🐙 RepoChat — Chat with Any GitHub Repository

RepoChat is an AI-powered RAG (Retrieval-Augmented Generation) application that lets you have a conversation with any public GitHub repository. Understand codebases instantly, explore open issues, and navigate pull requests — all through natural language.

---

🎬 Demo

🎯 Use Cases

• New contributors — Understand an unfamiliar codebase without reading every file
• Open source contributors — Find unassigned issues, explore PRs, identify where to start
• Developers evaluating libraries — Ask questions instead of digging through docs

---

✨ Features

• 🔍 Codebase understanding — Ask about architecture, specific files, functions, and tech stack
• 🐛 Issue exploration — Query open issues, filter by assignee or label
• 🔀 PR navigation — Understand what each PR changes and who authored it
• 💬 Conversational memory — Follow-up questions maintain full chat context
• 🎯 Scope filtering — Search across all content or filter to code / issues / PRs only
• ⚡ Fast inference — Powered by Groq's LPU for near-instant responses
• 📦 Smart caching — Already-loaded repos reload instantly from disk
• 🚀 Demo repos — One-click load for micrograd, Flask, and Express

---

🏗️ Architecture

GitHub Repo URL
      ↓
GitHub API — get_git_tree() [1 API call for full tree]
      ↓
Parallel get_git_blob() + Issues + PRs [ThreadPoolExecutor, 20 workers]
      ↓
Files + Issues + PRs (raw documents with metadata)
      ↓
Code-aware Chunker (RecursiveCharacterTextSplitter)
      ↓
MiniLM Embeddings (HuggingFace)
      ↓
ChromaDB Vector Store (cosine similarity, MMR retrieval)
      ↓
LangChain LCEL Chain (history-aware retriever + stuff documents)
      ↓
Groq LLM (llama-3.1-8b-instant)
      ↓
Streamlit Chat UI

---

🛠️ Tech Stack

Layer	Technology
Frontend	Streamlit
LLM	Groq API — llama-3.1-8b-instant
Embeddings	HuggingFace MiniLM (all-MiniLM-L6-v2)
Vector Store	ChromaDB (MMR retrieval, cosine similarity)
Orchestration	LangChain LCEL
GitHub Integration	PyGithub
Language	Python 3.10+

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/Asifmd45/RepoChat.git
cd RepoChat

2. Create virtual environment

python -m venv venv
venv\Scripts\activate  # Windows
source venv/bin/activate  # Mac/Linux

3. Install dependencies

pip install -r requirements.txt

4. Set up environment variables

Create a .env file in the root:

GITHUB_TOKEN=your_github_personal_access_token
GROQ_API_KEY=your_groq_api_key

• GitHub token: github.com/settings/tokens — select repo scope
• Groq API key: console.groq.com

5. Run the app

streamlit run app.py

---

💬 Example Questions

Codebase:

• What does this project do?
• What is the overall architecture?
• How does authentication work?
• Explain the main entry point

Issues:

• Show me unassigned open issues
• Any good first issues available?
• What bugs are currently reported?

Pull Requests:

• What PRs are currently open?
• Who is actively contributing?

---

📁 Project Structure

RepoChat/
├── app.py                  # Streamlit UI
├── github_ingestor.py      # GitHub API — tree + blob fetching, issues, PRs
├── chunker.py              # Code-aware document chunking
├── vector_store.py         # ChromaDB setup and MMR retrieval
├── rag_chain.py            # LangChain LCEL chain with Groq
├── .env                    # API keys (not committed)
├── .gitignore
├── requirements.txt
└── README.md

---

🔑 How It Works

1. Size check — Tree API call (1 second) counts valid files upfront; repos over 600 files are flagged before ingestion starts
2. Ingestion — get_git_tree(recursive=True) fetches the entire file tree in one API call; file content fetched in parallel via get_git_blob(sha)
3. Chunking — Type-aware strategies: code splits at function boundaries, markdown at headers, issues stay whole
4. Embedding — Each chunk embedded with MiniLM and stored in ChromaDB with metadata (type, source, issue number, assignee)
5. Caching — Already-indexed repos load from disk instantly on repeat visits
6. Retrieval — MMR fetches 20 candidates, returns 8 diverse relevant chunks
7. Generation — History-aware chain reformulates follow-up questions; Groq LLM generates grounded answers with sources

---

⚠️ Limitations

• Only works with public GitHub repositories
• Repos over 600 files are blocked upfront (GitHub API rate limit protection)
• Groq free tier: 6,000 tokens/minute, 500,000 tokens/day
• Only open issues and PRs are indexed (closed/merged not included)

---

👤 Author

Asif MD — github.com/Asifmd45