[bug] Autogen Agentchat Tracing - too many spans produced

[s-yeddula]

Describe the bug
Some users have brought up that the Autogen Agentchat tracing is unnecessarily verbose. Too many spans are created.

Here is an example of a user issue: https://arizeai.slack.com/archives/C04R3GXC8HK/p1751498388636159

To Reproduce
https://arize.com/docs/phoenix/integrations/frameworks/autogen/autogen-agentchat-tracing
Followed instructions from code snippets on docs

Expected behavior
Some spans don't carry much information or could be combined for a clearer trace view. Not sure how specifics work here, just wanted to raise this issue because I've see multiple people bring it up.

Screenshots
Example: https://storage.googleapis.com/arize-phoenix-assets/assets/images/agentchat-phoenix.png

[dosubot]

The Autogen Agentchat tracing integration in OpenInference creates a span for every call to key methods (agent runs, tool runs, group chat streams, sending/publishing messages) by wrapping these methods at runtime, which leads to a large number of spans for complex workflows. There isn't currently a built-in configuration or logic to combine, filter, or reduce the number of spans created—each wrapped method call results in a separate span, and the TraceConfig only controls masking or redacting span attributes, not span count or structure [source].

You can use the suppress_tracing context manager to manually pause tracing and suppress span creation within specific code blocks, but this requires explicit intervention and doesn't provide fine-grained control over verbosity at the instrumentation level [source].

Recent updates have added more detailed attributes to tool call spans and improved root span discovery for agent traces, which may help with trace clarity but do not directly reduce the number of spans produced [source][source].

If you need to reduce verbosity, the only current workaround is to selectively suppress tracing in parts of your workflow. Otherwise, this would require upstream changes to the instrumentation logic to support span grouping or verbosity controls.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[caroger]

Thanks for filing the issue @s-yeddula.

Do you have more examples of users reporting/complains around the verbosity of the spans? And which version of the autogen-agentchat auto instrumentor were the users on?

The link to the community slack thread and to the docs seems to be on an version that is prior to #1790 which aimed to address the verbosity complaint that we got from the initial release.