Observability

The platform ships a self-hosted observability stack (metrics, logs, traces) with per-organization isolation — one customer's telemetry is never visible to another. Administrators read dashboards; nothing is sent to a third-party cloud.

Who can see this

Org admins see their organization's dashboards; platform admins see all. Dashboards open from the console (or directly in Grafana if your platform exposes it).

What you can see

• Usage & cost — tokens and USD by organization, project, group, user, and model.
• Cache savings — semantic-cache hit rate and the spend it avoided.
• MCP activity — tool-call volume per project (when MCP is in use).
• Request health — throughput and error/blocked rates.

Per-organization isolation

Each organization's metrics, logs, and traces are kept in an isolated tenant, and dashboards are scoped to the viewer's organization. This keeps multi-tenant telemetry private by construction. Retention for metrics, logs, and traces is set by the platform operator — see Platform observability.

Next steps

• Budgets & limits — act on what the dashboards show.
• Audit log — who changed what.