v0.7.3

Fixed

• Updated pip install in Dockerfile to use unsafe-best-match strategy for compatibility with torch CPU version.

v0.7.2

Changed

• Added PyTorch CPU extra index URL to dependencies for a smaller final image size.

v0.7.1

Fixed

• Dockerfile bug regarding uv command

v0.7.0

Added

• Groq LLM Provider Support: Integration with Groq as a new LLM provider option
• OpenAI-Compatible Providers: Support for OpenRouter and GitHub Models through OpenAI-compatible interface
• Centralized Logging System: Introduced AllyLogger for unified logging throughout the codebase with improved error visibility
• UV Package Manager: Added uv as Python dependency manager in Dockerfile
• ChromaDB as Core Dependency: Integrated ChromaDB as a fundamental requirement for the project
• Torch & Transformers: Added PyTorch and Transformers libraries to requirements
• Enhanced Help Section: Updated help with previously missing commands

Changed

• UI Overhaul: Complete UI rework with vibrant unified color theme and improved readability
  • Redesigned help panel and input prompt styling
  • Updated output formatting in AgentUI
  • Enhanced code generation UI readability
  • Refactored choice selector using prompt_toolkit library
• Windows Installer: Completely reworked Windows installation process with tweaks for better compatibility
• Linux/Mac Setup: Updated setup scripts and removed logging to stdout
• Dependencies: Updated requirements.txt to latest LangChain and provider package versions, with deduplication
• Configuration Updates:
  • Updated default provider and model settings
  • Revised default temperature settings for providers
• Documentation:
  • Updated README with new provider additions
  • Improved technical documentation for logging and error handling
  • Documented Groq provider and clarified provider/model options
  • Logo color edits
• Codebase Refactoring:
  • Improved CLI class readability and consistency
  • Enhanced DataBaseClient instantiation for better readability
  • Condensed warning and error message display logic
  • Improved invalid directory error handling formatting
  • Clarified working directory notes and improved note handling
• Docling Support: Discontinued docling support
• .gitignore: Updated for uv, pyproject.toml, and Python version management

Fixed

• Ollama Provider Validation: Skip API key/model validation for Ollama provider to prevent unnecessary errors
• Thinking Block Leak: Resolved issue with thinking blocks leaking into output
• UI Overlap Issue: Fixed concurrent tool call permission selectors UI overlap when model launches multiple tool calls per AIMessage
• Error Handling: Improved resilience to errors with better error visibility in logs

v0.6.1

Changed

• Suppressed onnxruntime warnings and added pymupdf-layout
• Fixed a UI bug when waiting for message generation
• (Docling) Enhanced document scraper reusability and file handling
• Updated and pinned Python dependencies

v0.6.0

Added

• Support for OpenAI and NLP Cloud embedding providers
• Configurable scraping method (simple/docling) with CLI and config integration
• 'Simple' and 'docling' scrapers for file extraction
• Dependencies for document scraping: python-docx and pymupdf4llm
• Status messages for RAG and response generation
• LAST_N_TURNS constant for message history management
• BATCH_SIZE constant for batch processing
• Collection name validation with "too short" error handling
• Batch processing and deduplication support for embeddings
• gRPC warning suppression for Google LLM

Changed

• Replaced initial_prompt_suffix with recurring_prompt_suffix in CLI
• Updated default model and provider to Google Gemini
• Refactored web searcher tools and enabled fetch_tool for agents
• Updated agent configuration to use UTF-8 file reading
• Improved BaseAgent command handling and RAG integration
• Refactored and modularized file scraping for RAG embeddings
• Refactored and enhanced Hugging Face embedding function
• Improved handling of embedding and indexing commands
• Updated code generation orchestration for better streaming and error handling
• Updated constants and UI messages for embedding and session changes
• Ensured environment variables are loaded early in main.py
• Improved setup.sh for better POSIX compatibility
• Simplified Dockerfile installation of Python dependencies
• Updated documentation and example environment for embedding and API keys
• Cleaned up configuration formatting and provider/model settings
• Refined Docling setup and dependencies in embedding scrapers
• Tuned embedding database query defaults and constants
• Updated README for improved clarity, instructions, badge styles, and layout
• Tweaked UI and agent code for minor issues
• Reordered imports for consistency

Fixed

• Corrected RAG ToolMessage type and implemented filtering of RAG messages
• Fixed embedding function type signatures and improved OllamaEmbedder
• Replaced global flag with file-based setup check for thread safety in Docling
• Fixed various README issues
• Removed temporary message in CLI

v0.5.21

Fixed

• Fixed Docker hub image name in Github actions.

v0.5.2

Added:

• Auto pushing to Docker hub on new release using Github actions.

v0.5.1

Fixed:

• Failed model change request used to exit the program with an error message. New behavior allows the user to enter in a correct model name or use the oldest viable one before change if any.

v0.5.0

Added

• Docker support

Fixed

• bug: Not being able to change model mid conversation due to a missing argument.