We would like to automatically instrument llm calls made with llm (without going through the ag.span). This would allow us to instrument all the informations in the call.
This would allow us also to provide a simple code snippets for people wanting to fetch the prompt and instrument it in their code base (not using the hosted app).

For this we need to add an integration with litellm.

First task is to see what options are there to integrate with litellm.

_{From SyncLinear.com | AGE-159}