Agent Starter Kit

agent-starter-kit

[](https://github.com/asadullah48/agent-starter-kit/actions/workflows/ci.yml) [](LICENSE)

A minimal, **working** AI agent built on the Claude API -- organized so it stays maintainable once you add more tools, memory, APIs, tests, or teammates, instead of living in one growing Python file. Runs against Claude out of the box, or completely free and local via Ollama.

Live demo

**https://agent-starter-kit-six.vercel.app**

• `/health` -- liveness check
• `/docs` -- interactive OpenAPI docs (try /chat from here)
• POST /chat runs in a safe "demo mode" (explains itself instead of erroring) because no ANTHROPIC_API_KEY is set on the deployment -- a public endpoint with a real key attached would let any visitor spend it. Clone the repo and add your own key (or LLM_PROVIDER=ollama) to try the full agent. Note the deployment is also stateless per-request, so multi-turn memory only works locally, not against the hosted demo.

What's here

One folder = one responsibility:

agent-starter-kit/
??? README.md               Documentation & setup (this file)
??? LICENSE                 MIT
??? requirements.txt         Runtime dependencies
??? requirements-dev.txt      + pytest, for running the test suite
??? .env.example               API keys & environment variables (copy to .env)
??? .gitignore
??? docker-compose.yml         Runs the API in a container
??? Dockerfile
??? vercel.json                 Serverless function config for the live demo
??? main.py                     CLI entry point (chat REPL or --api server)
??? pytest.ini
??? api/index.py                 Vercel entrypoint (re-exports src/api/app.py)
??? .github/workflows/ci.yml     Runs the test suite on every push/PR