:

"},{"location":"configuration/#environment-variables-reference","title":"Environment Variables Reference","text":""},{"location":"configuration/#required-at-least-one-llm","title":"Required (at least one LLM)","text":""},{"location":"configuration/#llm-configuration-variables","title":"LLM Configuration Variables","text":""},{"location":"configuration/#embedding-configuration-variables","title":"Embedding Configuration Variables","text":""},{"location":"configuration/#web-search-configuration-variables","title":"Web Search Configuration Variables","text":""},{"location":"configuration/#pubmed-configuration-variables","title":"PubMed Configuration Variables","text":""},{"location":"configuration/#agent-configuration-variables","title":"Agent Configuration Variables","text":""},{"location":"configuration/#budget-configuration-variables","title":"Budget Configuration Variables","text":""},{"location":"configuration/#rag-configuration-variables","title":"RAG Configuration Variables","text":""},{"location":"configuration/#chromadb-configuration-variables","title":"ChromaDB Configuration Variables","text":""},{"location":"configuration/#external-services-variables","title":"External Services Variables","text":""},{"location":"configuration/#logging-configuration-variables","title":"Logging Configuration Variables","text":""},{"location":"configuration/#validation","title":"Validation","text":"

Settings are validated on load using Pydantic validation:

"},{"location":"configuration/#validation-examples","title":"Validation Examples","text":"

The max_iterations field has range validation:

The llm_provider field has literal validation:

"},{"location":"configuration/#error-handling","title":"Error Handling","text":"

Configuration errors raise ConfigurationError from src/utils/exceptions.py:

```22:25:src/utils/exceptions.py class ConfigurationError(DeepCriticalError): \"\"\"Raised when configuration is invalid.\"\"\"

pass\n

```

"},{"location":"configuration/#error-handling-example","title":"Error Handling Example","text":"

python from src.utils.config import settings from src.utils.exceptions import ConfigurationError try: api_key = settings.get_api_key() except ConfigurationError as e: print(f\"Configuration error: {e}\")

"},{"location":"configuration/#common-configuration-errors","title":"Common Configuration Errors","text":"
  1. Missing API Key: When get_api_key() is called but the required API key is not set
  2. Invalid Provider: When llm_provider is set to an unsupported value
  3. Out of Range: When numeric values exceed their min/max constraints
  4. Invalid Literal: When enum fields receive unsupported values
"},{"location":"configuration/#configuration-best-practices","title":"Configuration Best Practices","text":"
  1. Use .env File: Store sensitive keys in .env file (add to .gitignore)
  2. Check Availability: Use properties like has_openai_key before accessing API keys
  3. Handle Errors: Always catch ConfigurationError when calling get_api_key()
  4. Validate Early: Configuration is validated on import, so errors surface immediately
  5. Use Defaults: Leverage sensible defaults for optional configuration
"},{"location":"configuration/#future-enhancements","title":"Future Enhancements","text":"

The following configurations are planned for future phases:

  1. Additional LLM Providers: DeepSeek, OpenRouter, Gemini, Perplexity, Azure OpenAI, Local models
  2. Model Selection: Reasoning/main/fast model configuration
  3. Service Integration: Additional service integrations and configurations
"},{"location":"contributing/","title":"Contributing to The DETERMINATOR","text":"

Thank you for your interest in contributing to The DETERMINATOR! This guide will help you get started.

Note on Project Names: \"The DETERMINATOR\" is the product name, \"DeepCritical\" is the organization/project name, and \"determinator\" is the Python package name.

"},{"location":"contributing/#git-workflow","title":"Git Workflow","text":""},{"location":"contributing/#repository-information","title":"Repository Information","text":""},{"location":"contributing/#dual-repository-setup","title":"Dual Repository Setup","text":"

This project uses a dual repository setup:

"},{"location":"contributing/#remote-configuration","title":"Remote Configuration","text":"

When cloning, set up remotes as follows:

# Clone from GitHub\ngit clone https://github.com/DeepCritical/GradioDemo.git\ncd GradioDemo\n\n# Add HuggingFace remote (optional, for deployment)\ngit remote add huggingface-upstream https://huggingface.co/spaces/DataQuests/DeepCritical\n

Important: Never push directly to main or dev on HuggingFace. Always work through GitHub PRs. GitHub is the source of truth; HuggingFace is for deployment/demo only.

"},{"location":"contributing/#package-manager","title":"Package Manager","text":"

This project uses uv as the package manager. All commands should be prefixed with uv run to ensure they run in the correct environment.

"},{"location":"contributing/#installation","title":"Installation","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync all dependencies including dev extras\nuv sync --all-extras\n\n# Install pre-commit hooks\nuv run pre-commit install\n
"},{"location":"contributing/#development-commands","title":"Development Commands","text":"
# Installation\nuv sync --all-extras              # Install all dependencies including dev\nuv run pre-commit install          # Install pre-commit hooks\n\n# Code Quality Checks (run all before committing)\nuv run ruff check src tests       # Lint with ruff\nuv run ruff format src tests      # Format with ruff\nuv run mypy src                   # Type checking\nuv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire  # Tests with coverage\n\n# Testing Commands\nuv run pytest tests/unit/ -v -m \"not openai\" -p no:logfire              # Run unit tests (excludes OpenAI tests)\nuv run pytest tests/ -v -m \"huggingface\" -p no:logfire                 # Run HuggingFace tests\nuv run pytest tests/ -v -p no:logfire                                  # Run all tests\nuv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire  # Tests with terminal coverage\nuv run pytest --cov=src --cov-report=html -p no:logfire                # Generate HTML coverage report (opens htmlcov/index.html)\n\n# Documentation Commands\nuv run mkdocs build                # Build documentation\nuv run mkdocs serve                # Serve documentation locally (http://127.0.0.1:8000)\n
"},{"location":"contributing/#test-markers","title":"Test Markers","text":"

The project uses pytest markers to categorize tests. See Testing Guidelines for details:

Note: The -p no:logfire flag disables the logfire plugin to avoid conflicts during testing.

"},{"location":"contributing/#getting-started","title":"Getting Started","text":"

  1. Fork the repository on GitHub: DeepCritical/GradioDemo

  2. Clone your fork:

git clone https://github.com/yourusername/GradioDemo.git\ncd GradioDemo\n
  1. Install dependencies:
uv sync --all-extras\nuv run pre-commit install\n
  1. Create a feature branch:
git checkout -b yourname-feature-name\n

  1. Make your changes following the guidelines below

  2. Run checks:

uv run ruff check src tests\nuv run mypy src\nuv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire\n
  1. Commit and push:
git commit -m \"Description of changes\"\ngit push origin yourname-feature-name\n
  1. Create a pull request on GitHub
"},{"location":"contributing/#development-guidelines","title":"Development Guidelines","text":""},{"location":"contributing/#code-style","title":"Code Style","text":""},{"location":"contributing/#error-handling","title":"Error Handling","text":""},{"location":"contributing/#testing","title":"Testing","text":""},{"location":"contributing/#implementation-patterns","title":"Implementation Patterns","text":""},{"location":"contributing/#prompt-engineering","title":"Prompt Engineering","text":""},{"location":"contributing/#code-quality","title":"Code Quality","text":""},{"location":"contributing/#mcp-integration","title":"MCP Integration","text":""},{"location":"contributing/#mcp-tools","title":"MCP Tools","text":""},{"location":"contributing/#gradio-mcp-server","title":"Gradio MCP Server","text":""},{"location":"contributing/#common-pitfalls","title":"Common Pitfalls","text":"
  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Hallucinated citations: Always validate references
  4. Global mutable state: Use ContextVar or pass via parameters
  5. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
  6. Rate limiting: Always implement for external APIs
  7. Error chaining: Always use from e when raising exceptions
"},{"location":"contributing/#key-principles","title":"Key Principles","text":"
  1. Type Safety First: All code must pass mypy --strict
  2. Async Everything: All I/O must be async
  3. Test-Driven: Write tests before implementation
  4. No Hallucinations: Validate all citations
  5. Graceful Degradation: Support free tier (HF Inference) when no API keys
  6. Lazy Loading: Don't require optional dependencies at import time
  7. Structured Logging: Use structlog, never print()
  8. Error Chaining: Always preserve exception context
"},{"location":"contributing/#pull-request-process","title":"Pull Request Process","text":"
  1. Ensure all checks pass: uv run ruff check src tests && uv run mypy src && uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire
  2. Update documentation if needed
  3. Add tests for new features
  4. Update CHANGELOG if applicable
  5. Request review from maintainers
  6. Address review feedback
  7. Wait for approval before merging
"},{"location":"contributing/#project-structure","title":"Project Structure","text":""},{"location":"contributing/#questions","title":"Questions?","text":"

Thank you for contributing to The DETERMINATOR!

