🔭 GitVision
AI-powered GitHub repository intelligence platform.
GitVision syncs your GitHub repositories, generates AI-powered commit summaries, builds vector embeddings of your codebase, and lets you chat with your code using RAG (Retrieval-Augmented Generation). Think of it as a smart copilot that deeply understands your project's structure and history.
✨ Features
🧠 AI-Powered Insights
- Commit Summaries — Gemini-generated plain-English summaries, produced on demand for the 100 most recent commits
- RAG Code Chat — Ask natural-language questions about your codebase; answers are grounded in your actual source code via vector similarity search
📊 Repository Dashboard
- Commit Activity — Interactive timeline of project commit history
- Tech Stack Breakdown — Language composition with visual percentage bars
- Contributors — Author-level commit analytics
- Stats at a Glance — Stars, forks, branches, files, and total commits
🐛 Issues & Pull Requests
- Live Sync — Issues and PRs fetched from GitHub with full metadata
- Comment Threads — Inline display of GitHub discussion threads with avatars
📂 Code Browser
- Full File Tree — Browse the repository file structure in-app
- Syntax-Highlighted Viewer — Read source code with proper language highlighting
💬 Chat with Your Code
- Contextual Conversations — Multiple chat sessions per project with persistent history
- Vector Search — Queries are matched against 768-dimension embeddings (
qwen/qwen3-embedding-8b via OpenRouter) using HNSW indexing
- Source Citations — Responses reference specific files from your codebase
🏗️ Tech Stack
📁 Project Structure
├── app/ # Next.js App Router pages & API routes
│ ├── (auth)/ # Sign-in / sign-up flows (Clerk)
│ ├── (main)/ # Authenticated app shell
│ │ ├── chat/ # RAG chat interface
│ │ ├── code-viewer/ # Repository file browser
│ │ ├── create-project/ # New project onboarding
│ │ └── dashboard/ # Project analytics dashboard
│ └── api/ # API endpoints (chat, embeddings, inngest, tRPC, webhooks)
├── db/ # Drizzle schema & migrations
├── src/
│ ├── features/ # Feature-based architecture
│ │ ├── auth/ # User & auth utilities
│ │ ├── chat/ # Chat UI components & logic
│ │ ├── dashboard/ # Dashboard widgets & analytics
│ │ ├── landing/ # Marketing landing page
│ │ ├── projects/ # Project management
│ │ └── rag/ # Embedding generation & vector search
│ ├── lib/ # Shared libraries (inngest, trpc, AI clients)
│ └── shared/ # Shared UI components, hooks, utils
New to the codebase? Read docs/architecture.md
first. It covers the layering above, the ingestion and chat flows with
diagrams, and the three structural tensions worth knowing before you change
anything (no transactions, no row-level security, two ownership guards).
Security engineering is real and mostly invisible; the defences and the tests
that pin them are written up in docs/SECURITY.md.
🚀 Getting Started
Prerequisites
- Node.js ≥ 18 (or Bun)
- Neon database (free tier works)
- Clerk account for authentication
- GitHub Personal Access Token — for repo syncing
- Google Gemini API Key — for chat responses and commit summaries
- OpenRouter API Key — for embedding generation
1. Clone the repository
git clone https://github.com/Mayank-Radadiya/GitVision.git
cd GitVision
2. Install dependencies
bun install
The repository is a bun workspace — it ships a bun.lock and no package-lock.json
or yarn.lock. Use bun, or the lockfile will drift from what CI installs.
3. Configure environment variables
Create a .env file in the project root:
# Clerk Authentication
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_...
CLERK_SECRET_KEY=sk_...
NEXT_PUBLIC_CLERK_SIGN_IN_URL=/sign-in
NEXT_PUBLIC_CLERK_SIGN_UP_URL=/sign-up
NEXT_PUBLIC_CLERK_SIGN_IN_FALLBACK_REDIRECT_URL=/dashboard
NEXT_PUBLIC_CLERK_SIGN_UP_FALLBACK_REDIRECT_URL=/dashboard
CLERK_WEBHOOK_SECRET=whsec_...
