🚢 RepoHarbor

RepoHarbor lets you chat with a GitHub repository.
Provide a repo URL, the backend indexes its files, generates embeddings, and the app retrieves relevant code to answer questions about the repository.

---

✨ Features

• Index a public GitHub repository
• Chunk repository files and generate embeddings
• Store embeddings and metadata in MongoDB
• Retrieve relevant code snippets using vector similarity
• Ask questions about a repository through a chat interface
• Context-aware responses generated using an LLM
• Persistent conversations per repository

---

🏗 Architecture

User
 │
 ▼
Frontend (React + Vite)
 │
 │  API Requests
 ▼
Backend (Node.js + Express)
 │
 ├── GitHub API
 │     Fetch repository files
 │
 ├── Chunking + Embedding
 │     Gemini embeddings
 │
 ├── MongoDB
 │     Store:
 │     • repositories
 │     • code chunks
 │     • embeddings
 │
 └── LLM (Groq)
       Generate response using retrieved context

Flow:

Repository URL
      │
      ▼
Fetch repo files
      │
      ▼
Chunk code files
      │
      ▼
Generate embeddings
      │
      ▼
Store in MongoDB
      │
      ▼
User question
      │
      ▼
Vector search
      │
      ▼
Send context + question to LLM
      │
      ▼
Return answer

---

🧰 Tech Stack

Layer	Tools
Frontend	React
Build Tool	Vite
Styling	TailwindCSS
State	Zustand
Backend	Node.js
API Framework	Express
Language	TypeScript
Database	MongoDB
Embeddings	Google Gemini
LLM	Groq
Repository Data	GitHub API

---

📂 Repository Structure

repoharbor
│
├── backend
│   ├── src
│   │   ├── config
│   │   ├── controllers
│   │   ├── middlewares
│   │   ├── models
│   │   ├── routes
│   │   ├── services
│   │   ├── utils
│   │   └── index.ts
│   │
│   ├── dist
│   ├── package.json
│   ├── tsconfig.json
│   └── .env
│
├── frontend
│   ├── src
│   │   ├── api
│   │   ├── assets
│   │   ├── components
│   │   ├── features
│   │   ├── layouts
│   │   ├── lib
│   │   ├── pages
│   │   ├── store
│   │   ├── App.tsx
│   │   ├── main.tsx
│   │   └── index.css
│   │
│   ├── public
│   ├── dist
│   ├── index.html
│   └── .env
│
└── README.md

---

⚙ Environment Variables

Backend .env

MONGO_URI=
GITHUB_TOKEN=
GROQ_API_KEY=
FRONTEND_URL=

SESSION_SECRET=
ACCESS_JWT_SECRET=
REFRESH_JWT_SECRET=

ACCESS_JWT_EXPIRATION=
REFRESH_JWT_EXPIRATION=

GITHUB_CLIENT_ID=
GITHUB_CLIENT_SECRET=
GITHUB_CALLBACK_URL=

Frontend .env

VITE_API_URL=

---

🚀 Running Locally

Clone the repo

git clone https://github.com/yourusername/repoharbor.git
cd repoharbor

Install dependencies

Backend

cd backend
npm install

Frontend

cd frontend
npm install

Run backend

npm run dev

Run frontend

npm run dev

Open:

http://localhost:5173

---

🤝 Contributing

If you want to experiment or improve something, feel free to open a PR or issue.
No strict process — just keep it clean and reasonable.

⚠️ Heads up: Gemini Embeddings Rate Limits

If you're running RepoHarbor locally, you'll likely hit Gemini's embedding API rate limits pretty fast (it's strict on the free tier). Instead of waiting around, swap in one of these alternatives:

• Nomic Embed — open source, strong performance, runs fully local via Ollama. Zero API calls, zero limits.
• Ollama + mxbai-embed-large — pull and run embeddings locally in one command. Dead simple setup.
• OpenAI text-embedding-3-small — generous free credits, easy drop-in swap, solid quality.
• Voyage AI — made for code/retrieval tasks specifically, free tier is decent for local dev.

Swapping should be a one-liner — just match the output dimensions if you're using a vector store.