All systems operational

Every 5 minutes an automated check asks each service above a direct, cheap question: does the app answer on every copy we run, does sign-in load, does the call relay answer a connection request, is the meeting bot running, do our AI and transcription providers answer, is email leaving and not queueing up, is the push provider reachable, are connected calendars and mailboxes syncing, and does the API refuse a call without a key as it should.

Each bar is one UTC day and shows the worst answer that day, so a failure that lasted five minutes still paints the day: hover a bar to see how long it lasted. A pale bar means no checks were recorded that day, which is a different fact from a service being fine, and we will not draw it the same way.

We count readings against the schedule instead of publishing an uptime percentage. A check that did not run records nothing, so a percentage computed from recorded readings would get BETTER during an outage that also stopped the checks. Counting passed checks out of the checks the schedule called for, and saying how many were missed, is the honest version of the same information.

Incidents and maintenance on this page are written by a person, not generated. Our checks cannot see everything, so a green row and an open incident are not a contradiction; while an incident names a service, that service shows at least Degraded.