I know we are also discussing this in discord

This didn't fix my problem. I think the issue is that Agno (rightfully so) is doing a lot of work to manage the event loop for the motor client and the result of my having a singleton instance is likely causing problems for agno. My guess is that maybe AgentOS does something at the begnning now and the event loop becomes stale on the singleton instance? (I initialize AgentOS in my lifecycle method for fastapi)

regardless, I fixed the issue by simply allowing Agno to manage the client.

However - as noted in discord - if Agno plans on moving to pymongo async client. It will require my own client and my hope is that it will work fine in that scenario

This fix will be included in the upcoming release, I believe? I'm not seeing it in the existing releases.