Escalation policies

/dashboard/alerts/escalation is the policy editor for paging the right person.

How escalation works

When a critical alert fires:

1. Alert lands in the inbox + goes to whatever channels the routing matrix configured.
2. Sikurd starts a countdown — 10 minutes by default.
3. If nobody acknowledges the alert within the window, the policy escalates: pushes a notification to the next person in the chain, fires a Slack mention, or initiates an AI voice call.
4. The escalation can have multiple levels — 10 min → 20 min → 30 min, each with a different recipient.

What a policy contains

• Name + description.
• Scope — which instances does this policy apply to? "All" or a picker.
• Triggers — which alert types + severities trigger it.
• Levels — ordered list of (wait minutes, recipient, channel).

Example policy

On-call rotation — week 1 Jane

• Scope: All instances
• Triggers: Critical alerts (Instance Down, Service Down)
• Level 1 (0 min): Push + Slack #ops-alerts
• Level 2 (10 min): AI voice call to Jane
• Level 3 (25 min): AI voice call to Mark (backup on-call)

Acknowledging stops the escalation

When anyone clicks Acknowledge on the alert (in dashboard, mobile, or by replying to the push), the escalation stops.

Resolving acknowledges

Resolving an alert also implicitly acknowledges it.

Quiet hours apply

Escalation respects the tenant-wide quiet hours from the routing matrix — Critical alerts still escalate; Warning and Info pause until the window ends.

Multiple policies

You can have several. They evaluate in priority order. The first matching policy wins.

Voice calls (Pro+)

AI voice calls use Dialora. The call agent reads "Alert: [type] on [instance name]" — terse so the recipient knows what's wrong but no internal details leak over the phone line. Configure the agent and the on-call phone numbers in Settings → Integrations → Dialora.