"},{"location":"contributing/code-quality/","title":"Code Quality & Documentation","text":"

This document outlines code quality standards and documentation requirements for The DETERMINATOR.

"},{"location":"contributing/code-quality/#linting","title":"Linting","text":""},{"location":"contributing/code-quality/#type-checking","title":"Type Checking","text":""},{"location":"contributing/code-quality/#pre-commit","title":"Pre-commit","text":"

Pre-commit hooks run automatically on commit to ensure code quality. Configuration is in .pre-commit-config.yaml.

"},{"location":"contributing/code-quality/#installation","title":"Installation","text":"
# Install dependencies (includes pre-commit package)\nuv sync --all-extras\n\n# Set up git hooks (must be run separately)\nuv run pre-commit install\n

Note: uv sync --all-extras installs the pre-commit package, but you must run uv run pre-commit install separately to set up the git hooks.

"},{"location":"contributing/code-quality/#pre-commit-hooks","title":"Pre-commit Hooks","text":"

The following hooks run automatically on commit:

  1. ruff: Lints code and fixes issues automatically
  2. Runs on: src/ (excludes tests/, reference_repos/)

  3. Auto-fixes: Yes

  4. ruff-format: Formats code with ruff

  5. Runs on: src/ (excludes tests/, reference_repos/)

  6. Auto-fixes: Yes

  7. mypy: Type checking

  8. Runs on: src/ (excludes folder/)

  9. Additional dependencies: pydantic, pydantic-settings, tenacity, pydantic-ai

  10. pytest-unit: Runs unit tests (excludes OpenAI and embedding_provider tests)

  11. Runs: tests/unit/ with -m \"not openai and not embedding_provider\"

  12. Always runs: Yes (not just on changed files)

  13. pytest-local-embeddings: Runs local embedding tests

  14. Runs: tests/ with -m \"local_embeddings\"
  15. Always runs: Yes
"},{"location":"contributing/code-quality/#manual-pre-commit-run","title":"Manual Pre-commit Run","text":"

To run pre-commit hooks manually (without committing):

uv run pre-commit run --all-files\n
"},{"location":"contributing/code-quality/#troubleshooting","title":"Troubleshooting","text":""},{"location":"contributing/code-quality/#documentation","title":"Documentation","text":""},{"location":"contributing/code-quality/#building-documentation","title":"Building Documentation","text":"

Documentation is built using MkDocs. Source files are in docs/, and the configuration is in mkdocs.yml.

# Build documentation\nuv run mkdocs build\n\n# Serve documentation locally (http://127.0.0.1:8000)\nuv run mkdocs serve\n

The documentation site is published at: https://deepcritical.github.io/GradioDemo/

"},{"location":"contributing/code-quality/#docstrings","title":"Docstrings","text":"

Example:

"},{"location":"contributing/code-quality/#code-comments","title":"Code Comments","text":""},{"location":"contributing/code-quality/#see-also","title":"See Also","text":""},{"location":"contributing/code-style/","title":"Code Style & Conventions","text":"

This document outlines the code style and conventions for The DETERMINATOR.

"},{"location":"contributing/code-style/#package-manager","title":"Package Manager","text":"

This project uses uv as the package manager. All commands should be prefixed with uv run to ensure they run in the correct environment.

"},{"location":"contributing/code-style/#installation","title":"Installation","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync all dependencies including dev extras\nuv sync --all-extras\n
"},{"location":"contributing/code-style/#running-commands","title":"Running Commands","text":"

All development commands should use uv run prefix:

# Instead of: pytest tests/\nuv run pytest tests/\n\n# Instead of: ruff check src\nuv run ruff check src\n\n# Instead of: mypy src\nuv run mypy src\n

This ensures commands run in the correct virtual environment managed by uv.

"},{"location":"contributing/code-style/#type-safety","title":"Type Safety","text":""},{"location":"contributing/code-style/#pydantic-models","title":"Pydantic Models","text":""},{"location":"contributing/code-style/#async-patterns","title":"Async Patterns","text":" 
loop = asyncio.get_running_loop()\nresult = await loop.run_in_executor(None, cpu_bound_function, args)\n
"},{"location":"contributing/code-style/#common-pitfalls","title":"Common Pitfalls","text":"
  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Global mutable state: Use ContextVar or pass via parameters
  4. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
"},{"location":"contributing/code-style/#see-also","title":"See Also","text":""},{"location":"contributing/error-handling/","title":"Error Handling & Logging","text":"

This document outlines error handling and logging conventions for The DETERMINATOR.

"},{"location":"contributing/error-handling/#exception-hierarchy","title":"Exception Hierarchy","text":"

Use custom exception hierarchy (src/utils/exceptions.py):

"},{"location":"contributing/error-handling/#error-handling-rules","title":"Error Handling Rules","text":" 
logger.error(\"Operation failed\", error=str(e), context=value)\n
"},{"location":"contributing/error-handling/#logging","title":"Logging","text":""},{"location":"contributing/error-handling/#logging-examples","title":"Logging Examples","text":"
logger.info(\"Starting search\", query=query, tools=[t.name for t in tools])\nlogger.warning(\"Search tool failed\", tool=tool.name, error=str(result))\nlogger.error(\"Assessment failed\", error=str(e))\n
"},{"location":"contributing/error-handling/#error-chaining","title":"Error Chaining","text":"

Always preserve exception context:

try:\n    result = await api_call()\nexcept httpx.HTTPError as e:\n    raise SearchError(f\"API call failed: {e}\") from e\n
"},{"location":"contributing/error-handling/#see-also","title":"See Also","text":""},{"location":"contributing/implementation-patterns/","title":"Implementation Patterns","text":"

This document outlines common implementation patterns used in The DETERMINATOR.

"},{"location":"contributing/implementation-patterns/#search-tools","title":"Search Tools","text":"

All tools implement SearchTool protocol (src/tools/base.py):

Example pattern:

class MySearchTool:\n    @property\n    def name(self) -> str:\n        return \"mytool\"\n    \n    @retry(stop=stop_after_attempt(3), wait=wait_exponential(...))\n    async def search(self, query: str, max_results: int = 10) -> list[Evidence]:\n        # Implementation\n        return evidence_list\n
"},{"location":"contributing/implementation-patterns/#judge-handlers","title":"Judge Handlers","text":""},{"location":"contributing/implementation-patterns/#agent-factory-pattern","title":"Agent Factory Pattern","text":""},{"location":"contributing/implementation-patterns/#state-management","title":"State Management","text":""},{"location":"contributing/implementation-patterns/#singleton-pattern","title":"Singleton Pattern","text":"

Use @lru_cache(maxsize=1) for singletons:

"},{"location":"contributing/implementation-patterns/#see-also","title":"See Also","text":""},{"location":"contributing/prompt-engineering/","title":"Prompt Engineering & Citation Validation","text":"

This document outlines prompt engineering guidelines and citation validation rules.

