We start it for you. The incident already has a timeline, the right humans, the recovery actions, the metrics it tripped on. Postmortems open with all of that already filled in — you write the parts only you can write.
/var on keywest reached 94 % at 02:14, postgres entered fsync stalls, p95 climbed to 720 ms on api-prod, 5xx propagated across two probe regions. Backup-age alert at 02:30 was a downstream effect of the same root cause.
Everything that came from the incident timeline lands pre-populated. You add the human-only parts — context, lessons, owners — and ship.
Auto-drafted from the incident's alert chain and AI Triage notes. Editable.
Imported from the incident — alerts, recipe runs, recoveries, manual notes. Times kept exact.
Traffic, error-rate and SLO numbers pulled from the affected window. Customer-facing? You decide.
Starts with the AI Triage suggestion, refined by the writer. No vibes — every claim cites a log line or a metric.
Lifted from the recovery actions in the timeline (recipe runs, restarts, mitigations) — you confirm.
The detection gaps the timeline exposes — alerts that should have fired earlier, paths that should have rate-limited.
Owner + due date + status. Tracked across postmortems so the same item doesn't reappear forever.
The only purely-human field. Empty until you write it. We don't pretend to write this for you.
From timeline to publishable in under thirty minutes — even at 03:42.