Runbooks on keys
128 across 22 servicesRunbook automation · AI operator
One buttonper runbook.
Wire an incident procedure once. DEADMAN turns it into a key anyone on call can press. The model does the steps, the log shows its work, and the rollback is the same key pressed twice.
Press the red key. Nothing real is touched.
Red key · restart the payments worker
Ready
run log · svc/payments@staging
idle: press the key
$deadman run rb_payments_restart
Median time to press
9s from page to first stepLongest manual blocker
41min before the team moved to keysSteps run out of scope
0 every step is bound to a scope you setExample figures · median across 40 teams · Q3 2026
Readiness
Runbook readiness audit
Audit not started. Half your procedures are still a doc someone has to read at 3am.
0%Registry
Every procedure is a key
Works with your stack
- Kubernetes
- PagerDuty
- Datadog
- Terraform
- GitHub
- Slack
- SOC 2 Type II
- Append-only audit log
- Scoped, short-lived tokens
- SSO and SCIM
We had a forty-page wiki for payments incidents. Now it's six keys, and our newest engineer pressed the right one on her first night on call.
Playbook
Three moves, in order
Scope
A key that cannot touch prod is a key anyone can press.
Bind every runbook to a service, an environment and a window before it reaches the board.
Automate
Replace the doc nobody reads at 3am with one key.
The model drafts the step list from your procedure and waits for an approver who is already awake.
Escalate
Do not let a blocker age into an incident review.
A halted run pages the next name on the rota with the failing step attached, not a summary of it.