A developer cannot ship while this is unavailable. Outages here trigger the documented fallback: engineers continue in Claude Code or against direct provider access until routing is restored.
The Codie harness and model router. The only component in a developer's critical path — if this is unavailable, coding is blocked and the documented fallback applies.
Reporting and background automation. Sessions are captured locally by the CLI and reconcile on reconnect, so unavailability here delays visibility without blocking work or losing data.
Dashboards, reporting, and the web portal. Unavailability here does not block developers from shipping.
The endpoint the CLI reports sessions to. Captured sessions persist locally and reconcile on reconnect, so an outage here delays reporting without losing data.
Dispatch for background agent tasks and pull-request automation.
Each component exposes an unauthenticated health endpoint, polled every 5 min — about 288 samples per component per day. Nothing on this page is set by hand.
Fully-healthy probes divided by probes recorded, summed across raw samples rather than averaged across days. A reachable-but-degraded response is excluded from uptime, which biases the figure down rather than up.
A day that recorded under 80% of its expected probes is marked incomplete and hatched, never averaged into a clean number. Days before monitoring began are shown as unmonitored.
Platform and capture probes run inside the same infrastructure they measure, so a total outage of that infrastructure stops the probes rather than recording a failure. Those windows surface as incomplete days. Moving these probes to an independent third-party monitor is planned.
No failed probes recorded since Aug 5, 2026.
Duration is inferred from consecutive failed probes, so it is accurate to within one 5-minute interval. For incident detail or an availability report covering a specific contract period, get in touch.