Know the API path before you ship
Use the live service-status view for current incidents, then validate your own endpoint, model access, timeout, and retry behavior. A status page complements application monitoring; it does not replace it.
Live checks
Open the live service-status dashboard and make a non-sensitive request to the endpoint your application uses. Record the response status, request ID, model, and latency.
Build your own monitor
- Use a low-cost, bounded health prompt or a models request.
- Track success rate, p50/p95 latency, timeouts, and provider errors by model.
- Alert on your own error budget and pause retries during an incident.
After an incident
Retry only when your client policy allows it, inspect idempotency and duplicate billing implications, and rerun a representative smoke test before restoring full traffic.
Frequently asked questions
Where is the canonical status page?
Use /service-status/ from the main site for the live dashboard. This page is a crawlable explanation and quick link.
Does a green status page guarantee my request works?
No. Account permissions, model availability, client configuration, and regional network conditions can still affect a request.
Should I retry every 5xx?
Use bounded exponential backoff, respect request semantics, and avoid retry storms. Follow the API error and timeout guidance.