# Database (Neon Serverless PostgreSQL)
DATABASE_URL=postgresql://...
# GitHub
GITHUB_TOKEN=github_pat_...
# AI / LLM
GEMINI_API_KEY=AIza...
OPENROUTER_API_KEY=sk-or-...
bun run test:e2e needs the Clerk keys to point at a development instance
(pk_test_ / sk_test_). e2e/global-setup.ts refuses anything else before it
calls out, so a test run cannot mint a session against a live one. It creates a
fixture user on first run and writes e2e/.auth/user.json — gitignored, because
that file is a live credential. Every spec loads that state, so the signed-in
specs cannot start without a session. Override the identity with
CLERK_E2E_USER_EMAIL / CLERK_E2E_USER_PASSWORD.
4. Set up the database
# Generate migration files
bun db:generate
# Push schema to Neon
bun db:push
5. Run the development server
# Start Next.js (Turbopack)
bun dev
# In a separate terminal — start Inngest dev server
bun inngest
Open http://localhost:3000 to see the app.
📜 Available Scripts
| Command | Description |
|---|
bun dev | Start Next.js dev server with Turbopack |
bun build | Production build |
bun start | Start production server |
bun lint | Run ESLint |
bun typecheck | Run tsc --noEmit |
bun test | Run the Vitest suite once |
bun test:watch | Vitest in watch mode |
bun test:coverage | Vitest with a coverage report |
bun test:e2e | Run the Playwright end-to-end suite |
bun db:generate | Generate Drizzle migration files |
bun db:init | Create the database and apply migrations |
bun db:push | Init, then push the schema straight to the database |
bun db:migrate | Apply the committed migration files |
bun db:backup | Write a verified SQL dump |
bun db:studio | Open Drizzle Studio (DB GUI) |
bun inngest | Start Inngest dev server for background jobs |
bun clean | Remove build artifacts and caches |
db:push and db:migrate both change the database. db:push reconciles from
the schema files and is the right choice for local development; db:migrate
replays the committed SQL migrations and is the right choice anywhere shared.
🗄️ Database Schema
The app uses a relational schema with vector extensions:
- users — Clerk-synced user profiles with credit system
- projects — GitHub repo metadata, language breakdown, embedding status
- project_files — Full repository file contents with SHA-256 hashing
- code_embeddings — 768-dim
qwen/qwen3-embedding-8b vectors (OpenRouter) with HNSW cosine similarity index
- commits — Commit history with AI-generated summaries
- issues / issue_comments — GitHub issues & PRs, and their comments
- project_chats / chat_messages — Normalized chat history per project
🔒 Authentication
GitVision uses Clerk for authentication with webhook-based user sync:
- Users sign up / sign in via Clerk's hosted UI
- A Clerk webhook syncs user data to the local
users table
app/(main)/layout.tsx calls auth.protect() for every page under /dashboard, /chat and /code-viewer; each API route and tRPC procedure checks the session itself (24 of 24 tRPC procedures use protectedProcedure)
🤖 How RAG Chat Works
1. Project files are streamed from GitHub via Octokit tarball API
2. Files are chunked and embedded using qwen/qwen3-embedding-8b via OpenRouter (768 dims)
3. Embeddings are stored in Neon with pgvector HNSW indexes
4. User query → embed → cosine similarity search → top-k chunks
5. Top chunks + query are sent to Gemini `gemini-2.0-flash-001` as grounded context
6. Streamed AI response with file-path citations
🤝 Contributing
Contributions are welcome! Please follow the standard fork → branch → PR workflow.
- Fork the repository
- Create a feature branch:
git checkout -b feat/amazing-feature
- Commit your changes:
git commit -m 'feat: add amazing feature'
- Push to the branch:
git push origin feat/amazing-feature
- Open a Pull Request
📄 License
Not currently licensed. This repository is private and has no LICENSE file, so
all rights are reserved by default. Add a LICENSE before publishing it anywhere.
