LightFARS

Lightweight Fully Automated Research System
_{轻量级全自动研究系统}

简体中文 | English

---

💡 Inspiration: This project is inspired by FARS (Fully Automated Research System), an end-to-end AI research system that autonomously completes the entire research workflow. LightFARS is a lightweight implementation built entirely with LangChain 1.0+.

---

📖 Overview

LightFARS is an end-to-end AI research system that autonomously completes the entire research workflow:

• 💡 Ideation Agent: Literature search, hypothesis generation, research proposal
• 📋 Planning Agent: Task decomposition, experiment design
• 🧪 Experiment Agent: Code generation, experiment execution, data analysis
• ✍️ Writing Agent: Paper writing, report generation

🎯 Key Features

• ✅ LangChain 1.0+ Native: Built entirely with LangChain's latest APIs
  • create_react_agent() - Agent creation
  • @tool decorator - Tool definition
  • ChatPromptTemplate - Prompt templates
  • StateGraph - Workflow orchestration
• ✅ Multi-Agent Architecture: Four specialized agents working collaboratively
• ✅ Shared File System: Agents communicate through file-based coordination (inspired by FARS)
• ✅ JSON-Driven Tasks: Structured task execution with progress tracking
• ✅ Fully Chinese Support: All prompts and outputs in Chinese

🔄 How It Works

[User Input: Research Directions]
              ↓
[Ideation Agent]
  - Search arXiv papers
  - Generate research hypothesis
  - Write research proposal
              ↓
[Planning Agent]
  - Decompose experiment tasks
  - Design evaluation metrics
  - Create task plan (JSON)
              ↓
[Experiment Agent]
  - Generate experiment code
  - Execute experiments
  - Collect results
              ↓
[Writing Agent]
  - Write complete paper
  - Generate Markdown/LaTeX
              ↓
[Final Output: Research Paper]

📦 Installation

1. Create Conda Environment

conda create -n lightfars python=3.11 -y
conda activate lightfars

2. Install Dependencies

cd lightfars
pip install -r requirements.txt

3. Configure Environment Variables

# Copy example config
cp config/.env.example config/.env

# Edit config file with your API keys
# Required: LLM_API_KEY, LLM_BASE_URL, LLM_MODEL_ID

🚀 Quick Start

Run Example Project

# Activate environment
conda activate lightfars

# Run main program
python main.py

Create New Project

# 1. Create project directory structure
mkdir -p projects/my-research/{input,idea/references,exp/results,exp/figures,paper/final,.state,logs,config}

# 2. Write research directions
cat > projects/my-research/input/research_directions.md << EOF
# Research Directions

Describe your research directions here...
EOF

# 3. Modify project_dir in main.py
# project_dir = "projects/my-research"

# 4. Run
python main.py

📁 Project Structure

lightfars/
├── config/                 # Configuration files
│   ├── .env.example       # Environment template
│   └── settings.py        # Settings loader
│
├── src/                   # Source code
│   ├── agents/           # Agent implementations
│   │   ├── ideation.py   # Ideation Agent
│   │   ├── planning.py   # Planning Agent
│   │   ├── experiment.py # Experiment Agent
│   │   └── writing.py    # Writing Agent
│   │
│   ├── tools/            # Tool definitions
│   │   ├── literature.py # Literature search
│   │   └── file_ops.py   # File operations
│   │
│   ├── prompts/          # Prompt templates
│   │   └── templates.py  # All agent prompts
│   │
│   ├── workflows/        # Workflows
│   │   └── research_flow.py  # LangGraph workflow
│   │
│   └── utils/            # Utility functions
│       └── llm.py        # LLM initialization
│
├── projects/             # Project directories
│   └── prompt-engineering-research/  # Example project
│       ├── input/        # Input data
│       ├── idea/         # Ideation outputs
│       ├── exp/          # Experiment outputs
│       └── paper/        # Writing outputs
│
├── main.py               # Main entry point
├── requirements.txt      # Dependencies
└── README.md             # This file

⚙️ Configuration

Supported LLM Providers

• OpenAI: LLM_PROVIDER=openai
• Anthropic: LLM_PROVIDER=anthropic
• DeepSeek: LLM_PROVIDER=deepseek
• Qwen (DashScope): LLM_PROVIDER=openai with DashScope base URL

Environment Variables

# LLM Configuration
LLM_PROVIDER=openai
LLM_API_KEY=your-api-key
LLM_BASE_URL=https://api.openai.com/v1
LLM_MODEL_ID=gpt-4o
LLM_TEMPERATURE=0.0
LLM_MAX_TOKENS=4000

# Search API
ARXIV_API_ENABLED=true
ARXIV_MAX_RESULTS=20

📊 Output Example

After completion, the project directory will contain:

projects/my-research/
├── idea/
│   ├── proposal.md          # Research proposal (10-15 pages)
│   ├── plan.json            # Structured plan
│   └── references/          # Literature library
│
├── exp/
│   ├── task_plan.json       # Task list
│   ├── results/             # Experiment data
│   ├── figures/             # Visualizations
│   └── analysis.md          # Experiment analysis
│
└── paper/
    └── final/
        ├── report.md        # Markdown report
        └── paper.tex        # LaTeX paper

📸 Screenshots

Research Workflow

Experiment Results

Generated Paper

🎨 Customization

Add New Tools

from langchain_core.tools import tool

@tool
def my_custom_tool(param: str) -> str:
    """Tool description
    
    Args:
        param: Parameter description
    
    Returns:
        Return value description
    """
    # Implementation
    return result

Modify Prompts

Edit prompt templates in src/prompts/templates.py.

🤝 Contributing

Issues and Pull Requests are welcome!

📄 License

MIT License

🙏 Acknowledgments

• FARS (Fully Automated Research System) - Inspiration source
• LangChain - LangChain 1.0+ framework

---

_{Built with ❤️ by q198132}