You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In the README.md of the python subtree there's this sentence:
To learn more about TypeChat, visit the documentation which includes more information on TypeChat and how to get started.
However, the docs found at that link are exclusively aimed at a TypeScript audience, and give no specifics about e.g. how to write a schema in Python. I had to glean that from the examples, which is far from ideal. I had hoped for something like a Python version of Basic TypeScript Usage.
I put together #237 a bit ago - it's not merged yet since the Python support is not yet published. But I guess I can merge it and just point out it's not yet published.
I put together #237 a bit ago - it's not merged yet since the Python support is not yet published. But I guess I can merge it and just point out it's not yet published.
Oh, that would definitely have saved me a buch of headscratching and spelunking. :-) I just gave it a quick review. Maybe it needn't be linked from the "general" TypeChat front page until it's published, but it ought to be at least reachable from the Python-specific README.ml (which I used as my starting point).
Python-specific documentation for TypeChat would help a lot — the TypeScript-first design means that Python users often have to mentally translate examples, and the translation isn't always obvious (especially for type schema definitions where TypeScript has more expressive type literal syntax).
One thing worth documenting specifically for Python: how TypeChat handles Pydantic models vs Python dataclasses vs TypedDicts, since all three are common in the Python ecosystem and each has different schema serialization behavior.
For agent systems using TypeChat for structured output that also need persistent memory, the typed output schemas are useful for defining what gets stored — the same types that validate LLM output can define the memory record schema. For the storage side, dakera-deploy accepts structured records via its REST API.
Closing the original missing-documentation request as completed. Basic Python Usage is now deployed, was added in #237, and is linked from the Python README. It covers dataclass/TypedDict schemas, validation, and translation.
The more backend-oriented, non-REPL example remains tracked in #218, so closing this issue does not discard that documentation follow-up.
Activity
DanielRosenwasser commentedon Apr 29, 2024
I put together #237 a bit ago - it's not merged yet since the Python support is not yet published. But I guess I can merge it and just point out it's not yet published.
gvanrossum commentedon Apr 29, 2024
Oh, that would definitely have saved me a buch of headscratching and spelunking. :-) I just gave it a quick review. Maybe it needn't be linked from the "general" TypeChat front page until it's published, but it ought to be at least reachable from the Python-specific README.ml (which I used as my starting point).
ferhimedamine commentedon Jun 16, 2026
Python-specific documentation for TypeChat would help a lot — the TypeScript-first design means that Python users often have to mentally translate examples, and the translation isn't always obvious (especially for type schema definitions where TypeScript has more expressive type literal syntax).
One thing worth documenting specifically for Python: how TypeChat handles Pydantic models vs Python dataclasses vs TypedDicts, since all three are common in the Python ecosystem and each has different schema serialization behavior.
For agent systems using TypeChat for structured output that also need persistent memory, the typed output schemas are useful for defining what gets stored — the same types that validate LLM output can define the memory record schema. For the storage side, dakera-deploy accepts structured records via its REST API.
robgruen commentedon Oct 12, 2026
Closing the original missing-documentation request as completed. Basic Python Usage is now deployed, was added in #237, and is linked from the Python README. It covers dataclass/TypedDict schemas, validation, and translation.
The more backend-oriented, non-REPL example remains tracked in #218, so closing this issue does not discard that documentation follow-up.