SciLink

AI-Powered Scientific Research Automation Platform

SciLink employs a system of intelligent agents to automate experimental design, data analysis, and iterative optimization workflows. Built around large language models with domain-specific tools, these agents act as AI research partners that can plan experiments, analyze results across multiple modalities, and suggest optimal next steps.

---

Overview

SciLink provides three complementary agent systems that cover the full scientific research cycle:

System	Purpose	Key Capabilities
Planning Agents	Experimental design & optimization	Hypothesis generation, Bayesian optimization, literature-aware planning
Analysis Agents	Multi-modal data analysis	Microscopy, spectroscopy, particle segmentation, curve fitting
Simulation Agents	Computational modeling	DFT calculations, classical MD (LAMMPS), structure recommendations

All systems support configurable autonomy levels: from co-pilot mode where humans lead and AI assists, to fully autonomous operation where the agent chains all tools independently.

---

Installation

pip install scilink

Environment Variables

Set API keys for your preferred LLM provider:

# Google Gemini (default)
export GEMINI_API_KEY="your-key"

# OpenAI
export OPENAI_API_KEY="your-key"

# Anthropic
export ANTHROPIC_API_KEY="your-key"

# Internal proxy (if applicable)
export SCILINK_API_KEY="your-key"

---

Quick Start

Planning a New Experiment

# Interactive planning session
scilink plan

# With specific settings
scilink plan --autonomy supervised --data-dir ./results --knowledge-dir ./papers

Analyzing Experimental Data

# Interactive analysis session
scilink analyze

# With data file
scilink analyze --data ./sample.tif --metadata ./metadata.json

---

---

Python API

from scilink.agents.planning_agents import PlanningAgent, BOAgent
from scilink.agents.exp_agents import AnalysisOrchestratorAgent, AnalysisMode

# Generate an experimental plan
planner = PlanningAgent(model_name="gemini-3-pro-preview")
plan = planner.propose_experiments(
    objective="Optimize lithium extraction yield",
    knowledge_paths=["./literature/"],
    primary_data_set={"file_path": "./composition_data.xlsx"}
)

# Analyze microscopy data
analyzer = AnalysisOrchestratorAgent(
    analysis_mode=AnalysisMode.SUPERVISED
)
result = analyzer.chat("Analyze ./stem_image.tif and generate scientific claims")

---

Planning Agents

The Planning Agents module provides an AI-powered research orchestration system that automates experimental design, data analysis, and iterative optimization workflows.

Architecture

PlanningOrchestratorAgent (main coordinator)
├── PlanningAgent (scientific strategy)
│   ├── Dual KnowledgeBase (Docs KB + Code KB)
│   ├── RAG Engine (retrieval-augmented generation)
│   └── Literature Agent (external search)
├── ScalarizerAgent (raw data → scalar metrics)
└── BOAgent (Bayesian optimization)

Agent	Purpose
PlanningOrchestratorAgent	Coordinates the full experimental workflow via natural language
PlanningAgent	Generates experimental strategies using dual knowledge bases
ScalarizerAgent	Converts raw data (CSV, Excel) into optimization-ready metrics
BOAgent	Suggests optimal parameters via Bayesian Optimization

Autonomy Levels

• Co-Pilot (default): Human leads, AI assists. Reviews every step.
• Supervised: AI leads, human reviews plans/code only.
• Autonomous: Full autonomy, no human review.

CLI Usage

# Start interactive planning session
scilink plan

# Supervised mode with workspace config
scilink plan --autonomy supervised \
  --data-dir ./experimental_results \
  --knowledge-dir ./papers \
  --code-dir ./opentrons_api

# Use a specific model
scilink plan --model claude-opus-4-5

Interactive Session Example

$ scilink plan

📋 What's your research objective?
Your objective: Optimize lithium extraction from brine

🔧 Initializing agent...
✅ Agent ready!

============================================================
💬 CHAT SESSION STARTED
============================================================

👤 You: Generate a plan using papers in ./literature/

🤖 Agent: I'll generate an experimental plan using your literature.

  ⚡ Tool: Generating Initial Plan...
    📚 Knowledge sources: ['./literature/']
    ✅ Retrieved 8 document chunks.

