🧠 Codebase Genius

The primary objective of this project is to create an agentic system that accepts a GitHub repository URL and produces quality Markdown documentation. The system is designed to be particularly effective for repositories written in Python and Jac. A key feature is the automatic generation of visual diagrams to represent the codebase's structure and flow.

🎬 Demo Video

Codebase.Genius.mp4

➡️ High-Level Workflow

The process is broken down into a sequence of clear steps. The agent will first understand the "what" and "where" of the code and then dive deeper to understand the "how."

• Clone the Repo: Download the GitHub repository.
• Get File and Folder Structure: Map out the repository's layout.
• Retrieve and Analyze README.md: Read the README for a project overview.
• High-Level Planning: Create a documentation plan using the initial data.
• Iteratively Analyze Code Content: Parse source files to understand the code.
• Generate the Documentation: Create a comprehensive Markdown document.

🏛️ Proposed Architecture: A Multi-Agent System 🤖

To accomplish this, a multi-agent architecture was used. Think of it as a team of specialized AI agents managed with a supervisor.

Components

• Repo Mapper: 🗺️ Analyzes structure and README.
• Code Analyzer: 👨‍💻 Parses and understands source code.
• DocGenie: 📝 Produces the documentation and diagrams.

Agent Responsibilities

• Repo Mapper 📂

  • This agent was made responsible for high-level repository mapping.
  • File Tree Generator: 🌳 A structured view of the file system was built, ignoring unnecessary files and folders (e.g., .git, node_modules, etc.).
  • Readme Summarizer: 📖 This was created to extract a concise summary from the README.md file to provide context for the documentation process.

• Code Analyzer 🔬

  • This agent was designed to perform in-depth code analysis.
  • It uses tools such as Tree-sitter for parsing.
  • It identifies functions, classes, and their relationships.
  • It builds the foundation for understanding code logic and interaction.

• DocGenie ✍️

  • This agent was designed to be responsible for generating documentation.
  • It converts structured code insights into human-readable Markdown.

📥Inputs & Outputs

• Input: ➡️ A GitHub repository URL (public repo for MVP).
• Output: 📄 Markdown (.md) files saved locally, containing comprehensive documentation of the repository.

🚀 Getting Started

Prerequisites

• jaclang Jac and Jaseci
• streamlit
• Mistral API key

Star with Streamlit

1. Backend Setup: Terminal 1

   python -m venv jac-env
   source jac-env/bin/activate  # or `jac-env\Scripts\activate` on Windows 
   pip install -r requirements.txt
   cd backend
   export "MISTRAL_API_KEY=your-api-key-here" > .env
   export 'MISTRAL_API_KEY'="your api key"
   export 'HF_KEY'="your hf key"
   export REQUIRE_AUTH_BY_DEFAULT=false
   jac serve server.jac

2. Frontend Setup: Terminal 2

   source jac-env/bin/activate
   cd frontend
   pip install jac streamlit
   jac streamlit client.jac

🔧 Configuration

The following environment variables can be set in the .env file in the rag-system directory:

• MISTRAL_API_KEY: Your Mistral AI API key.
• HF_KEY : your HuggingFace Access Token Key
• CHUNK_SIZE: The size of the chunks the document is split into (default: 1000).
• CHUNK_OVERLAP: The number of overlapping characters between chunks (default: 100).
• TOP_K: The number of chunks to retrieve from the vector store (default: 10).

🙏 Acknowledgments

• Mistral AI for the language model
• LangChain for the RAG implementation
• Chroma for Vector Database

🔗 API Reference

Your application exposes these main endpoints:

• POST /user/register — Create a new user account
• POST /user/login — Login and get an access token
• POST /walker/upload_file — Upload files (requires authentication)
• POST /walker/interact — Chat with the AI (requires authentication)

Visit http://localhost:8000/docs to see the full API documentation.