"},{"location":"contributing/prompt-engineering/#judge-prompts","title":"Judge Prompts","text":""},{"location":"contributing/prompt-engineering/#hypothesis-prompts","title":"Hypothesis Prompts","text":""},{"location":"contributing/prompt-engineering/#report-prompts","title":"Report Prompts","text":""},{"location":"contributing/prompt-engineering/#citation-validation","title":"Citation Validation","text":""},{"location":"contributing/prompt-engineering/#citation-validation-rules","title":"Citation Validation Rules","text":"
  1. Every reference URL must EXACTLY match a provided evidence URL
  2. Do NOT invent, fabricate, or hallucinate any references
  3. Do NOT modify paper titles, authors, dates, or URLs
  4. If unsure about a citation, OMIT it rather than guess
  5. Copy URLs exactly as provided - do not create similar-looking URLs
"},{"location":"contributing/prompt-engineering/#evidence-selection","title":"Evidence Selection","text":""},{"location":"contributing/prompt-engineering/#see-also","title":"See Also","text":""},{"location":"contributing/testing/","title":"Testing Requirements","text":"

This document outlines testing requirements and guidelines for The DETERMINATOR.

"},{"location":"contributing/testing/#test-structure","title":"Test Structure","text":""},{"location":"contributing/testing/#test-markers","title":"Test Markers","text":"

The project uses pytest markers to categorize tests. These markers are defined in pyproject.toml:

"},{"location":"contributing/testing/#running-tests-by-marker","title":"Running Tests by Marker","text":"
# Run only unit tests (excludes OpenAI tests by default)\nuv run pytest tests/unit/ -v -m \"not openai\" -p no:logfire\n\n# Run HuggingFace tests\nuv run pytest tests/ -v -m \"huggingface\" -p no:logfire\n\n# Run all tests\nuv run pytest tests/ -v -p no:logfire\n\n# Run only local embedding tests\nuv run pytest tests/ -v -m \"local_embeddings\" -p no:logfire\n\n# Exclude slow tests\nuv run pytest tests/ -v -m \"not slow\" -p no:logfire\n

Note: The -p no:logfire flag disables the logfire plugin to avoid conflicts during testing.

"},{"location":"contributing/testing/#mocking","title":"Mocking","text":""},{"location":"contributing/testing/#tdd-workflow","title":"TDD Workflow","text":"
  1. Write failing test in tests/unit/
  2. Implement in src/
  3. Ensure test passes
  4. Run checks: uv run ruff check src tests && uv run mypy src && uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire
"},{"location":"contributing/testing/#test-command-examples","title":"Test Command Examples","text":"
# Run unit tests (default, excludes OpenAI tests)\nuv run pytest tests/unit/ -v -m \"not openai\" -p no:logfire\n\n# Run HuggingFace tests\nuv run pytest tests/ -v -m \"huggingface\" -p no:logfire\n\n# Run all tests\nuv run pytest tests/ -v -p no:logfire\n
"},{"location":"contributing/testing/#test-examples","title":"Test Examples","text":"
@pytest.mark.unit\nasync def test_pubmed_search(mock_httpx_client):\n    tool = PubMedTool()\n    results = await tool.search(\"metformin\", max_results=5)\n    assert len(results) > 0\n    assert all(isinstance(r, Evidence) for r in results)\n\n@pytest.mark.integration\nasync def test_real_pubmed_search():\n    tool = PubMedTool()\n    results = await tool.search(\"metformin\", max_results=3)\n    assert len(results) <= 3\n
"},{"location":"contributing/testing/#test-coverage","title":"Test Coverage","text":""},{"location":"contributing/testing/#terminal-coverage-report","title":"Terminal Coverage Report","text":"
uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire\n

This shows coverage with missing lines highlighted in the terminal output.

"},{"location":"contributing/testing/#html-coverage-report","title":"HTML Coverage Report","text":"
uv run pytest --cov=src --cov-report=html -p no:logfire\n

This generates an HTML coverage report in htmlcov/index.html. Open this file in your browser to see detailed coverage information.

"},{"location":"contributing/testing/#coverage-goals","title":"Coverage Goals","text":""},{"location":"contributing/testing/#see-also","title":"See Also","text":""},{"location":"getting-started/examples/","title":"Examples","text":"

This page provides examples of using The DETERMINATOR for various research tasks.

"},{"location":"getting-started/examples/#basic-research-query","title":"Basic Research Query","text":""},{"location":"getting-started/examples/#example-1-drug-information","title":"Example 1: Drug Information","text":"

Query: 

What are the latest treatments for Alzheimer's disease?\n

What The DETERMINATOR Does: 1. Searches PubMed for recent papers 2. Searches ClinicalTrials.gov for active trials 3. Evaluates evidence quality 4. Synthesizes findings into a comprehensive report

"},{"location":"getting-started/examples/#example-2-clinical-trial-search","title":"Example 2: Clinical Trial Search","text":"

Query: 

What clinical trials are investigating metformin for cancer prevention?\n

What The DETERMINATOR Does:

  1. Searches ClinicalTrials.gov for relevant trials
  2. Searches PubMed for supporting literature
  3. Provides trial details and status
  4. Summarizes findings
"},{"location":"getting-started/examples/#advanced-research-queries","title":"Advanced Research Queries","text":""},{"location":"getting-started/examples/#example-3-comprehensive-review","title":"Example 3: Comprehensive Review","text":"

Query:

Review the evidence for using metformin as an anti-aging intervention, \nincluding clinical trials, mechanisms of action, and safety profile.\n

What The DETERMINATOR Does: 1. Uses deep research mode (multi-section) 2. Searches multiple sources in parallel 3. Generates sections on: - Clinical trials - Mechanisms of action - Safety profile 4. Synthesizes comprehensive report

"},{"location":"getting-started/examples/#example-4-hypothesis-testing","title":"Example 4: Hypothesis Testing","text":"

Query: 

Test the hypothesis that regular exercise reduces Alzheimer's disease risk.\n

What The DETERMINATOR Does: 1. Generates testable hypotheses 2. Searches for supporting/contradicting evidence 3. Performs statistical analysis (if Modal configured) 4. Provides verdict: SUPPORTED, REFUTED, or INCONCLUSIVE

"},{"location":"getting-started/examples/#mcp-tool-examples","title":"MCP Tool Examples","text":""},{"location":"getting-started/examples/#using-search_pubmed","title":"Using search_pubmed","text":"
Search PubMed for \"CRISPR gene editing cancer therapy\"\n
"},{"location":"getting-started/examples/#using-search_clinical_trials","title":"Using search_clinical_trials","text":"
Find active clinical trials for \"diabetes type 2 treatment\"\n
"},{"location":"getting-started/examples/#using-search_all","title":"Using search_all","text":"
Search all sources for \"COVID-19 vaccine side effects\"\n
"},{"location":"getting-started/examples/#using-analyze_hypothesis","title":"Using analyze_hypothesis","text":"
Analyze whether vitamin D supplementation reduces COVID-19 severity\n
"},{"location":"getting-started/examples/#code-examples","title":"Code Examples","text":""},{"location":"getting-started/examples/#python-api-usage","title":"Python API Usage","text":"
from src.orchestrator_factory import create_orchestrator\nfrom src.tools.search_handler import SearchHandler\nfrom src.agent_factory.judges import create_judge_handler\n\n# Create orchestrator\nsearch_handler = SearchHandler()\njudge_handler = create_judge_handler()\n
# Run research query\nquery = \"What are the latest treatments for Alzheimer's disease?\"\nasync for event in orchestrator.run(query):\n    print(f\"Event: {event.type} - {event.data}\")\n
"},{"location":"getting-started/examples/#gradio-ui-integration","title":"Gradio UI Integration","text":"
import gradio as gr\nfrom src.app import create_research_interface\n\n# Create interface\ninterface = create_research_interface()\n\n# Launch\ninterface.launch(server_name=\"0.0.0.0\", server_port=7860)\n
"},{"location":"getting-started/examples/#research-patterns","title":"Research Patterns","text":""},{"location":"getting-started/examples/#iterative-research","title":"Iterative Research","text":"