============================================================
✅ PROPOSED EXPERIMENTAL PLAN
============================================================

🔬 EXPERIMENT 1: pH-Controlled Selective Precipitation
--------------------------------------------------------------------------------
> 🎯 Hypothesis:
> Adjusting pH to 10-11 will selectively precipitate Mg(OH)₂ while retaining Li⁺

--- 🧪 Experimental Steps ---
 1. Prepare 50mL aliquots of brine sample
 2. Add NaOH dropwise while monitoring pH
 3. Filter precipitate through 0.45μm membrane
 4. Analyze filtrate via ICP-OES

📝 Press [ENTER] to approve or type feedback:

👤 You: Add implementation code using ./opentrons_api/

🤖 Agent: [calls generate_implementation_code]
    → Builds Code KB from ./opentrons_api/
    → Maps steps to API calls
    → Generates Python scripts
    ✅ Scripts saved to ./output_scripts/

👤 You: Analyze ./results/batch_001.csv and run optimization

🤖 Agent: [calls analyze_file]
    → Generates analysis script
    → Returns: {"metrics": {"yield": 78.5}}

  [calls run_optimization]
    → Bayesian Optimization with 3 data points
    → Returns: {"recommended_parameters": {"temp": 85.2, "pH": 6.8}}

👤 You: /quit
👋 Session saved at: ./campaign_session

CLI Commands

Command	Description
/help	Show available commands
/tools	List all available agent tools
/files	List files in workspace
/state	Show current agent state
/autonomy [level]	Show or change autonomy level
/checkpoint	Save session checkpoint
/quit	Exit session

Python API

Using the Orchestrator

from scilink.agents.planning_agents.planning_orchestrator import (
    PlanningOrchestratorAgent, 
    AutonomyLevel
)

orchestrator = PlanningOrchestratorAgent(
    objective="Optimize reaction yield",
    autonomy_level=AutonomyLevel.SUPERVISED,
    data_dir="./experimental_results",
    knowledge_dir="./papers"
)

response = orchestrator.chat("Generate initial plan and analyze batch_001.csv")

Using Individual Agents

PlanningAgent - Experimental Design

from scilink.agents.planning_agents import PlanningAgent

agent = PlanningAgent(model_name="gemini-3-pro-preview")

plan = agent.propose_experiments(
    objective="Screen precipitation conditions for magnesium recovery",
    knowledge_paths=["./literature/", "./protocols.pdf"],
    code_paths=["./opentrons_api/"],
    primary_data_set={"file_path": "./composition_data.xlsx"},
    enable_human_feedback=True
)

# Iterate based on results
updated_state = agent.update_plan_with_results(
    results=["./results/batch_001.csv", "./plots/yield_curve.png"]
)

ScalarizerAgent - Data Analysis

from scilink.agents.planning_agents import ScalarizerAgent

scalarizer = ScalarizerAgent(model_name="gemini-3-pro-preview")

result = scalarizer.scalarize(
    data_path="./data/hplc_run_001.csv",
    objective_query="Calculate peak area and purity percentage",
    enable_human_review=True
)

print(f"Metrics: {result['metrics']}")
# {'peak_area': 12504.2, 'purity_percent': 98.5}

BOAgent - Bayesian Optimization

from scilink.agents.planning_agents import BOAgent

bo = BOAgent(model_name="gemini-3-pro-preview")

result = bo.run_optimization_loop(
    data_path="./optimization_data.csv",
    objective_text="Maximize yield while minimizing cost",
    input_cols=["Temperature", "pH", "Concentration"],
    input_bounds=[[20, 80], [6, 10], [0.1, 2.0]],
    target_cols=["Yield"],
    batch_size=1
)

print(f"Next parameters: {result['next_parameters']}")
# {'Temperature': 65.2, 'pH': 8.3, 'Concentration': 1.2}

---

Experimental Analysis Agents

The Analysis Agents module provides automated scientific data analysis across multiple modalities—microscopy, spectroscopy, particle segmentation, and curve fitting.

Architecture

AnalysisOrchestratorAgent (main coordinator)
├── FFTMicroscopyAnalysisAgent (ID: 0)
├── SAMMicroscopyAnalysisAgent (ID: 1)
├── HyperspectralAnalysisAgent (ID: 2)
└── CurveFittingAgent (ID: 3)

ID	Agent	Use Case
0	FFTMicroscopyAnalysisAgent	Microstructure via FFT/NMF—grains, phases, atomic-resolution
1	SAMMicroscopyAnalysisAgent	Particle segmentation—counting, size distributions
2	HyperspectralAnalysisAgent	Spectroscopic datacubes—EELS-SI, EDS, Raman imaging
3	CurveFittingAgent	1D fitting—XRD, UV-Vis, PL, DSC, kinetics

Autonomy Levels

• Co-Pilot (default): Human leads, AI assists. Reviews agent selection.
• Supervised: AI leads, human approves major decisions.
• Autonomous: Full autonomy for complete analysis pipelines.

