Microsoft Foundry provides built-in application/agent tracing based on OpenTelemetry that captures every step of a request as correlated spans — including both the model invocation via the Responses API and any tool calls made through Foundry connections, such as Azure AI Content Safety. When application tracing is enabled in the project (typically by connecting an Application Insights resource), Foundry automatically instruments and correlates these steps into a single end-to-end trace that can be viewed in the Foundry portal's Traces view, giving visibility into inputs, outputs, latency, and errors at each stage of the workflow. Enabling logging only within the Content Safety client SDK would isolate that single call and not correlate it with the model request, Foundry Local applies to local model execution rather than this cloud-hosted workflow, and simply routing requests through the Azure OpenAI endpoint does not by itself produce any tracing or visibility into the workflow steps.
Community Discussion