Skip to main content
The Autonomy section under AI → Settings → Autonomy is the single place where organisation admins decide how much Ankra’s AI does without a human in the loop. Everything on this page is organisation-wide.
Autonomy is opt-in. A new organisation starts with auto-remediation disabled and no active agents - nothing runs autonomously until an admin turns it on here.

​
Stopping the AI

The Stopping the AI card at the top of the page has two stops, from least to most disruptive. Both apply to everyone in the organisation and take effect immediately.
  1. Pause autonomous actions - chat, diagnosis and read-only investigation keep working, but the AI cannot change anything without a person approving it first. It switches off the auto-remediation policy and pauses every running agent. Resuming restores exactly what the pause switched off - it never enables anything that was off before the pause. The pause is stored on the server, so any admin can resume it from anywhere.
  2. Stop all AI - the emergency stop. Nothing keeps working: it cancels every running AI session and agent run, expires pending approvals, and blocks chat for everyone until an administrator resumes it. Use it during an incident, not to tighten a policy.
Both are also in the CLI as ankra ai autonomy pause and ankra ai autonomy stop-all, with resume and start-all to undo them - see ankra ai.

​
Auto-Remediation Policy

The policy is off on a new organisation: firing alerts are not handed to an agent until an admin switches it on. When enabled, the remediation agent can act on detected incidents, within these limits:
  • Autonomy level - Observe only (diagnose and report, never touch the cluster), Propose (every corrective action goes to Slack for approval; the default), or Autonomous (low-risk reversible actions run on their own; the rest ask first).
  • Approvals - approval requests go to a Slack webhook; you can name the users allowed to approve.
  • Limits - a cap on actions per incident (5 by default) and a cooldown before the same alert on the same cluster is picked up again (30 minutes by default).
  • Per-action rules - override individual tools (for example delete_pod or restart_deployment) to Run automatically, Ask on Slack, or Never run. Run automatically means that action runs with no approval card at all.
  • Cluster scope - apply the policy to all clusters or an explicit allow-list.
Enabling the policy asks for confirmation and shows what the chosen level authorises before saving.

​
Ticket Desk

The ticket desk auto-files board tickets from proactive insights. Pick which severities create tickets and where ticket notifications go (a Slack channel or Teams conversation). See the AI Board for how filed tickets are worked. A drafted plan waits for a verdict from a reviewer agent. Bind the reviewer here, or turn the review requirement off so plans go straight to a person - otherwise plans park in review with nothing able to clear them.

​
Board Staffing

Board tickets are only worked by an agent you designate here. Until one is, every ticket the AI files is escalated straight to a person. Platform-created agents need the board identity before they can be designated; both live on the same card.

​
Proactive Insight Thresholds

Tune when the platform flags cluster health issues - the restart-count threshold and how recent restarts must be to count. Leave a field empty to use the platform default. The Analyse cluster health issues automatically toggle in the same section turns the analysis on or off entirely; see AI Insights.

​
Permissions

Everything on this page requires organisation admin. Members can view the current policy but not change it.

​
Next

See how these controls fit with chat modes, connection modes and agent autonomy in What it can change without asking.