Skip to main content

Monitoring

The Monitoring group answers "what are the agents doing, what is it costing, and can we prove it". All five pages require org_admin or above; the group is hidden for lower roles.

Token Usage​

/monitoring ("Token Usage Monitoring") tracks LLM token consumption.

  • Period selector: last 24 hours, last 7 days, last 30 days.
  • Group-by selector: by model, by agent, by user, by purpose, or by tool — the usage table re-aggregates instantly, with a Refresh button for live updates.
  • A User Quotas table lists per-user daily and monthly token limits with current usage. Limits are edited inline; quota rows are created automatically the first time a user interacts with an agent.

Observability​

/observability is the operational health dashboard, organised into tabs:

  • Overview — error rate (7 days), session volume (14 days), top agents, and skills by status.
  • Heatmap — a 30-day tool invocation heatmap showing which tools run when.
  • Anomaly — anomaly detection rules, with an Add Anomaly Rule dialog for alerting on unusual behaviour.
  • Performance — agent performance analytics across sessions.
  • OTel — OpenTelemetry status and the list of configured exporters, for shipping traces and metrics to your own observability stack.
  • Cost — cost optimisation recommendations derived from usage patterns.

Sandbox Executions​

/observability/executions records every tool call across the org. Click a row to see stdout, stderr and metadata for sandboxed runs.

  • Filters: tool, status, agent, from/to datetime, and a Sandbox only toggle that limits results to tool calls that ran inside a container.
  • Refresh re-queries; Reset clears all filters.
  • Failed API states are surfaced explicitly, so an empty list always means "zero rows", never a silent error.

Audit Log​

/audit is the append-only administrative audit trail.

  • Filters: resource type (users, agents, gateways, skills, teams, sessions) and after/before date bounds.
  • The table shows time, actor, action, resource type, resource ID and IP address, paginated in pages of 20.
  • Entries with no actor are attributed to system.

Compliance​

/compliance groups the regulatory tooling into four tabs:

  • Retention — retention enforcement status plus an Add Retention Policy dialog for defining how long each class of data is kept.
  • Erasure — GDPR erasure requests: raise a request for a subject and track it through processing.
  • Reports — compliance report generation, including an access review report and a 30-day data access report.
  • Certifications — the list of compliance certifications configured for the deployment.

Deeper operational guidance for these surfaces lives in Observability and audit and Security and compliance.