AI triage
Ask a model why a pod is failing. Kubebay gathers the evidence, shows exactly what it would send, and sends it only when you say so.
AI triage adds a Triage tab to pods. It gathers the evidence about one pod, asks a model what’s wrong, and shows each finding beside the evidence it cites. It’s off until you turn it on for a cluster.
Turning it on
In Settings → AI incident triage:
- Tick the clusters whose pods get a Triage tab.
- Set the Endpoint, such as
https://api.anthropic.com, and the Model. - Add an API key. Save key stores it in your OS keychain. Without a keychain, set
KUBEBAY_TRIAGE_API_KEYin Kubebay’s environment instead. The key is never shown again, never stored in Kubebay’s settings, and Remove key deletes it. - Click Turn on.
Using it
Open a pod’s drawer, then Triage.
- Gather evidence collects a bounded bundle: the pod, its owners, warning events, recent rollouts and logs. Credentials in it are masked.
- Kubebay shows What would be sent: the endpoint, the model, about how many tokens, how many values were masked, and whether logs were cut to fit. Copy evidence copies it, if you’d rather paste it somewhere yourself.
- Nothing has been sent yet. Send to model makes exactly that request, once.
- The answer lists likely causes, each citing the evidence it’s based on, then Next steps, for you to review. If the evidence doesn’t support a cause, it says so, and says what evidence would help.
The model only reads. Nothing in the cluster changes. Every preview and every send is written to the Audit Log.
The answer is generated by a model from the evidence shown. Check it before you act on it.
Where it’s available
Triage is part of Kubebay on your own machine. It isn’t available when the engine runs in a cluster or with OIDC sign-in.