Single-loop research with search-judge-synthesize cycles:

from src.orchestrator.research_flow import IterativeResearchFlow\n
async for event in flow.run(query):\n    # Handle events\n    pass\n
"},{"location":"getting-started/examples/#deep-research","title":"Deep Research","text":"

Multi-section parallel research:

from src.orchestrator.research_flow import DeepResearchFlow\n
async for event in flow.run(query):\n    # Handle events\n    pass\n
"},{"location":"getting-started/examples/#configuration-examples","title":"Configuration Examples","text":""},{"location":"getting-started/examples/#basic-configuration","title":"Basic Configuration","text":"
# .env file\nLLM_PROVIDER=openai\nOPENAI_API_KEY=your_key_here\nMAX_ITERATIONS=10\n
"},{"location":"getting-started/examples/#advanced-configuration","title":"Advanced Configuration","text":"
# .env file\nLLM_PROVIDER=anthropic\nANTHROPIC_API_KEY=your_key_here\nEMBEDDING_PROVIDER=local\nWEB_SEARCH_PROVIDER=duckduckgo\nMAX_ITERATIONS=20\nDEFAULT_TOKEN_LIMIT=200000\nUSE_GRAPH_EXECUTION=true\n
"},{"location":"getting-started/examples/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/installation/","title":"Installation","text":"

This guide will help you install and set up DeepCritical on your system.

"},{"location":"getting-started/installation/#prerequisites","title":"Prerequisites","text":""},{"location":"getting-started/installation/#installation-steps","title":"Installation Steps","text":""},{"location":"getting-started/installation/#1-install-uv-recommended","title":"1. Install uv (Recommended)","text":"

uv is a fast Python package installer and resolver. Install it using the standalone installer (recommended):

Unix/macOS/Linux: 

curl -LsSf https://astral.sh/uv/install.sh | sh\n

Windows (PowerShell): 

powershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n

Alternative methods: 

# Using pipx (recommended if you have pipx installed)\npipx install uv\n\n# Or using pip\npip install uv\n

After installation, restart your terminal or add ~/.cargo/bin to your PATH.

"},{"location":"getting-started/installation/#2-clone-the-repository","title":"2. Clone the Repository","text":"
git clone https://github.com/DeepCritical/GradioDemo.git\ncd GradioDemo\n
"},{"location":"getting-started/installation/#3-install-dependencies","title":"3. Install Dependencies","text":"

Using uv (recommended):

uv sync\n

Using pip:

pip install -e .\n
"},{"location":"getting-started/installation/#4-install-optional-dependencies","title":"4. Install Optional Dependencies","text":"

For embeddings support (local sentence-transformers):

uv sync --extra embeddings\n

For Modal sandbox execution:

uv sync --extra modal\n

For Magentic orchestration:

uv sync --extra magentic\n

Install all extras:

uv sync --all-extras\n
"},{"location":"getting-started/installation/#5-configure-environment-variables","title":"5. Configure Environment Variables","text":"

Create a .env file in the project root:

# Required: At least one LLM provider\nLLM_PROVIDER=openai  # or \"anthropic\" or \"huggingface\"\nOPENAI_API_KEY=your_openai_api_key_here\n\n# Optional: Other services\nNCBI_API_KEY=your_ncbi_api_key_here  # For higher PubMed rate limits\nMODAL_TOKEN_ID=your_modal_token_id\nMODAL_TOKEN_SECRET=your_modal_token_secret\n

See the Configuration Guide for all available options.

"},{"location":"getting-started/installation/#6-verify-installation","title":"6. Verify Installation","text":"

Run the application:

uv run gradio run src/app.py\n

Open your browser to http://localhost:7860 to verify the installation.

"},{"location":"getting-started/installation/#development-setup","title":"Development Setup","text":"

For development, install dev dependencies:

uv sync --all-extras --dev\n

Install pre-commit hooks:

uv run pre-commit install\n
"},{"location":"getting-started/installation/#troubleshooting","title":"Troubleshooting","text":""},{"location":"getting-started/installation/#common-issues","title":"Common Issues","text":"

Import Errors: - Ensure you've installed all required dependencies - Check that Python 3.11+ is being used

API Key Errors: - Verify your .env file is in the project root - Check that API keys are correctly formatted - Ensure at least one LLM provider is configured

Module Not Found: - Run uv sync or pip install -e . again - Check that you're in the correct virtual environment

Port Already in Use: - Change the port in src/app.py or use environment variable - Kill the process using port 7860

"},{"location":"getting-started/installation/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/mcp-integration/","title":"MCP Integration","text":"

The DETERMINATOR exposes a Model Context Protocol (MCP) server, allowing you to use its search tools directly from Claude Desktop or other MCP clients.

"},{"location":"getting-started/mcp-integration/#what-is-mcp","title":"What is MCP?","text":"

The Model Context Protocol (MCP) is a standard for connecting AI assistants to external tools and data sources. The DETERMINATOR implements an MCP server that exposes its search capabilities as MCP tools.

"},{"location":"getting-started/mcp-integration/#mcp-server-url","title":"MCP Server URL","text":"

When running locally:

http://localhost:7860/gradio_api/mcp/\n
"},{"location":"getting-started/mcp-integration/#claude-desktop-configuration","title":"Claude Desktop Configuration","text":""},{"location":"getting-started/mcp-integration/#1-locate-configuration-file","title":"1. Locate Configuration File","text":"

macOS: 

~/Library/Application Support/Claude/claude_desktop_config.json\n

Windows: 

%APPDATA%\\Claude\\claude_desktop_config.json\n

Linux: 

~/.config/Claude/claude_desktop_config.json\n

"},{"location":"getting-started/mcp-integration/#2-add-the-determinator-server","title":"2. Add The DETERMINATOR Server","text":"

Edit claude_desktop_config.json and add:

{\n  \"mcpServers\": {\n    \"determinator\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#3-restart-claude-desktop","title":"3. Restart Claude Desktop","text":"

Close and restart Claude Desktop for changes to take effect.

"},{"location":"getting-started/mcp-integration/#4-verify-connection","title":"4. Verify Connection","text":"

In Claude Desktop, you should see The DETERMINATOR tools available: - search_pubmed - search_clinical_trials - search_biorxiv - search_all - analyze_hypothesis

"},{"location":"getting-started/mcp-integration/#available-tools","title":"Available Tools","text":""},{"location":"getting-started/mcp-integration/#search_pubmed","title":"search_pubmed","text":"

Search peer-reviewed biomedical literature from PubMed.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search PubMed for \"metformin diabetes\"\n

"},{"location":"getting-started/mcp-integration/#search_clinical_trials","title":"search_clinical_trials","text":"

Search ClinicalTrials.gov for interventional studies.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search clinical trials for \"Alzheimer's disease treatment\"\n

"},{"location":"getting-started/mcp-integration/#search_biorxiv","title":"search_biorxiv","text":"

Search bioRxiv/medRxiv preprints via Europe PMC.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search bioRxiv for \"CRISPR gene editing\"\n

"},{"location":"getting-started/mcp-integration/#search_all","title":"search_all","text":"

Search all sources simultaneously (PubMed, ClinicalTrials.gov, Europe PMC).