CLI Usage

# Start interactive analysis session
scilink analyze

# With data and metadata
scilink analyze --data ./sample.tif --metadata ./metadata.json

# Autonomous mode
scilink analyze --mode autonomous --data ./spectrum.npy

Interactive Session Example

$ scilink analyze --data ./stem_image.tif

============================================================
🔬 SCILINK ANALYSIS ORCHESTRATOR
============================================================

👤 You: Examine my data and suggest an analysis approach

🤖 Agent: I'll examine your data file.

  ⚡ Tool: Examining data at ./stem_image.tif...

  • Type: microscopy
  • Shape: 2048 x 2048 pixels
  • Suggested agents: FFTMicroscopyAnalysisAgent (0) or SAMMicroscopyAnalysisAgent (1)

👤 You: Preview the image and select the best agent

🤖 Agent: Looking at the image, I can see atomic-resolution features with a crystalline lattice.

  ⚡ Tool: Setting agent to 0...

Selected **FFTMicroscopyAnalysisAgent** for microstructure analysis.

👤 You: Convert this to metadata: HAADF-STEM of MoS2, 50nm FOV, 300kV

🤖 Agent: ⚡ Tool: Converting metadata...
    ✅ Metadata saved

👤 You: Run the analysis

🤖 Agent: ⚡ Tool: Running analysis...
    Analysis ID: stem_image_FFT_20250202_143215_001

**Detailed Analysis:**
The HAADF-STEM image reveals MoS2 with predominantly 2H phase structure.
FFT analysis identified four distinct spatial frequency patterns...

**Scientific Claims Generated:** 3

👤 You: What follow-up measurements do you recommend?

🤖 Agent: 
1. **[Priority 1] EELS Spectrum Imaging** - Target sulfur vacancy clusters
2. **[Priority 2] 4D-STEM Strain Mapping** - Quantify grain boundary strain
3. **[Priority 3] Time-Series Imaging** - Assess defect evolution

CLI Commands

Command	Description
/help	Show available commands
/tools	List orchestrator tools
/agents	List analysis agents with descriptions
/status	Show session state
/mode [level]	Show or change analysis mode
/checkpoint	Save checkpoint
/schema	Show metadata JSON schema
/quit	Exit session

Python API

Using the Orchestrator

from scilink.agents.exp_agents import AnalysisOrchestratorAgent, AnalysisMode

orchestrator = AnalysisOrchestratorAgent(
    base_dir="./my_analysis",
    analysis_mode=AnalysisMode.SUPERVISED
)

response = orchestrator.chat("Examine ./data/sample.tif")
response = orchestrator.chat("Select agent 0 and run analysis")

Using Individual Agents

FFTMicroscopyAnalysisAgent

from scilink.agents.exp_agents import FFTMicroscopyAnalysisAgent

agent = FFTMicroscopyAnalysisAgent(
    output_dir="./fft_output",
    enable_human_feedback=True
)

# Single image
result = agent.analyze("sample.tif", system_info=metadata)

# Batch/series
result = agent.analyze(
    ["frame_001.tif", "frame_002.tif"],
    series_metadata={"series_type": "time", "values": [0, 10], "unit": "s"}
)

# Get recommendations
recommendations = agent.recommend_measurements(analysis_result=result)

SAMMicroscopyAnalysisAgent

from scilink.agents.exp_agents import SAMMicroscopyAnalysisAgent

agent = SAMMicroscopyAnalysisAgent(
    output_dir="./sam_output",
    sam_settings={"min_area": 100, "sam_parameters": "sensitive"}
)

result = agent.analyze("nanoparticles.tif")
print(f"Particles: {result['summary']['successful']}")
print(f"Mean area: {result['statistics']['mean_area_pixels']:.1f} px²")

HyperspectralAnalysisAgent

from scilink.agents.exp_agents import HyperspectralAnalysisAgent

agent = HyperspectralAnalysisAgent(
    output_dir="./hyperspectral_output",
    run_preprocessing=True
)

# 3D datacube: (height, width, energy_channels)
result = agent.analyze(
    "eels_spectrum_image.npy",
    system_info={"experiment": {"technique": "EELS-SI"}},
    structure_image_path="haadf_reference.tif"  # Optional correlation
)

CurveFittingAgent

from scilink.agents.exp_agents import CurveFittingAgent

agent = CurveFittingAgent(
    output_dir="./curve_output",
    use_literature=True,  # Search for fitting models
    r2_threshold=0.95
)

result = agent.analyze(
    "pl_spectrum.csv",
    system_info={"experiment": {"technique": "Photoluminescence"}},
    hints="Focus on band-edge emission"
)

print(f"Model: {result['model_type']}")
print(f"R²: {result['fit_quality']['r_squared']:.4f}")

