Skip to main content

Monitoring

Monitoring provides runtime and execution observability for agents in production.

Monitoring Signals​

  • Agent lifecycle and health status.
  • Tool execution outcomes and latency.
  • Cost and token usage trends.
  • MCP/toolset readiness indicators.

Practical Workflow​

  1. Launch agent with monitoring enabled.
  2. Inspect status and telemetry during execution.
  3. Correlate failures with outputs, tool calls, and approvals.
  4. Feed findings back into specs and guardrails.

Best Practices​

  • Monitor every production agent by default.
  • Keep alert thresholds aligned with workload criticality.
  • Tie monitoring dashboards to incident and on-call processes.

See also​