Local GitHub RAG

Local GitHub RAG is a Streamlit application that helps developers understand codebases through natural language queries using local LLMs. It clones GitHub repositories locally and implements RAG (Retrieval Augmented Generation) to provide context-aware responses about your code.

Features

• 🔍 Clone and analyze GitHub repositories locally
• 💬 Ask questions about the codebase in natural language
• 🤖 Powered by local LLMs through Ollama
• 📝 Optional GitHub issues integration
• 🗄️ Persistent vector storage for quick repository switching
• 🔄 Support for multiple repositories
• 🎯 Efficient code context retrieval using multi-query generation

Prerequisites

• Python 3.8+
• Ollama installed locally
• Git

Required Models

The following models need to be available through Ollama:

• qwen2.5-coder:14b for code understanding and response generation
• nomic-embed-text for text embeddings

Installation

# Clone the repository
git clone git@github.com:pixelsoccupied/local-github-rag.git
cd local-github-rag

# Create and activate virtual environment
python -m venv .venv
source .venv/bin/activate  # On macOS/Linux

# Install dependencies
pip install -r requirements.txt

# Start the server
streamlit run main.py

Usage

1. Start the Streamlit application:

streamlit run app.py

2. Choose one of three options:

   • Process a new repository by entering a GitHub URL
   • Use an already downloaded repository from the ./repos directory
   • Switch to a previously processed repository

3. Optional: Include GitHub issues in the analysis by providing a GitHub token

4. Start asking questions about your codebase!

Environment Variables

• GITHUB_TOKEN (optional): For fetching GitHub issues and higher API rate limits

How It Works

1. Repository Processing:

   • Clones the GitHub repository locally
   • Processes code files and optionally GitHub issues
   • Splits content into chunks for efficient retrieval
   • Creates embeddings using the nomic-embed-text model
   • Stores vectors in a local Chroma database

2. Question Answering:

   • Uses multi-query retrieval to find relevant code contexts
   • Generates comprehensive responses using the Qwen coder model
   • Provides specific code examples and issue references when applicable

Supported File Types

• Python (.py)
• Markdown (.md)
• Text (.txt)
• Go (.go)
• JavaScript (.js)
• Java (.java)
• C++ (.cpp, .h)
• Rust (.rs)
• AsciiDoc (.adoc)

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT License

Acknowledgments

• Built with Streamlit
• Powered by Ollama
• Uses LangChain for RAG implementation
• Vector storage by Chroma