Support for Async in Google GenAI SDK - LLM Observability

[maltehedderich]

Hi Posthog Team,

recently you added basic Google GenAI SDK support see #201. Thanks a lot for that!

Unfortunately the whole async part of the client seems still unsupported, as the aio part is missing. Would be awesome to have that also in the Posthog Client replacement.

Cheers,

Malte

Edit: I just noticed the files API is also missing.

[kdenz]

Have the same need too

[LucasAlvesSoares]

Needed that too! #324 .Not sure how long it takes to be reviewed but feel free to use mine for now

[clr182]

+1 https://posthoghelp.zendesk.com/agent/tickets/40880

[staru09]

Is this been taken care of?

[ethanporcaro]

This would be extremely helpful with the pydantic-ai framework.

[ethanporcaro]

I found a workaround way to do it. You may need to tweak the methods you carry over. This is unstable so use at your own risk.

from posthog.ai.gemini import AsyncClient
from google.genai.client import Client

google_client = Client(
    api_key=os.getenv("GEMINI_API_KEY", ""),
)
ph_client = AsyncClient(
    api_key=os.getenv("GEMINI_API_KEY", ""),
    posthog_client=posthog_client
)
google_client.aio.models.generate_content = ph_client.models.generate_content
google_client.aio.models.generate_content_stream = ph_client.models.generate_content_stream
client = google_client

It basically uses the original Gemini client, then replaces the actual generation methods with the posthog tracked ones.

[marandaneto]

Gemini async model calls exist, but not full drop-in compatibility (Client().aio, files API missing).