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results per source (default: 10)

Example: 

Search all sources for \"COVID-19 vaccine efficacy\"\n

"},{"location":"getting-started/mcp-integration/#analyze_hypothesis","title":"analyze_hypothesis","text":"

Perform secure statistical analysis using Modal sandboxes.

Parameters: - hypothesis (string): Hypothesis to analyze - data (string, optional): Data description or code

Example: 

Analyze the hypothesis that metformin reduces cancer risk\n

"},{"location":"getting-started/mcp-integration/#using-tools-in-claude-desktop","title":"Using Tools in Claude Desktop","text":"

Once configured, you can ask Claude to use DeepCritical tools:

Use DeepCritical to search PubMed for recent papers on Alzheimer's disease treatments.\n

Claude will automatically: 1. Call the appropriate DeepCritical tool 2. Retrieve results 3. Use the results in its response

"},{"location":"getting-started/mcp-integration/#troubleshooting","title":"Troubleshooting","text":""},{"location":"getting-started/mcp-integration/#connection-issues","title":"Connection Issues","text":"

Server Not Found: - Ensure DeepCritical is running (uv run gradio run src/app.py) - Verify the URL in claude_desktop_config.json is correct - Check that port 7860 is not blocked by firewall

Tools Not Appearing: - Restart Claude Desktop after configuration changes - Check Claude Desktop logs for errors - Verify MCP server is accessible at the configured URL

"},{"location":"getting-started/mcp-integration/#authentication","title":"Authentication","text":"

If DeepCritical requires authentication: - Configure API keys in DeepCritical settings - Use HuggingFace OAuth login - Ensure API keys are valid

"},{"location":"getting-started/mcp-integration/#advanced-configuration","title":"Advanced Configuration","text":""},{"location":"getting-started/mcp-integration/#custom-port","title":"Custom Port","text":"

If running on a different port, update the URL:

{\n  \"mcpServers\": {\n    \"deepcritical\": {\n      \"url\": \"http://localhost:8080/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#multiple-instances","title":"Multiple Instances","text":"

You can configure multiple DeepCritical instances:

{\n  \"mcpServers\": {\n    \"deepcritical-local\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    },\n    \"deepcritical-remote\": {\n      \"url\": \"https://your-server.com/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/quick-start/","title":"Single Command Deploy","text":"

Deploy with docker instandly with a single command :

docker run -it -p 7860:7860 --platform=linux/amd64 \\\n    -e DB_KEY=\"YOUR_VALUE_HERE\" \\\n    -e SERP_API=\"YOUR_VALUE_HERE\" \\\n    -e INFERENCE_API=\"YOUR_VALUE_HERE\" \\\n    -e MODAL_TOKEN_ID=\"YOUR_VALUE_HERE\" \\\n    -e MODAL_TOKEN_SECRET=\"YOUR_VALUE_HERE\" \\\n    -e NCBI_API_KEY=\"YOUR_VALUE_HERE\" \\\n    -e SERPER_API_KEY=\"YOUR_VALUE_HERE\" \\\n    -e CHROMA_DB_PATH=\"./chroma_db\" \\\n    -e CHROMA_DB_HOST=\"localhost\" \\\n    -e CHROMA_DB_PORT=\"8000\" \\\n    -e RAG_COLLECTION_NAME=\"deepcritical_evidence\" \\\n    -e RAG_SIMILARITY_TOP_K=\"5\" \\\n    -e RAG_AUTO_INGEST=\"true\" \\\n    -e USE_GRAPH_EXECUTION=\"false\" \\\n    -e DEFAULT_TOKEN_LIMIT=\"100000\" \\\n    -e DEFAULT_TIME_LIMIT_MINUTES=\"10\" \\\n    -e DEFAULT_ITERATIONS_LIMIT=\"10\" \\\n    -e WEB_SEARCH_PROVIDER=\"duckduckgo\" \\\n    -e MAX_ITERATIONS=\"10\" \\\n    -e SEARCH_TIMEOUT=\"30\" \\\n    -e LOG_LEVEL=\"DEBUG\" \\\n    -e EMBEDDING_PROVIDER=\"local\" \\\n    -e OPENAI_EMBEDDING_MODEL=\"text-embedding-3-small\" \\\n    -e LOCAL_EMBEDDING_MODEL=\"BAAI/bge-small-en-v1.5\" \\\n    -e HUGGINGFACE_EMBEDDING_MODEL=\"sentence-transformers/all-MiniLM-L6-v2\" \\\n    -e HF_FALLBACK_MODELS=\"Qwen/Qwen3-Next-80B-A3B-Thinking,Qwen/Qwen3-Next-80B-A3B-Instruct,meta-llama/Llama-3.3-70B-Instruct,meta-llama/Llama-3.1-8B-Instruct,HuggingFaceH4/zephyr-7b-beta,Qwen/Qwen2-7B-Instruct\" \\\n    -e HUGGINGFACE_MODEL=\"Qwen/Qwen3-Next-80B-A3B-Thinking\" \\\n    registry.hf.space/dataquests-deepcritical:latest python src/app.py\n   ```\n\n## Quick start guide\n\nGet up and running with The DETERMINATOR in minutes.\n\n## Start the Application\n\n```bash\ngradio src/app.py\n

Open your browser to http://localhost:7860.

"},{"location":"getting-started/quick-start/#first-research-query","title":"First Research Query","text":"
  1. Enter a Research Question

Type your research question in the chat interface, for example: - \"What are the latest treatments for Alzheimer's disease?\" - \"Review the evidence for metformin in cancer prevention\" - \"What clinical trials are investigating COVID-19 vaccines?\"

  1. Submit the Query

Click \"Submit\" or press Enter. The system will: - Generate observations about your query - Identify knowledge gaps - Search multiple sources (PubMed, ClinicalTrials.gov, Europe PMC) - Evaluate evidence quality - Synthesize findings into a report

  1. Review Results

Watch the real-time progress in the chat interface: - Search operations and results - Evidence evaluation - Report generation - Final research report with citations

"},{"location":"getting-started/quick-start/#authentication","title":"Authentication","text":""},{"location":"getting-started/quick-start/#huggingface-oauth-recommended","title":"HuggingFace OAuth (Recommended)","text":"
  1. Click \"Sign in with HuggingFace\" at the top of the app
  2. Authorize the application
  3. Your HuggingFace API token will be automatically used
  4. No need to manually enter API keys
"},{"location":"getting-started/quick-start/#manual-api-key","title":"Manual API Key","text":"
  1. Open the Settings accordion
  2. Enter your API key:
  3. OpenAI API key
  4. Anthropic API key
  5. HuggingFace API key
  6. Click \"Save Settings\"
  7. Manual keys take priority over OAuth tokens
"},{"location":"getting-started/quick-start/#understanding-the-interface","title":"Understanding the Interface","text":""},{"location":"getting-started/quick-start/#chat-interface","title":"Chat Interface","text":""},{"location":"getting-started/quick-start/#status-indicators","title":"Status Indicators","text":""},{"location":"getting-started/quick-start/#settings","title":"Settings","text":""},{"location":"getting-started/quick-start/#example-queries","title":"Example Queries","text":""},{"location":"getting-started/quick-start/#simple-query","title":"Simple Query","text":"
What are the side effects of metformin?\n
"},{"location":"getting-started/quick-start/#complex-query","title":"Complex Query","text":"
Review the evidence for using metformin as an anti-aging intervention, \nincluding clinical trials, mechanisms of action, and safety profile.\n
"},{"location":"getting-started/quick-start/#clinical-trial-query","title":"Clinical Trial Query","text":"
What are the active clinical trials investigating Alzheimer's disease treatments?\n
"},{"location":"getting-started/quick-start/#next-steps","title":"Next Steps","text":""},{"location":"overview/architecture/","title":"Architecture Overview","text":"

The DETERMINATOR is a powerful generalist deep research agent system that uses iterative search-and-judge loops to comprehensively investigate any research question. It stops at nothing until finding precise answers, only stopping at configured limits (budget, time, iterations). The system automatically determines if medical knowledge sources are needed and adapts its search strategy accordingly. It supports multiple orchestration patterns, graph-based execution, parallel research workflows, and long-running task management with real-time streaming.

