Observability
See what your agents do in production: what they log, how long each step takes, what each model call cost, and whether the process is healthy.
- Monitor agents: structured logs, OpenTelemetry traces and metrics, a Prometheus endpoint, health checks and structured agent traces.
- LLM calls: capture every model call as an
LLMCallLogevent with prompt, response, usage, cost and timing, or trace it as a GenAI span. - Sampling: decide per event whether to capture, drop or down-sample, and cap volume per tenant.
To score what agents answer rather than how they run, see Evaluation.