Observability

• Readiness. GET /ready on the gateway and admin checks every dependency and returns each check in its body; use it for readiness probes and alerts. GET /health is liveness only.
• Logs. Each service logs to standard output. Credentials and secret-like fields are redacted before logging; upstream error details stay in the log rather than being returned to agents.
• Traces and metrics. Set observability.otlp_endpoint to an OTLP HTTP collector to export traces (one span per HTTP request) and metrics. Nothing is exported when it is unset. Needs the otel extra, which the published image includes.
• Gateway statistics. GET /stats reports cache hit counts, active connections and registered sources.
• The audit log is the record of what agents did; the usage view summarises it. See Audit and usage.
• Worker health. The Workers page shows each worker’s heartbeat, and Ingestion shows job state.

Service-level objectives and recovery targets are in the availability runbook.