Token usage and genai calls not seen in ai foundary tracing for AI Agents

[sapinderpalsingh]

I configured azure monitor open telemetry to an ai agent application and configured AZURE_TRACING_GEN_AI_CONTENT_RECORDING_ENABLED=true in my environment variables but i still can't see the LLM calls in AI foundary tracing. While the same is much cleaner when I integrate the application insights with semantic kernel chatcompletion agent and setting SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS_SENSITIVE=true

Sample output with SK chatcompletion agent:

Sample output with AI agent

Keeping or removing AZURE_TRACING_GEN_AI_CONTENT_RECORDING_ENABLED=true in .env file doesn't change the output.

[github-actions]

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @hectorhdzg @jeremydvoss @lzchen.

[github-actions]

Hello @sapinderpalsingh. I'm an AI assistant for the azure-sdk-for-python repository. I have some suggestions that you can try out while the team gets back to you.

• Verify that the environment variable AZURE_TRACING_GEN_AI_CONTENT_RECORDING_ENABLED=true is correctly set and available in the running context (similar to what is done for the SK chatcompletion agent with SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS_SENSITIVE=true).

• Confirm that your application initializes and enables the Azure Monitor OpenTelemetry instrumentation before any AI agent calls are made. For example, using configure_azure_monitor(connection_string=<your_connection_string>) as shown in this guide.

• Enable additional diagnostics (e.g. printing traces to stdout) to verify that spans are being created. Check out the snippet that uses project_client.telemetry.enable(destination=sys.stdout) in the documentation for reference.

• Compare the configuration with the semantic kernel setup to identify any differences in how your agents report token usage or LLM calls.

• If the issue persists, consider verifying that you are using the latest versions of the Azure AI Projects SDK and its dependencies, as experimental tracing features could have updates or bug fixes in newer releases.

The team will get back to you shortly, hopefully this helps in the meantime.

[M-Hietala]

The message contents are part of event attributes. There is work being done to make the event contents visible on the trace UI.