google-agents-cli-observability

Cloud Trace works out of the box for ADK agents via OpenTelemetry. Agent Runtime exports traces automatically, while scaffolded Cloud Run and GKE projects configure exporters through setup_telemetry(). View traces in Cloud Console under Trace explorer.

Set LOGS_BUCKET_NAME and OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT environment variables. Deployed scaffolded projects get these from Terraform; locally, export them before running agents-cli playground. Logs go to GCS as NDJSON and to BigQuery via log sinks.

Cloud Trace provides always-on distributed tracing of execution flow, latency, and errors via OpenTelemetry spans. BigQuery Agent Analytics is an opt-in plugin that logs structured agent events to BigQuery for conversational analytics, dashboards, and LLM-as-judge evaluations.

Verify setup_telemetry() runs at application startup and the service account has the cloudtrace.agent role. For tool spans, check the trace explorer filters since tool executions appear under execute_tool spans.

Yes, ADK supports AgentOps, Arize AX, Phoenix, MLflow, Monocle, Weave, and Freeplay through OpenTelemetry or custom instrumentation. Each requires provider-specific environment variables such as API keys and endpoints.

Run agents-cli infra single-project before the first agents-cli deploy for agent_runtime targets. Terraform owns the entire Reasoning Engine resource, so applying it after an SDK-based deploy creates a state mismatch that Terraform cannot reconcile.