"},{"location":"overview/architecture/#core-architecture","title":"Core Architecture","text":""},{"location":"overview/architecture/#orchestration-patterns","title":"Orchestration Patterns","text":"
  1. Graph Orchestrator (src/orchestrator/graph_orchestrator.py):
  2. Graph-based execution using Pydantic AI agents as nodes
  3. Supports both iterative and deep research patterns
  4. Node types: Agent, State, Decision, Parallel
  5. Edge types: Sequential, Conditional, Parallel
  6. Conditional routing based on knowledge gaps, budget, and iterations
  7. Parallel execution for concurrent research loops
  8. Event streaming via AsyncGenerator[AgentEvent] for real-time UI updates

  9. Fallback to agent chains when graph execution is disabled

  10. Deep Research Flow (src/orchestrator/research_flow.py):

  11. Pattern: Planner \u2192 Parallel Iterative Loops (one per section) \u2192 Synthesis
  12. Uses PlannerAgent to break query into report sections
  13. Runs IterativeResearchFlow instances in parallel per section via WorkflowManager
  14. Synthesizes results using LongWriterAgent or ProofreaderAgent
  15. Supports both graph execution (use_graph=True) and agent chains (use_graph=False)
  16. Budget tracking per section and globally

  17. State synchronization across parallel loops

  18. Iterative Research Flow (src/orchestrator/research_flow.py):

  19. Pattern: Generate observations \u2192 Evaluate gaps \u2192 Select tools \u2192 Execute \u2192 Judge \u2192 Continue/Complete
  20. Uses KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent, WriterAgent
  21. JudgeHandler assesses evidence sufficiency
  22. Iterates until research complete or constraints met (iterations, time, tokens)

  23. Supports graph execution and agent chains

  24. Magentic Orchestrator (src/orchestrator_magentic.py):

  25. Multi-agent coordination using agent-framework-core
  26. ChatAgent pattern with internal LLMs per agent
  27. Uses MagenticBuilder with participants: searcher, hypothesizer, judge, reporter
  28. Manager orchestrates agents via OpenAIChatClient
  29. Requires OpenAI API key (function calling support)
  30. Event-driven: converts Magentic events to AgentEvent for UI streaming

  31. Supports long-running workflows with max rounds and stall/reset handling

  32. Hierarchical Orchestrator (src/orchestrator_hierarchical.py):

  33. Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge
  34. Adapts Magentic ChatAgent to SubIterationTeam protocol
  35. Event-driven via asyncio.Queue for coordination

  36. Supports sub-iteration patterns for complex research tasks

  37. Legacy Simple Mode (src/legacy_orchestrator.py):

  38. Linear search-judge-synthesize loop
  39. Uses SearchHandlerProtocol and JudgeHandlerProtocol
  40. Generator-based design yielding AgentEvent objects
  41. Backward compatibility for simple use cases
"},{"location":"overview/architecture/#long-running-task-support","title":"Long-Running Task Support","text":"

The system is designed for long-running research tasks with comprehensive state management and streaming:

  1. Event Streaming:
  2. All orchestrators yield AgentEvent objects via AsyncGenerator
  3. Real-time UI updates through Gradio chat interface
  4. Event types: started, searching, search_complete, judging, judge_complete, looping, synthesizing, hypothesizing, complete, error

  5. Metadata includes iteration numbers, tool names, result counts, durations

  6. Budget Tracking (src/middleware/budget_tracker.py):

  7. Per-loop and global budget management
  8. Tracks: tokens, time (seconds), iterations
  9. Budget enforcement at decision nodes
  10. Token estimation (~4 chars per token)
  11. Early termination when budgets exceeded

  12. Budget summaries for monitoring

  13. Workflow Manager (src/middleware/workflow_manager.py):

  14. Coordinates parallel research loops
  15. Tracks loop status: pending, running, completed, failed, cancelled
  16. Synchronizes evidence between loops and global state
  17. Handles errors per loop (doesn't fail all if one fails)
  18. Supports loop cancellation and timeout handling

  19. Evidence deduplication across parallel loops

  20. State Management (src/middleware/state_machine.py):

  21. Thread-safe isolation using ContextVar for concurrent requests
  22. WorkflowState tracks: evidence, conversation history, embedding service
  23. Evidence deduplication by URL
  24. Semantic search via embedding service
  25. State persistence across long-running workflows

  26. Supports both iterative and deep research patterns

  27. Gradio UI (src/app.py):

  28. Real-time streaming of research progress
  29. Accordion-based UI for pending/done operations
  30. OAuth integration (HuggingFace)
  31. Multiple backend support (API keys, free tier)
  32. Handles long-running tasks with progress indicators
  33. Event accumulation for pending operations
"},{"location":"overview/architecture/#graph-architecture","title":"Graph Architecture","text":"

The graph orchestrator (src/orchestrator/graph_orchestrator.py) implements a flexible graph-based execution model:

Node Types:

Edge Types:

Graph Patterns:

Execution Flow:

  1. Graph construction from nodes and edges
  2. Graph validation (no cycles, all nodes reachable)
  3. Graph execution from entry node
  4. Node execution based on type
  5. Edge evaluation for next node(s)
  6. Parallel execution via asyncio.gather()
  7. State updates at state nodes
  8. Event streaming for UI
"},{"location":"overview/architecture/#key-components","title":"Key Components","text":""},{"location":"overview/architecture/#research-team--parallel-execution","title":"Research Team & Parallel Execution","text":"

The system supports complex research workflows through:

  1. WorkflowManager: Coordinates multiple parallel research loops
  2. Creates and tracks ResearchLoop instances
  3. Runs loops in parallel via asyncio.gather()
  4. Synchronizes evidence to global state

  5. Handles loop failures gracefully

  6. Deep Research Pattern: Breaks complex queries into sections

  7. Planner creates report outline with sections
  8. Each section runs as independent iterative research loop
  9. Loops execute in parallel
  10. Evidence shared across loops via global state

  11. Final synthesis combines all section results

  12. State Synchronization: Thread-safe evidence sharing

  13. Evidence deduplication by URL
  14. Global state accessible to all loops
  15. Semantic search across all collected evidence
  16. Conversation history tracking per iteration
"},{"location":"overview/architecture/#configuration--modes","title":"Configuration & Modes","text":"

Note: The UI provides separate controls for orchestrator mode and graph research mode. When using graph-based orchestrators (iterative/deep/auto), the graph research mode determines the specific pattern used within the graph execution.

