AI-first full-stack framework. Describe your app, let AI build it.
pnpm create fastreact my-app- Frontend: React 19 + Vite + TypeScript + Tailwind CSS v4 + shadcn/ui
- Backend: FastAPI on Modal (serverless Python)
- AI Agent: Autonomous coding agent that builds your app from a description
- Deployment: Vercel (frontend) + Modal (backend)
- Describe your app in plain English
- AI creates a feature list from your description
- AI implements features one by one, testing each
- You review and iterate as it builds
Based on Anthropic's autonomous coding architecture.
- Node.js 18+
- pnpm
- Python 3.12+
- uv (Python package manager)
- Claude Code (logged in)
- Modal CLI (logged in)
- Vercel CLI (logged in)
- Create your app:
pnpm create fastreact my-appThe CLI verifies prerequisites (Claude Code, Modal, Vercel), then prompts for:
- Project name
- App description (plain English)
- Proxy auth tokens (optional)
- Enter the project directory (required):
cd my-app- Run the AI agent:
pnpm run agentThe agent will:
- Session 1 (Initializer): Create
feature_list.jsonfrom your description - Session 2+ (Coding): Implement features one by one
- Monitor progress:
feature_list.json- Track which features are doneclaude-progress.txt- Session notes from the AIgit log- See commits for each feature
cd my-app
pnpm run dev # Start frontend + backend dev servers- Frontend: http://localhost:5173
- Backend: Modal serve (hot-reloading)
my-app/
├── frontend/ # React + Vite + Tailwind
├── backend/ # FastAPI on Modal
├── agent/ # AI coding agent
│ ├── agent.py # Main agent script
│ ├── prompts/ # Customizable prompts
│ └── .env.example # Environment config template
├── app_spec.md # Your app description
├── feature_list.json # AI-generated feature list
└── claude-progress.txt # AI session notes
Edit the prompts in agent/prompts/:
initializer_prompt.md- How features are broken downcoding_prompt.md- How features are implemented
- AI-first: Describe what you want, AI builds it
- Full-stack: React frontend + FastAPI backend
- Serverless: Deploy to Modal with zero infrastructure
- Modern stack: Vite, Tailwind v4, shadcn/ui, TypeScript
- Incremental: AI commits after each feature, easy to review
cd backend
modal deploy modal_app.pycd frontend
vercelSet environment variables in Vercel:
VITE_API_URL- Modal production URLVITE_MODAL_KEY- Proxy auth token IDVITE_MODAL_SECRET- Proxy auth token secret
MIT