Synchronous Python Client

[mattzh72]

We would like to have a synchronous Python client. Our code is primarily synchronous, so writing a ton of async -> sync wrapper code is tedious and bug prone.

Letta team is also happy to contribute this to the codebase.

[Nov1c444]

I'm also very interested in having this feature integrated into the official SDK repository, so is there any progress on this at the moment?

[JIAQIA]

Is there an official plan for this feature? We would also be happy to contribute to it. And we really need it.

[Kludex]

There's no official plan.

[felixweinberger]

Is this still an issue @mattzh72 @Nov1c444 @JIAQIA? Could you provide some additional information on why this would be valuable to you?

Avoiding a few lines of wrapper code at the application layer is likely not enough reason for a heavy investment here.

[JIAQIA]

I have some tasks running in Celery, but I need to be careful because Celery’s support for Async is not very good. I could use some wrapper code to encapsulate and use async, but introducing additional uncertainty into the entire system is not something we can fully control. For example, we have our own event loop management work, and these tasks are not a major focus of our current architecture. Moreover, I’m not entirely sure about Celery’s compatibility, so I’d prefer not to take this risk.

In addition, our system currently supports both synchronous and asynchronous native interfaces. In many cases, we can’t fully control which mode developers might use, and once they start mixing the two, managing and coordinating them becomes more difficult. Our internal libraries are also quite complex — for example, we use the Blinker module, and when combined with LLM tool triggers, this makes our call order and logic not 100% deterministic.

Also, since Python 3.14 already has an official version that removes the GIL, I wonder if it might make development easier based on that version. I personally don’t mind exposing synchronous primitives to users who are already on 3.14.

[Kludex]

I think this is a valid request, but it requires too much work (to the point that I think this is just never going to happen).