"},{"location":"overview/features/","title":"Features","text":"

The DETERMINATOR provides a comprehensive set of features for AI-assisted research:

"},{"location":"overview/features/#core-features","title":"Core Features","text":""},{"location":"overview/features/#multi-source-search","title":"Multi-Source Search","text":""},{"location":"overview/features/#mcp-integration","title":"MCP Integration","text":""},{"location":"overview/features/#authentication","title":"Authentication","text":""},{"location":"overview/features/#secure-code-execution","title":"Secure Code Execution","text":""},{"location":"overview/features/#semantic-search--rag","title":"Semantic Search & RAG","text":""},{"location":"overview/features/#orchestration-patterns","title":"Orchestration Patterns","text":"

Orchestrator Modes: - simple: Legacy linear search-judge loop - advanced (or magentic): Multi-agent coordination (requires OpenAI API key) - iterative: Knowledge-gap-driven research with single loop - deep: Parallel section-based research with planning - auto: Intelligent mode detection based on query complexity

Graph Research Modes (used within graph orchestrator): - iterative: Single research loop pattern - deep: Multi-section parallel research pattern - auto: Auto-detect pattern based on query complexity

Execution Modes: - use_graph=True: Graph-based execution with parallel and conditional routing - use_graph=False: Agent chains with sequential execution (backward compatible)

"},{"location":"overview/features/#real-time-streaming","title":"Real-Time Streaming","text":""},{"location":"overview/features/#budget-management","title":"Budget Management","text":""},{"location":"overview/features/#state-management","title":"State Management","text":""},{"location":"overview/features/#multimodal-input--output","title":"Multimodal Input & Output","text":""},{"location":"overview/features/#advanced-features","title":"Advanced Features","text":""},{"location":"overview/features/#agent-system","title":"Agent System","text":""},{"location":"overview/features/#search-tools","title":"Search Tools","text":""},{"location":"overview/features/#error-handling","title":"Error Handling","text":""},{"location":"overview/features/#configuration","title":"Configuration","text":""},{"location":"overview/features/#testing","title":"Testing","text":""},{"location":"overview/features/#ui-features","title":"UI Features","text":""},{"location":"overview/features/#gradio-interface","title":"Gradio Interface","text":""},{"location":"overview/features/#mcp-server","title":"MCP Server","text":""},{"location":"overview/features/#development-features","title":"Development Features","text":""},{"location":"overview/features/#code-quality","title":"Code Quality","text":""},{"location":"overview/features/#documentation","title":"Documentation","text":""},{"location":"overview/quick-start/","title":"Quick Start","text":"

Get started with DeepCritical in minutes.

"},{"location":"overview/quick-start/#installation","title":"Installation","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync dependencies\nuv sync\n
"},{"location":"overview/quick-start/#run-the-ui","title":"Run the UI","text":"
# Start the Gradio app\nuv run gradio run src/app.py\n

Open your browser to http://localhost:7860.

"},{"location":"overview/quick-start/#basic-usage","title":"Basic Usage","text":""},{"location":"overview/quick-start/#1-authentication-required","title":"1. Authentication (REQUIRED)","text":"

Authentication is mandatory - you must authenticate before using the application. The app will display an error message if you try to use it without authentication.

HuggingFace OAuth Login (Recommended): - Click the \"Sign in with HuggingFace\" button at the top of the app - Your HuggingFace API token will be automatically used for AI inference - No need to manually enter API keys when logged in

Manual API Key (Alternative): - Set environment variable HF_TOKEN or HUGGINGFACE_API_KEY before starting the app - The app will automatically use these tokens if OAuth login is not available - Supports HuggingFace API keys only (OpenAI/Anthropic keys are not used in the current implementation)

"},{"location":"overview/quick-start/#2-start-a-research-query","title":"2. Start a Research Query","text":"
  1. Enter your research question in the chat interface
  2. Text Input: Type your question directly
  3. Image Input: Click the \ud83d\udcf7 icon to upload images (OCR will extract text)
  4. Audio Input: Click the \ud83c\udfa4 icon to record or upload audio (STT will transcribe to text)
  5. Click \"Submit\" or press Enter
  6. Watch the real-time progress as the system:
  7. Generates observations
  8. Identifies knowledge gaps
  9. Searches multiple sources
  10. Evaluates evidence
  11. Synthesizes findings
  12. Review the final research report
  13. Audio Output: If enabled, the final response will include audio synthesis (TTS)

Multimodal Features: - Configure image/audio input and output in the sidebar settings - Image OCR and audio STT/TTS can be enabled/disabled independently - TTS voice and speed can be customized in the Audio Output settings

"},{"location":"overview/quick-start/#3-mcp-integration-optional","title":"3. MCP Integration (Optional)","text":"

Connect DeepCritical to Claude Desktop:

  1. Add to your claude_desktop_config.json: 

    {\n  \"mcpServers\": {\n    \"deepcritical\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    }\n  }\n}\n

  2. Restart Claude Desktop

  3. Use DeepCritical tools directly from Claude Desktop
"},{"location":"overview/quick-start/#available-tools","title":"Available Tools","text":"

Note: The application automatically uses all available search tools (Neo4j, PubMed, ClinicalTrials.gov, Europe PMC, Web search, RAG) based on query analysis. Neo4j knowledge graph search is included by default for biomedical queries.

"},{"location":"overview/quick-start/#next-steps","title":"Next Steps","text":""}]} \ No newline at end of file diff --git a/site/sitemap.xml b/site/sitemap.xml index 04c8e2e..839e4da 100644 --- a/site/sitemap.xml +++ b/site/sitemap.xml @@ -2,138 +2,126 @@ https://deepcritical.github.io/GradioDemo/ - 2025-12-01 + 2025-12-02 - https://deepcritical.github.io/GradioDemo/CONTRIBUTING/ - 2025-12-01 + https://deepcritical.github.io/GradioDemo/LICENSE/ + 2025-12-02 - https://deepcritical.github.io/GradioDemo/LICENSE/ - 2025-12-01 + https://deepcritical.github.io/GradioDemo/MKDOCS_IMPROVEMENTS_ASSESSMENT/ + 2025-12-02 https://deepcritical.github.io/GradioDemo/team/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/api/agents/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/api/models/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/api/orchestrators/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/api/services/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/api/tools/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/architecture/agents/ - 2025-12-01 - - - https://deepcritical.github.io/GradioDemo/architecture/graph-orchestration/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/architecture/graph_orchestration/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/architecture/middleware/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/architecture/orchestrators/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/architecture/services/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/architecture/tools/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/architecture/workflow-diagrams/ - 2025-12-01 - - - https://deepcritical.github.io/GradioDemo/architecture/workflows/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/configuration/ - 2025-12-01 - - - https://deepcritical.github.io/GradioDemo/configuration/CONFIGURATION/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/contributing/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/contributing/code-quality/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/contributing/code-style/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/contributing/error-handling/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/contributing/implementation-patterns/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/contributing/prompt-engineering/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/contributing/testing/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/getting-started/examples/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/getting-started/installation/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/getting-started/mcp-integration/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/getting-started/quick-start/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/overview/architecture/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/overview/features/ - 2025-12-01 + 2025-12-02 https://deepcritical.github.io/GradioDemo/overview/quick-start/ - 2025-12-01 + 2025-12-02 \ No newline at end of file diff --git a/site/team/index.html b/site/team/index.html index e023793..eda114d 100644 --- a/site/team/index.html +++ b/site/team/index.html @@ -1 +1 @@ - Team - The DETERMINATOR
Skip to content

Team

DeepCritical is developed by a team of researchers and developers working on AI-assisted research.

Team Members

ZJ

Mario Aderman

Joseph Pollack

Virat Chauran

Anna Bossler

About

The DeepCritical team met online in the Alzheimer's Critical Literature Review Group in the Hugging Science initiative. We're building the agent framework we want to use for AI-assisted research to turn the vast amounts of clinical data into cures.

Contributing

We welcome contributions! See the Contributing Guide for details.

\ No newline at end of file + Team - The DETERMINATOR
Skip to content

Team

DeepCritical is developed by a team of researchers and developers working on AI-assisted research.

Team Members

ZJ

Mario Aderman

Joseph Pollack

Virat Chauran

Anna Bossler

About

The DeepCritical team met online in the Alzheimer's Critical Literature Review Group in the Hugging Science initiative. We're building the agent framework we want to use for AI-assisted research to turn the vast amounts of clinical data into cures.

Contributing

We welcome contributions! See the Contributing Guide for details.

