Deepankar M.

ah nice! This seems to work. Thank you 🤖

I see, so this gets overwritten since SpanAttribute.METADATA is used as the key. Makes sense. RunLLM (or any humans 🙇) are there examples of how to query in phoenix using custom attributes? Because of the overwriting issue, I cannot use metadata for filtering and I can't find any example in phoenix docs to filter using custom attributes.

RunLLM attributes are not overwritten with subsequent calls right? Only metadata is overwritten?

💯 surprised that there are not many 👍s for this

I did a dummy version here - https://github.com/Arize-ai/openinference/pull/1380. will be using this class locally, until we have a better version added in the repo. Thank you 🙌

Created an issue to discuss this - https://github.com/Arize-ai/openinference/issues/1379

Hey Mikyo, do we have a NoOpOITracer compatible with the OITracer interface? I can use the NoOpTracer from opentelemetry.trace, but that and OpenInferenceSpan have some extra methods that are not compatible. While suppress_tracing is useful in selected contexts, when phoenix is not available (e.g.- unit tests), NoOpOITracer enables us to have a global tracer and leave the rest of the code as-is.

I'll create an issue tomorrow. Thank you 🙏

We are building a tool calling agent that is multi-step (multiple tool calls one after the other) and multi-turn (human might add more messages). We have many tools that are part of the conversation (one session) with each call to the LLM (hopefully) triggering a tool call + tool result sequence. So tools are as important as the messages from human/LLM.