Skip to main content
Most cluster problems start small - a pod restarting, a node filling up - and are noticed only when they page someone. AI Insights scans your clusters on a schedule, spots those problems early, and gives each one a root cause and a fix. When an alert does fire, the same kind of analysis is attached to the firing, so the first person to look starts with an explanation instead of a blank log view.
Proactive AI Insights is switched on from AI → Settings → Autonomy. Until it is on, a cluster’s AI Insights page reads AI Insights generation is disabled and no new issues are detected. The alert-triggered analysis below works without it.
Each insight carries:
  • Root cause analysis - what went wrong and why.
  • Remediation commands - kubectl commands you can copy.
  • Platform actions - fixes through Ankra, such as a Stack change or an add-on configuration change.
  • Conversation starters - prompts that open the AI Assistant with the insight loaded.

​
Turn it on

1

Open the Autonomy settings

Go to AI → Settings → Autonomy. Changing it needs organisation admin.
2

Enable proactive insights

In the Proactive insights section, turn on Analyse cluster health issues automatically. It is off by default. The same section holds the detection thresholds and notification routing.
Each cluster also has an Enable AI Insights switch in its Settings → General, on by default; turn it off to leave one cluster out of the scans. With proactive insights off, you can still ask the AI Assistant about a cluster at any time with ⌘J.

​
Scan frequency

Scans adapt to cluster health:

​
The insights list

Open a cluster and click AI Insights in the sidebar. The page has two tabs. Insights lists the cluster’s insights. Filter by status (All, Open, Acknowledged, Resolved) and severity (critical, warning, info), sort by Most Recent, By Severity, Oldest First or Most Recurring, and search titles and root causes. Select several insights to Acknowledge or Dismiss them together. Analytics summarises the last 7, 30 or 90 days: new and resolved issues, the net change, mean time to resolve, a severity trend, breakdowns by namespace and category, the top recurring issues, and whether cluster health is improving, degrading or stable.

​
An insight’s detail page

Click an insight to open it. The page has four tabs:
  • Overview - severity, the AI’s confidence in its diagnosis (a warning appears when it is low), the root cause analysis, affected resources with links to them, recent changes that may have triggered it, and conversation starters.
  • Remediation - remediation commands, recommended actions, and platform actions you can apply from the page.
  • Health - the cluster’s health, pod and node state, and container logs captured when the insight was detected.
  • History - earlier resolutions if the issue has come back, each with its resolution type, notes, and whether the fix held.
The page also lists Related Issues: other open insights in the same namespace or category, since several issues in one namespace often share a cause.

​
Status and resolution

Mark as Resolved asks for a resolution type (Manual fix, Configuration change, Rollback, Scale up, Restart) and optional notes. Ankra also keeps a health snapshot from the moment of resolution, so you can compare the cluster before and after the fix. Under Was this analysis helpful?, rate the analysis. A helpful rating makes Ankra reuse the analysis as a reference for similar issues later.

​
Keyboard shortcuts

On the insights list, 1 and 2 switch between the Insights and Analytics tabs and / focuses search. On a detail page, a acknowledges, i asks the AI, o, r, e and h open the Overview, Remediation, Health and History tabs, and [ goes back to the list. Shortcuts are off while the command palette, a dialog, or an input field has focus. Scripts and agents read insights with the get_cluster_insights MCP tool.

​
Alert-triggered analysis

When an alert fires, Ankra collects the state of the affected resources from the cluster - pod status and restarts, warning events, recent container logs, job results and node conditions - and asks the AI for an analysis. The analysis is attached to the firing. To read it, open the alert and its Firings tab, and view the analysis on the firing’s row. It contains:
  • Quick Summary - what happened, in a few lines.
  • Root Cause - the underlying problem, not just the symptom.
  • Key Insights - observations about the resources involved, performance, timing, and the errors found.
  • Affected Resources - each with a link to the resource.
  • Recommended Actions - a checklist you can tick off as you work through it.
  • Full Analysis - the complete write-up.
Start Conversation with AI opens the AI Assistant with the analysis loaded, to ask follow-up questions or have it carry out a recommended action. A cluster’s overview also lists its recent AI Analyses. Alert analyses also appear as incidents in AI → Inbox, one stream for approvals, incidents, insights and activity, and a firing can start auto-remediation under the remediation policy.

​
Next

Decide what the AI may do on its own when an alert fires: AI Autonomy.