// the find
feldroy/air
The first web framework designed for AI to write. Built on Python, FastAPI, Pydantic, and HTMX. By the authors of Two Scoops of Django.
Air is a FastAPI wrapper that adds typed HTML generation (via Python classes called Air Tags), simplified Jinja templating, and Pydantic-backed form handling. It's aimed at developers building server-rendered Python web apps, particularly those using AI coding assistants where the framework's explicit typing means less guessing. From the Two Scoops of Django authors, so the docs and project hygiene are above average.
The Air Tags system — writing HTML as typed Python classes with full autocomplete — is genuinely useful for AI-assisted development and not just marketing; the type system means an LLM can reason about nesting and attributes without hallucinating. The Jinja integration is first-class rather than bolted on, so you can mix template files and code-generated HTML in the same view without friction. Pydantic form validation via dependency injection is a clean pattern that avoids the usual FastAPI form handling boilerplate. The AGENTS.md file and llms-full.txt docs artifact show the authors actually thought through the AI tooling angle rather than just claiming it.
It's alpha at 0.48.x with 48+ changelog entries since 0.3.0 — the API surface is still moving and the caution warning is real; don't build anything you can't afford to partially rewrite. The 'designed for AI to write' pitch is a double-edged sword: it means the abstractions are explicit and verbose, which pays off in tooling but costs in ergonomics when writing by hand. HTMX support is listed as a feature but appears to be just utility helpers rather than any real integration — you're still wiring HTMX yourself. No async database or ORM story included; you get a web layer and nothing else, so you'll be assembling the stack yourself.