Description

I are currently not able to use the tracing from pydantic-ai in production because the spans contain prompts and completions, which is very sensitive data.

Adding a settings somewhere that disabled the tracing of sensitive data like prompts and completions would be great, and make us able to use the tracing from pydantic-ai out of the box.

A possible solution could be to add a parameter toInstrumentationSettings that controls whether sensitive data is included in the traces.

References

Other libraries, like OpenLLMetry, offer this functionality:https://www.traceloop.com/docs/openllmetry/privacy/traces