⚡ RepoDocs

Chat with any GitHub repository using AI. Paste a URL, ask questions, and get precise answers with exact file and line citations.

---

📖 What is RepoDocs?

RepoDocs is a lightning-fast, highly accurate RAG-powered codebase assistant. You paste a public GitHub repo URL — it indexes the entire codebase, embeds it into a local vector store, and lets you chat with it naturally without hallucinations.

Ask things like:

• "How does the authentication middleware work?"
• "Where are API rate limits handled?"
• "Explain the folder structure"
• "What does the main entry point do?"

Every single answer cites the exact file path and line numbers it pulled from.

---

🛠️ Tech Stack

RepoDocs is built entirely on a modern, lightweight open-source stack, avoiding expensive closed-source APIs where possible.

Layer	Technology
Frontend	React 18 · Vite · Tailwind CSS
Backend	Python · FastAPI
RAG Pipeline	LangChain · FAISS (Local Vector Database)
Embeddings	Hugging Face Inference API (sentence-transformers/all-MiniLM-L6-v2)
LLM Inference	Groq (llama-3.1-8b-instant)
Ingestion	GitHub REST API

---

✨ Features

• ⚡ Instant Indexing — Fetches and embeds any public repo in seconds using parallel processing.
• 🎯 Source Citations — Every answer cites the exact file path and line range it came from. No hallucinations.
• 🌊 Streaming Responses — Answers stream in real-time via Server-Sent Events (SSE).
• 🗂️ File Tree Sidebar — Browse all indexed files within the chat interface.
• 💡 Follow-up Chips — Suggested contextual questions after each answer to keep the exploration going.
• 🔒 Session Isolation — Each repo gets its own local FAISS index, cleaned up entirely on demand.

---

🏗️ Architecture

graph TD
    User([User]) --> |Paste URL| Frontend[React Frontend Vite]
    
    subgraph Backend [FastAPI Backend]
        API_Ingest[POST /api/ingest]
        API_Chat[POST /api/chat]
        
        GitHub[GitHub API]
        HuggingFace[Hugging Face API]
        FAISS[(FAISS Vector Store)]
        Groq[Groq API Llama 3]
    end
    
    Frontend --> API_Ingest
    API_Ingest --> |Fetch repo tree| GitHub
    API_Ingest --> |Generate Embeddings| HuggingFace
    API_Ingest --> |Store Vectors| FAISS
    
    Frontend --> API_Chat
    API_Chat --> |Similarity Search| FAISS
    API_Chat --> |Top 5 Chunks + Prompt| Groq
    API_Chat -.-> |Streamed Response| Frontend

Loading

---

📸 Screenshots

Indexing Repository	Chat Interface

---

🚀 Getting Started

Prerequisites

• Python 3.11+
• Node.js 18+
• Groq API Key
• Hugging Face Access Token (Make sure the token has "Inference" permissions enabled)

1. Local Setup

Clone the repo

git clone https://github.com/Cypher-redeye/RepoDocs.git
cd RepoDocs

Backend Configuration

cd backend
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt

Set Environment Variables

cp .env.example .env

Add your API keys to the .env file in the root directory:

GROQ_API_KEY=gsk_...
HF_TOKEN=hf_...
GITHUB_TOKEN=ghp_...           # Optional: increases GitHub API rate limit from 60 to 5000/hr
MAX_FILES=500                  # Max files to index per repo

Run the backend

uvicorn main:app --reload --port 8000

Frontend Configuration

cd ../frontend
npm install
npm run dev

App runs at http://localhost:5173

---

2. Docker Setup (Recommended)

You can easily run the entire stack with a single command using Docker. This setup automatically maps a persistent volume for the FAISS database.

docker compose up --build -d

• Frontend → http://localhost:3000
• Backend → http://localhost:8000

---

📚 API Reference

POST /api/ingest

Fetch, chunk, and embed a GitHub repository.

curl -X POST http://localhost:8000/api/ingest \
  -H "Content-Type: application/json" \
  -d '{"repo_url": "https://github.com/tiangolo/fastapi"}'

GET /api/status/{session_id}

Check ingestion progress in real-time.

POST /api/chat

Ask a question about the indexed codebase. Streams an SSE response containing source chunks, streaming tokens, and follow-up suggestions.

DELETE /api/session/{session_id}

Clean up a session and permanently delete its vector store from disk.

---

⚠️ Limitations

• Public repos only — private repos require a GitHub token with appropriate repository scopes.
• 500 file limit — to respect API limits, repos larger than 500 files are currently truncated.
• Stateless Cloud Deployments — if deployed to free serverless platforms (like Render's free tier), the local FAISS index will be wiped upon server spin-down unless a persistent disk is attached.

---

👨‍💻 Built By

Om — GitHub · LinkedIn

Part of a series of AI-powered developer tools. Also check out FairScan — an AI Bias Auditing Tool.

---

📄 License

MIT License — see LICENSE for details.