AI appsOpen source

Windlass

Open-source FastAPI starter for agent apps with tools and human approval

About this starter

Windlass is a Python backend for products built around agents that call tools. It uses FastAPI with LangGraph for agent flows, Postgres for state checkpoints, a tool registry with typed inputs, and an approval step so a human confirms risky actions like sending email. A small React front end shows runs step by step, with token counts and timings.

The maintainers are opinionated about observability: every run is stored and can be replayed with a different model for comparison. The limitation: it is not a full SaaS kit. Auth is a single API key, there is no billing and no teams, so you will pair it with another starter or build those parts yourself.

Highlights

  • 01Human approval step for risky tool calls
  • 02Replay any run against a different model
  • 03Step-by-step run viewer with token counts

Tags

Related starters

More in AI apps
  • Open source

    AI writing assistant starter: web app plus a browser extension, MIT licensed

    Free4.0
    • browser extension
    • writing
  • Wrenfield

    AI apps

    Subscription

    Django pipeline starter for AI document extraction, priced per page

    $29/mo4.3
    • extraction
    • ocr
  • Ragbag

    AI apps

    One-time

    Next.js RAG starter: upload documents, chat with citations, bill by usage

    $199 once4.6
    • rag
    • pgvector
Live demo · Starter Bench— View template →