AI Advice

The AI Advice tab is the newest and most proactive feature in Kure Monitor. While normal pod monitoring waits for a pod to crash, the Advice engine scans your healthy and running workloads to find architectural, scheduling, and hygiene issues before they cause outages.

How it works

Kure Monitor currently ships with 38+ advice detectors covering:

• Resource Hygiene: Missing CPU/Memory requests, improper limits, or CPU limit throttling risks.
• Scheduling & Availability: Over-concentrated pods, missing pod anti-affinity, or missing disruption budgets.
• Networking: Missing readiness probes or overly permissive network policies.
• Lifecycle & Storage: Unsafe termination grace periods or emptyDir abuse.

When Kure scans your cluster, it surfaces these findings in the Advice tab, categorized into Active and Ignored lists.

Lazy AI Explanations

To save on LLM API costs and execution time, Kure Monitor evaluates all 38 detectors using deterministic, blazingly-fast Go code.

The LLM is only invoked lazily on card expand. When you click on a specific advice finding, Kure sends the exact evidence dictionary to your configured LLM (one call per finding, which is then heavily cached).

This approach ensures:

1. Cost Efficiency: A scan with 500 findings does not cost 500 LLM requests up front.
2. High Accuracy: The prompt is strictly constrained to only reference data present in the finding’s evidence dict. The AI cannot invent replica counts, image names, ports, or labels. It simply explains why the evidence is problematic and how to fix it.