I love windmill as a way to trivially make python scripts into production-ready endpoints. Thanks for building it!

Idea: Any windmill Python script can take a pydantic model or return a pydantic model without breaking windmill's type inference.

But I was sad to see that the input spec inference only works for basic argument types, given that most ML-focused python developers I know rely on libraries like Pydantic (21k stars), Attrs or built-in dataclasses for defining data specs (in FastAPI (77k stars) for example).

I haven't looked at this for a minute, but I think pydantic supports conversion to json-schema, and this might "just work" for your argument type inference. Then, a python windmill script It would make me very very happy :D

(Note: there was a project geared towards python ML practitioners which does the pydantic bit, and I would love to use except it doesn't handle auth, hosting, etc which is where I see windmill shining).

Thanks for considering this idea.