Settings

Themes, fonts and density, Prometheus, the node shell image, the debug image, usage history, AI features, extensions, updates and saved views.

Settings is the last item in the sidebar’s Tools.

Display

  • Theme: 12 themes in three groups, plus System.

    • Apple: Dawn (light, the default), Dusk (dark), System (follows your OS), and the high-contrast Dusk HC and Dawn HC.
    • VSCode: VS Dark+ and VS Light+.
    • Community favourites: One Dark, Dracula, Nord, GitHub Dark, GitHub Light and Catppuccin (Mocha).

    Reset to system goes back to following your OS. Every theme meets WCAG contrast for text: 4.5:1, and 7:1 in the high-contrast themes.

  • Font size: S (13px), M (14px), L (15px) or XL (16px).

  • Font family: Roboto (the default), JetBrains Mono or System Mono.

  • Table density: Compact, Default or Relaxed.

Prometheus

Kubebay reads Prometheus for a pod’s graphs, right-sizing over 7 days, and GPU utilisation.

You usually don’t need to set anything. For each cluster, Kubebay looks for a Prometheus Service by its usual names and labels (prometheus-server, kube-prometheus-stack-prometheus, prometheus-k8s, thanos-query, app=prometheus, and so on), skipping Alertmanager, exporters and the operator. It then queries it through the cluster’s API server, with your own credentials, so no port-forward is needed. This card shows Auto-detected for <cluster>: monitoring/prometheus-server:80 with a Change button, which lists the other Prometheus services it found so you can pick one, or go back to the best match. Use my own URL in that dialog switches to a URL instead. The detection is cached for a few minutes and never holds up a page. If you aren’t allowed to list Services, or none looks like Prometheus, the card says none was found and you set a URL. A URL you set always wins over auto-detection. To set one, set the default Prometheus URL here, such as http://localhost:9090, for clusters that don’t have their own. Set a cluster’s own URL from its ⋮ menu on the Clusters page, since a port-forward usually reaches one cluster’s Prometheus.

Kubebay refuses a Prometheus URL that points at a cloud metadata or link-local address (such as http://169.254.169.254), at an address that isn’t http or https, or that carries a user name and password, and the setting isn’t saved. That keeps the engine from being aimed at a cloud node’s credentials. A local address such as http://localhost:9090 is fine on your own machine. When Kubebay runs in a cluster, loopback is refused too, since it’s the engine’s own pod.

Kubeconfig sources

Kubeconfig files are added and removed on the Clusters page, next to the clusters they bring in. Manage kubeconfig files takes you there. See Connect your clusters.

Node shell image

The image for the privileged pod behind node shell. Set it to a mirror in your own registry if the cluster can’t pull from registry.k8s.io.

Debug image

The image Debug with ephemeral container offers in a pod’s Terminal tab. Leave it empty for the default (busybox:1.36), or set one with the tools you want (nicolaka/netshoot) or a mirror in your own registry.

Usage history

Kubebay records hourly usage per namespace for the clusters you connect to, only while the app is open, and keeps it for 35 days. It stays on your computer, in Kubebay’s folder in your user config directory. It powers the Clusters page’s CPU sparkline, and the headroom forecast and unusual-usage cards on Cost / Waste.

Each cluster you’ve connected to is listed as recording or stopped. Stop and Resume control recording. Clear deletes a cluster’s history.

Usage history isn’t recorded when the engine runs in a cluster.

Warm clusters

How many connected clusters keep streaming live, and for how long one you haven’t looked at stays warm. Set Warm clusters (0 to 50, 0 for no limit) and Idle minutes (0 to 1440, 0 for never), then Save; the engine applies it at once, and lowering the number puts the extra least recently viewed clusters to sleep right away. Reset to defaults puts back 5 clusters and 30 minutes. The cluster you’re on and your pinned clusters never go to sleep. A value saved here wins over the --warm-clusters and --warm-idle flags. The fields are disabled, with the reason, when Kubebay runs in a cluster or with OIDC, where streams are shared between users. See Switching clusters.

AI assistants (MCP)

Lets an AI assistant such as Claude read the clusters you choose. See AI assistants.

AI incident triage

Adds a Triage tab to pods, which asks a model what’s wrong. See AI triage.

Extensions

Where ~/.kubebay/extensions.yaml is, what Kubebay loaded from it, and why it skipped any entries. Reload reads the file again. See Extensions.

Updates

In the desktop app only. Check for updates automatically turns the daily check on or off; the check only looks for a new version, it never downloads one. Check now checks at once, and Update now downloads and installs a version that was found. See Install.

Saved views

Every saved view: its name, table, namespaces and filter. Delete one, Export them all to a file, or Import… a file.