Setup for Cursor to understand Pydantic AI

[zdzarsky]

Question

Question to the community. Does anyone successfuly managed a cursor setup in which it does not mix some paradigms of langchain with PydanticAI?

I have tried adding the Pydantic AI docs but it still hallucinates as hell (i.e. pipes agent calls on a default model)

I truly love the simplicity of the tool, it would be awesome to have a setup in Cursor.

Additional Context

Cursor: 0.49.x
Pydantic AI: 0.1.9

[Kludex]

Can't you index the docs here?

[zdzarsky]

Yup, this is something I use from day one, but hallucination is a real issue for me. I am building versions of agents and benchmark them on the same task. I often prompt something like:

Build an agent like @my_agent_v10.py using @PydanticAI
Do X instead of Y
Do Z instead of A
...

Suddenly new agent becomes a class that is very similar to langchain use, but with some incorrect mixins from PydanticAI. I have a hypothesis, that due to the fact that AI knows very well what Pydantic is, there might be an issue with a name Pydantic AI, because it can be interpreted as "use AI (which means some framework for calling llms), control input and output with pydantic".

I feel it requires some additional set of rules or using different model than default. By asking this question I wondered if anyone in the community has similar issues.