I Built a Multi-Agent Starter Kit with LangGraph — 6 Patterns, 5 Providers, One Command

If you've built more than one LangGraph project, you know the drill. Supervisor setup. Provider config. Handoff tools. Persistence. Streaming endpoint. Same boilerplate, different repo. So I stopped rewriting it and packaged the whole thing. LangGraph Starter Kit npx create-langgraph-app Interactive CLI. Pick your provider, pick your patterns, get a project that runs. Or clone the full kit with everything included. 6 Patterns Each one is a standalone app you can use, modify, or delete: Supervisor — central coordinator routes tasks to worker agents Swarm — agents hand off to each other with transfer tools, no central brain Human-in-the-Loop — graph pauses for approval before destructive actions Structured Output — typed JSON responses validated by Zod Research Agent — web search + scraping, supervisor coordinates a researcher and writer RAG — in-memory vector store, semantic retrieval, no external DB 5 Providers LLM_PROVIDER=anthropic ANTHROPIC_API_KEY=sk-ant-... Two lines. Done. OpenAI