Recommended way to view full requests to model (without a Logfire account)?

[g-eoj]

Question

I often find it is best to see the full request sent to the model when trying to understand LLM responses. Examining message parts doesn't work to well for me. So if I want to see exactly what is sent to the model, is there a way to do it?

Thanks in advance!

Additional Context

No response

[DouweM]

@g-eoj The requests are built deep inside the model SDKs, so there's no way to get to see them except by instrumenting httpx like so: https://ai.pydantic.dev/logfire/#monitoring-http-requests.

You don't have to create a Logfire account though, you can use logfire.configure(send_to_logfire=False) to not send to Logfire, and you can use the additional_span_processors option to add a custom SpanProcessor that can do whatever you like, including writing the span attributes (that will have the request and response body) to a file or stdout.