Rémi C.

Hey is there possibility to display input messages for other kinds of span than the LLM kind I tried

class ChatMessage(TypedDict):
    role: str
    content: str

def trace_messages(messages: list[ChatMessage], span: Span):
    for i, msg in enumerate(messages):
        for field in ['role', 'content']:
            span.set_attribute(f"input.messages.{i}.{field}", msg[field])

as per https://arize.com/docs/ax/learn/tracing-concepts/openinference-semantic-conventions#transforming-messages-into-opentelemetry-span-attributes I know this is for AX but it's weird input.messages is not a field available in the https://github.com/Arize-ai/openinference/blob/main/spec/semantic_conventions.md maybe it's an issue in the AX doc page and this doesn't exists anywhere. Anyway it would be nice to have input messages for other thing than LLM (agents) for example Currently I'm using an LLM span kind for what is actually an agent (or chain) that processes an history but none of agent nor chain displays inputs.messages and llm.input_messages

Hello I have a question 🙂 (we are using phoenix v3.23) Whenever there is an LLM call, I see the following POST API call to the `/traces` endpoint, which is expected: ``` INFO: 127.0.0.1:37436 - "POST /v1/traces HTTP/1.1" 200 OK ``` However, I also observe a significant number of API calls to the GraphQL endpoint. Could you explain why so many calls are happening? Is it making frequent calls to send out trace logs for telemetry or another purpose? ``` INFO: 78.198.212.116:53183 - "POST /graphql HTTP/1.1" 200 OK ```

Hey do you have any idea why the persistance is not working with the latest version when I do python -m phoenix.server.main serve and I turn it off the traces disapear