\ No newline at end of file diff --git a/uv.lock b/uv.lock index 7980915..9d87f4d 100644 --- a/uv.lock +++ b/uv.lock @@ -1226,6 +1226,7 @@ dependencies = [ dev = [ { name = "mkdocs" }, { name = "mkdocs-codeinclude-plugin" }, + { name = "mkdocs-git-revision-date-localized-plugin" }, { name = "mkdocs-material" }, { name = "mkdocs-mermaid2-plugin" }, { name = "mkdocs-minify-plugin" }, @@ -1262,6 +1263,7 @@ modal = [ dev = [ { name = "mkdocs" }, { name = "mkdocs-codeinclude-plugin" }, + { name = "mkdocs-git-revision-date-localized-plugin" }, { name = "mkdocs-material" }, { name = "mkdocs-mermaid2-plugin" }, { name = "mkdocs-minify-plugin" }, @@ -1300,6 +1302,7 @@ requires-dist = [ { name = "md2pdf", specifier = ">=1.0.1" }, { name = "mkdocs", marker = "extra == 'dev'", specifier = ">=1.6.0" }, { name = "mkdocs-codeinclude-plugin", marker = "extra == 'dev'", specifier = ">=0.2.0" }, + { name = "mkdocs-git-revision-date-localized-plugin", marker = "extra == 'dev'", specifier = ">=1.2.0" }, { name = "mkdocs-material", marker = "extra == 'dev'", specifier = ">=9.0.0" }, { name = "mkdocs-mermaid2-plugin", marker = "extra == 'dev'", specifier = ">=1.1.0" }, { name = "mkdocs-minify-plugin", marker = "extra == 'dev'", specifier = ">=0.8.0" }, @@ -1346,6 +1349,7 @@ provides-extras = ["dev", "magentic", "embeddings", "modal"] dev = [ { name = "mkdocs", specifier = ">=1.6.1" }, { name = "mkdocs-codeinclude-plugin", specifier = ">=0.2.1" }, + { name = "mkdocs-git-revision-date-localized-plugin", specifier = ">=1.2.0" }, { name = "mkdocs-material", specifier = ">=9.7.0" }, { name = "mkdocs-mermaid2-plugin", specifier = ">=1.2.3" }, { name = "mkdocs-minify-plugin", specifier = ">=0.8.0" }, @@ -1805,6 +1809,30 @@ wheels = [ { url = "https://files.pythonhosted.org/packages/f7/ec/67fbef5d497f86283db54c22eec6f6140243aae73265799baaaa19cd17fb/ghp_import-2.1.0-py3-none-any.whl", hash = "sha256:8337dd7b50877f163d4c0289bc1f1c7f127550241988d568c1db512c4324a619", size = 11034, upload-time = "2022-05-02T15:47:14.552Z" }, ] +[[package]] +name = "gitdb" +version = "4.0.12" +source = { registry = "https://pypi.org/simple" } +dependencies = [ + { name = "smmap" }, +] +sdist = { url = "https://files.pythonhosted.org/packages/72/94/63b0fc47eb32792c7ba1fe1b694daec9a63620db1e313033d18140c2320a/gitdb-4.0.12.tar.gz", hash = "sha256:5ef71f855d191a3326fcfbc0d5da835f26b13fbcba60c32c21091c349ffdb571", size = 394684, upload-time = "2025-01-02T07:20:46.413Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/a0/61/5c78b91c3143ed5c14207f463aecfc8f9dbb5092fb2869baf37c273b2705/gitdb-4.0.12-py3-none-any.whl", hash = "sha256:67073e15955400952c6565cc3e707c554a4eea2e428946f7a4c162fab9bd9bcf", size = 62794, upload-time = "2025-01-02T07:20:43.624Z" }, +] + +[[package]] +name = "gitpython" +version = "3.1.45" +source = { registry = "https://pypi.org/simple" } +dependencies = [ + { name = "gitdb" }, +] +sdist = { url = "https://files.pythonhosted.org/packages/9a/c8/dd58967d119baab745caec2f9d853297cec1989ec1d63f677d3880632b88/gitpython-3.1.45.tar.gz", hash = "sha256:85b0ee964ceddf211c41b9f27a49086010a190fd8132a24e21f362a4b36a791c", size = 215076, upload-time = "2025-07-24T03:45:54.871Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/01/61/d4b89fec821f72385526e1b9d9a3a0385dda4a72b206d28049e2c7cd39b8/gitpython-3.1.45-py3-none-any.whl", hash = "sha256:8908cb2e02fb3b93b7eb0f2827125cb699869470432cc885f019b8fd0fccff77", size = 208168, upload-time = "2025-07-24T03:45:52.517Z" }, +] + [[package]] name = "google-auth" version = "2.43.0" @@ -3235,6 +3263,21 @@ wheels = [ { url = "https://files.pythonhosted.org/packages/9f/d4/029f984e8d3f3b6b726bd33cafc473b75e9e44c0f7e80a5b29abc466bdea/mkdocs_get_deps-0.2.0-py3-none-any.whl", hash = "sha256:2bf11d0b133e77a0dd036abeeb06dec8775e46efa526dc70667d8863eefc6134", size = 9521, upload-time = "2023-11-20T17:51:08.587Z" }, ] +[[package]] +name = "mkdocs-git-revision-date-localized-plugin" +version = "1.5.0" +source = { registry = "https://pypi.org/simple" } +dependencies = [ + { name = "babel" }, + { name = "gitpython" }, + { name = "mkdocs" }, + { name = "tzdata", marker = "sys_platform == 'win32'" }, +] +sdist = { url = "https://files.pythonhosted.org/packages/0f/c5/1d3c4e6ddae6230b89d09105cb79de711655e3ebd6745f7a92efea0f5160/mkdocs_git_revision_date_localized_plugin-1.5.0.tar.gz", hash = "sha256:17345ccfdf69a1905dc96fb1070dce82d03a1eb6b0d48f958081a7589ce3c248", size = 460697, upload-time = "2025-10-31T16:11:34.44Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/bc/51/fe0e3fdb16f6eed65c9459d12bae6a4e1f0bb4e2228cb037e7907b002678/mkdocs_git_revision_date_localized_plugin-1.5.0-py3-none-any.whl", hash = "sha256:933f9e35a8c135b113f21bb57610d82e9b7bcc71dd34fb06a029053c97e99656", size = 26153, upload-time = "2025-10-31T16:11:32.987Z" }, +] + [[package]] name = "mkdocs-material" version = "9.7.0" @@ -5845,6 +5888,15 @@ wheels = [ { url = "https://files.pythonhosted.org/packages/b7/ce/149a00dd41f10bc29e5921b496af8b574d8413afcd5e30dfa0ed46c2cc5e/six-1.17.0-py2.py3-none-any.whl", hash = "sha256:4721f391ed90541fddacab5acf947aa0d3dc7d27b2e1e8eda2be8970586c3274", size = 11050, upload-time = "2024-12-04T17:35:26.475Z" }, ] +[[package]] +name = "smmap" +version = "5.0.2" +source = { registry = "https://pypi.org/simple" } +sdist = { url = "https://files.pythonhosted.org/packages/44/cd/a040c4b3119bbe532e5b0732286f805445375489fceaec1f48306068ee3b/smmap-5.0.2.tar.gz", hash = "sha256:26ea65a03958fa0c8a1c7e8c7a58fdc77221b8910f6be2131affade476898ad5", size = 22329, upload-time = "2025-01-02T07:14:40.909Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/04/be/d09147ad1ec7934636ad912901c5fd7667e1c858e19d355237db0d0cd5e4/smmap-5.0.2-py3-none-any.whl", hash = "sha256:b30115f0def7d7531d22a0fb6502488d879e75b260a9db4d0819cfb25403af5e", size = 24303, upload-time = "2025-01-02T07:14:38.724Z" }, +] + [[package]] name = "sniffio" version = "1.3.1"