# Series with trend analysis
result = agent.analyze(
    ["pl_300K.csv", "pl_350K.csv", "pl_400K.csv"],
    series_metadata={"series_type": "temperature", "values": [300, 350, 400], "unit": "K"}
)

Metadata Conversion

from scilink.agents.exp_agents import generate_metadata_json_from_text

# Convert natural language to structured metadata
metadata = generate_metadata_json_from_text("./experiment_notes.txt")

# Input: "HAADF-STEM of MoS2 monolayer, 50nm FOV, 300kV"
# Output: {"experiment_type": "Microscopy", "experiment": {"technique": "HAADF-STEM"}, ...}

---

Novelty Assessment: From Data to Discovery

One of SciLink's key features is automated novelty assessment - instantly checking your experimental findings against the scientific literature to identify what's genuinely new.

This is powered by integration with FutureHouse AI agents, which provide deep literature search and scientific reasoning capabilities. The integration enables SciLink to search across thousands of scientific papers and assess whether your findings have been previously reported.

Interactive Example

👤 You: Analyze the STEM images and generate scientific claims

🤖 Agent: 
  ⚡ Running analysis...
     → Analysis ID: stem_MoS2_FFT_20250203_001

  **Scientific Claims Generated:** 3
  • MoS2 exhibits mixed 2H/1T phase coexistence at room temperature
  • Sulfur vacancy density at grain boundaries is 3.2 × 10¹³ cm⁻²
  • 1T phase domains are localized within 5nm of grain boundaries

👤 You: Assess novelty of these claims

🤖 Agent:
  ⚡ Searching literature via FutureHouse...

  📚 [Score 2/5] Mixed 2H/1T phase coexistence
     → Well-documented (Li et al. 2018, Wang et al. 2020)

  🤔 [Score 3/5] Sulfur vacancy density of 3.2 × 10¹³ cm⁻²
     → Similar measurements exist at different conditions

  🌟 [Score 4/5] 1T phase localized within 5nm of grain boundaries
     → Limited prior reports; spatial correlation not established

  Summary: 1 HIGH-NOVELTY finding identified

👤 You: What experiments should I run next?

🤖 Agent:
  ⚡ Generating novelty-informed recommendations...

  [Priority 1] Confirm 1T localization near grain boundaries (Score 4/5)
     → Statistical analysis across n>20 boundaries, EELS mapping

  [Priority 2] Differentiate vacancy density from prior work (Score 3/5)
     → Document methodology differences vs. Zhou et al. 2021

  💡 High-novelty findings prioritized for validation experiments

The Discovery Acceleration Loop

1. Analysis Agent processes data → generates scientific claims with searchable keywords
2. Novelty Assessment searches literature → scores each claim (1-5)
3. Recommendations prioritized by novelty → validation experiments for novel findings

Without SciLink: Days of manual analysis and literature searching
With SciLink: Know what's novel in minutes - while your experiment is still running

---

Output Structure

Planning Session

campaign_session/
├── optimization_data.csv      # Accumulated experimental data
├── plan.json                  # Current experimental plan
├── plan.html                  # Rendered plan visualization
├── checkpoint.json            # Session state for restoration
└── output_scripts/            # Generated automation code

Analysis Session

analysis_session/
├── results/
│   └── analysis_{dataset}_{agent}_{timestamp}/
│       ├── metadata_used.json
│       ├── analysis_results.json
│       ├── visualizations/
│       └── report.html
├── chat_history.json
└── checkpoint.json

---

Simulation Agents (Coming Soon)

The Simulation Agents module provides AI-powered computational modeling capabilities, bridging experimental observations with atomistic simulations.

Planned Capabilities

Agent	Purpose
DFTAgent	Density Functional Theory workflow automation
MDAgent	Classical molecular dynamics simulations via LAMMPS
SimulationRecommendationAgent	Recommends structures and simulation objectives based on experimental analysis (within available DFT/MD methods)

Key Features (In Development)

• Experiment-to-Simulation Pipeline: Automatically generate simulation input structures from microscopy analysis
• Defect Modeling: Create supercells with point defects, grain boundaries, and interfaces identified in images
• DFT Calculations: Electronic structure, formation energies, and spectroscopic signatures
• Classical MD Simulations: Large-scale dynamics, thermal properties, mechanical response via LAMMPS

Integration with Analysis Agents

The Simulation Agents will integrate directly with the Analysis Agents. Experimental analysis and interpretation will be used to recommend structures and simulation objectives that provide deeper insight into observed phenomena:

Note: This module is currently being refactored. Check back for updates.