Service status
Checked live, every time this page loads. Each component below is probed with a real call, not a ping.
How this is measured
Every check makes a real request to each dependency — a query against the database, a lookup against the model provider, a check that email is deliverable and that billing answers. A component is only “operational” if that call succeeded. Results are sampled about once a minute and kept for 72 hours, which is why the history above stops there rather than showing a year of green.
The same endpoint that powers this page is public at /api/health. It returns 503 when the database is unreachable, so you can point your own monitor at it.
Something broken that this page says is fine? Tell us — a check that misses a real outage is a bug in the check.