Kubexy AI Assistant

Kubexy is the AI assistant built into Kubexer. Ask it about your cluster in plain language — "why is this pod crashing?", "what's broken in the payments namespace?", "which pod uses the most CPU?" — and it inspects the live cluster and answers, citing the actual resources, statuses, and events it found.

Kubexy is BYOK (bring your own key): you connect your own AI provider, and your key stays on your machine.

Bring your own key (BYOK)

Kubexy doesn't ship a model or proxy through Kubexer's servers. You supply an API key for the provider you already use:

Anthropic (Claude)
OpenAI
Google (Gemini)
OpenAI-compatible endpoints via a custom base URL — e.g. Ollama Cloud (https://ollama.com/v1), OpenRouter, Azure OpenAI, or a local model.

Your key is encrypted at rest on your device (via the OS keychain) and is used to call your chosen provider directly. It is never sent to Kubexer. The model you pick must support tool/function calling — that's how Kubexy reads your cluster. The Test button in settings verifies both the key and tool support.

What it can do

Kubexy works through a set of read-only "tools" that pull live data the same way the rest of the app does:

Diagnose pods — explain CrashLoopBackOff, OOMKilled, ImagePullBackOff, exit codes, and restart loops (it even reads the previous/crashed container's logs).
Investigate namespaces — health summaries, failing pods, and recent warning events.
Resources & capacity — requests/limits vs live usage, node allocation, pods missing limits, and quota status.
Live usage — top pods/nodes by CPU or memory (requires metrics-server).
Networking & storage — service endpoints, ingress backends, network policies, and stuck PVCs.
Anything else — list and describe any resource, fetch logs, and read events.

Taking action (with confirmation)

By default Kubexy is read-only — it can look but not touch. If you enable actions in settings, it can also restart, scale, suspend, edit (YAML), and delete resources — but every change requires your explicit confirmation. Kubexy proposes the change, you see a confirmation card (with a before/after diff for edits), and nothing happens until you approve.

Privacy

Cluster data you ask about (pod statuses, logs, events) is sent to your chosen AI provider so the model can reason over it. Secret values are redacted before they ever leave your machine — Kubexy sees the key names, not the contents.

Using it

Open Settings → Kubexy, choose a provider, paste your API key, and click Test.
Open Kubexy from the sidebar and ask a question, or click Ask Kubexy in any Pod or Deployment drawer to start a diagnosis scoped to that resource.
Export any conversation to PDF from the chat header to share or keep a record.

Notes

Kubexy is a Pro feature. Because it's BYOK, you bring and pay for your own AI provider usage; Pro unlocks the assistant.