Skip to main content
Ankra AI investigates problems across your clusters, drafts fixes and, where you allow it, applies them. By default nothing that is already running changes until a person approves it: every change Ankra AI proposes in a conversation stops for your confirmation, and fixing alerts on its own is off until someone who manages Ankra AI turns it on. It runs on Ankra’s default models within a monthly allowance, or on your own model key. See Models and cost.

What Ankra AI does for you

  • Investigate a problem - press ⌘J (Ctrl+J on Windows and Linux) on any page to open Ask Ankra, and ask why. On a cluster’s pages the question carries that cluster. See AI Assistant.
  • Build or change a Stack - describe what you need and it proposes the add-ons, their order and starting values. See Building a Stack.
  • Explain a firing alert - an alert that fires gets an AI analysis with the likely cause and what to do. See Alert analysis.
  • Find problems before they page you - a health check of your clusters files each problem with a likely cause and the steps to fix it. See AI Insights.
  • Review pull requests - a review on every pull request in the repositories you choose, and answers when someone mentions @ankraai. See AI code review.
  • Work tickets - the Ankra agent investigates a ticket, prepares a plan, carries out the work a person approves and records the result. See AI Board.

Where everything lives

Ankra AI is a row in the sidebar’s Platform group (/next/ai). It opens its own menu. You see it when your role can read agent runs, alerts or the organisation. The menu’s Configuration section holds the organisation-wide settings. Your role needs to read the organisation to see it. Two places outside the menu belong to Ankra AI too:
  • Your account → Preferences (/next/account/preferences) holds your own settings: the mode a new conversation starts in, and What Ankra may do without asking you. See AI Assistant.
  • Needs you (/next/needs-you) lists every approval, fix and alert that waits on a person, once. See Activity & Inbox.

The Overview page

Ankra AI → Overview opens with four counts. Each one opens the list behind it: Below the counts:
  • Needs your team lists the decisions waiting on your team. Open on the board opens them on Work.
  • In progress lists up to six tickets with their number, cluster and next step. View board opens the rest.
  • Recent agent results shows the three latest finished runs. All runs opens Activity. The results cover the latest runs only, and a run that completed has not necessarily resolved its ticket.
  • Ankra agent says where ticket conversations are posted. Ticket desk settings changes it.
If tickets are switched off for your organisation, the Overview reads Tickets are not turned on and the menu has no Work row. Contact support to have them turned on.

What it can change without asking

Five controls bound what Ankra AI may do, each with a safe default. Whatever these settings allow, these always wait for a person:
  • Cluster lifecycle - creating, scaling, upgrading, stopping, moving or deleting a cluster, its node pools or its control plane.
  • Removals - a Stack change that would remove or replace a deployed add-on or manifest, or delete a Stack, and a Stack change whose validation failed.
  • Alert silences - muting an alert always needs a person.
Ankra AI never writes into protected namespaces (kube-*, ankra-*, and operator namespaces such as cert-manager, ingress-nginx, argocd and vault), in any mode, and secret values it detects in what it reads are redacted before they reach the model. The exception to all of the above is an auto-remediation rule set to Run on its own: that fix runs with no approval, so set it only for a fix you would approve every time.

Stopping Ankra AI

Ankra AI → Autonomy and approvals → Stop Ankra AI has two stops, from the gentler to the harder. Both apply to everyone in the organisation straight away, and need a role that manages Ankra AI.
  1. Pause autonomous actions - chat, finding causes and read-only investigation keep working, but Ankra AI cannot change anything until a person approves it. Pausing turns auto-remediation off and pauses every AI agent. Resume autonomous actions switches back on exactly what the pause switched off.
  2. Stop all AI - the emergency stop. It cancels every running AI conversation and agent run, expires approvals that are waiting, and blocks chat for everyone until someone resumes it with Resume all AI. It also keeps the analysis Ankra writes when an alert fires and the scheduled health check from calling a model: an alert that fires is still recorded, with a note that the AI analysis was skipped, and the health check waits until AI is resumed. While it is on, the page reads All Ankra AI is stopped.
  1. Open Ankra AI → Autonomy and approvals.
  2. Under Stop Ankra AI, press Pause autonomous actions… or Stop all AI….
  3. Enter a Reason and confirm.
See AI Autonomy.

Where you can use it

Every surface uses the same tools and the same permission checks, and answers with the permissions of the person asking. A channel or a mode never grants more than that person’s own access.

Models and cost

Ankra AI runs on Ankra’s default models, counted against a monthly allowance for your organisation, or on your own provider key, which bills your own account and never draws on the allowance. When the allowance runs out on Pay as you go, your credit pays while there is some. Otherwise a question is refused with a message saying so, or, in an organisation that has the DevOps AI model by Ankra, answered by that model within its own monthly allowance. The rest of Ankra keeps working. Models and cost has the models, the allowance on each plan and the spend cap.

How work flows

Work reaches the Work board from three directions: an alert fires, the health check finds a problem, or a person files a ticket (Create an Ankra AI ticket in Find, or /next/ai/tickets/new). The Ticket desk (Ankra AI → Settings → Configure work routing, /next/ai/tickets/desk) decides how this runs. Its Ankra agent tab shows the one shared agent that works tickets, with a status such as Available for automatic staffing or No eligible agent reported, and the Require an independent plan review checkbox: the review pass is separate from the work pass, and an agent cannot approve its own work. Its Communication tab chooses where new automatic tickets open their shared conversation in Slack or Teams. Organisation administrators can change the ticket desk; everyone else can read it.

Get started

  1. Press ⌘J on a cluster’s page and ask what needs attention. Ankra AI answers in the panel.
  2. To bring Ankra AI to your team’s chat and pull requests, an admin opens Ankra AI → Gateway and presses Set up the gateway. See Set up the AI Gateway.