Invitation for feedback

[gvanrossum]

Now that typeagent is released (as 0.3.0), I'm hoping to get some feedback from people who are interested in using it. If you think you have a potential use case, is anything blocking that use case? More docs? Features? Project management? A roadmap?

I want to point everyone to TODO.md, where I've collected a pretty random set of things I might want to work on. I'd love feedback on what priorities people like me to give to those. (I am intending to move from a single chaotic TODO list to issues in this repo with tags indicating priority, project size, category etc.)

Feedback can take any form: issues, PRs, comments in this thread, new discussion threads. Even personal email.

[bmerkle]

As i am a fan of DSLs and extensible languages which can also carry the domain semantic, i like the idea of structured RAG.
It has schema info and carries more semantic than just embedding the plain text.

Currently the approach uses schemas, have you considered using knowledge graph for the meta information ?
(e.g. https://github.com/microsoft/graphrag)

[gvanrossum]

@bmerkle -- Steve Lucco gave a great answer when this question was asked at PyBay. It's preserved in the recording: https://www.youtube.com/watch?v=-klESD7iB-s around 29:30.

[ahmetbulut]

Hi Guido,

Pleasure to meet you. What is the best way to integrate local LLMs (managed by a locally running LLM manager, e.g., ollama) into the typeagent ? As far as I understand, it is currently based on private & closed-source LLMs, e.g., OpenAI; and therefore it requires an API key to work.

Cheers, --Ahmet.

[gvanrossum]

Check how Pydantic AI handles that in its docs.