Anthony P.

In your experience do you think it would be better to just use a global trace feeding both service (phoenix and app otel service)?

Yeah I think this is the strategy I would take if I wanted to leverage the automatic tracing of strands, in lieu of openinference specific tracing support of Strands

Another approach that you may try is to configure multiple span processors off of the same strands telemetry, and then modify the spans in the processor that goes to phoenix into openinference (to the level of fidelity that you would like). This approach is nice in some ways because you will get full spans without needing to do manual tracing (as it would be handled by strands telemetry), but more cumbersome in a way in that you will need to write custom code in the phoenix processor that looks at each span and figures out which attributes to add or filter I suspect a custom implementation would look very close to this https://arize.com/docs/phoenix/tracing/how-to-tracing/advanced/modifying-spans with attributes matching these patterns https://github.com/Arize-ai/openinference/blob/main/spec/llm_spans.md

You can enable authentication against your Phoenix instance via environment variables https://arize.com/docs/phoenix/self-hosting/features/authentication#authentication But by default, you are correct, you do not need an API key

thank you

for what it is worth, getting to the web UI does imply that you have successfully self hosted Phoenix. But I agree, you should then be able to easily get up and running with an application that is traced properly into Phoenix. I will file some documentation triage and get it to the team this week

I can give the cookbook a spin today and try to replicate any issues

Phoenix is a simple http server at its core

that sounds possibly like a network configuration issue with docker

I would expect this quickstart to work for your use case, I am curious if you still have a specific error message available that you encountered

I can try to find you an appropriate tracing quickstart if you have an LLM / Programming Language preference. Phoenix at its core operates best when you have an application that can be instrumented and traced into Phoenix