fix(agno): handle streaming mode in InvokeAsync to prevent pickle error

[KarthikeyaKollu]

Fixes #214 - Error: cannot pickle '_thread.RLock' object when using OpenAIChat with Memori + Agno in streaming mode.

The issue occurred because InvokeAsync was trying to call handle_post_response() on the raw stream object, which contains unpicklable _thread.RLock objects.

The fix detects when stream=True is in kwargs and returns a wrapped async generator that:

1. Iterates through the stream and yields each chunk
2. Collects chunks into raw_response using merge_chunk
3. Only calls handle_post_response after the stream is fully consumed

This matches the behavior of InvokeAsyncStream but allows InvokeAsync to handle both streaming and non-streaming scenarios dynamically.

[devwdave]

@KarthikeyaKollu Thanks for the contribution! The team will review this PR later in the week and provide feedback if necessary.

[devwdave]

@KarthikeyaKollu Can you confirm that this issue persists if you state that your using streaming in the memori initialization?

mem = Memori(conn=conn_factory).llm.register(my_llm_client, stream=True)

[KarthikeyaKollu]

Hi @devwdave I've tested this thoroughly, and here are the results:

1. When I tried using mem = Memori(conn=conn_factory).llm.register(my_llm_client, stream=True) on the main branch, it failed with a TypeError, as the register method on main doesn't accept a stream argument.

2. I confirmed that the original issue still persists on the main branch when streaming is enabled correctly in the agent.arun(..., stream=True) call. It produces the cannot pickle '_thread.RLock' object error.

3. As this branch (fix/agno-openai-streaming-pickle-error-214) completely resolves the issue. The streaming works as expected without any pickle errors.

i feel things are working in the fix branch, let me know if you have anything!

[KarthikeyaKollu]

@devwdave

[devwdave]

@KarthikeyaKollu Thanks again for this contribution. We’ve been working through a few related fixes and believe this may have been addressed in a more recent release.

I noticed you were on v3.0.6 when the issue was opened, would you mind upgrading to the latest (v3.1.2) and letting us know if you can still reproduce the problem?