UPKira monitors websites, APIs, certificates, and jobs that must check in. An incident opens after a confirming observation, not after one unlucky timeout.
GET or HEAD, expected status codes, a timeout, a redirect cap, and an optional response-text check. The text check is a plain substring, not a regular expression. An optional slow threshold marks the monitor degraded and does not page you.
A certificate check warns at the days you choose, commonly 14, 7, and 1. The warning is a TLS event, separate from an outage.
A cron job or worker pings a token URL. Only a hash of the token is stored. Miss the interval plus the grace period and the monitor is due for an expiry check.
The monitor moves from up to suspect. A second probe, from another healthy region when one exists, confirms the failure and opens the incident. While the monitor is recovering, it needs consecutive successes before it is up again. A missed check never counts as up.
Email, Slack, Discord, and a generic webhook. Webhook bodies are signed. Quiet hours defer delivery instead of dropping it. Failed deliveries retry with backoff, then stop after five attempts so a dead endpoint cannot loop forever.
A public status page at your slug shows the monitors you attach to it. Maintenance windows park probes and keep that time out of the uptime percentage. Unknown, paused, and maintenance time are excluded from the denominator, so a new monitor is not reported as 100% up.