Klotse, reviewed: a Kubernetes desktop that asks for your kubeconfig and nothing else

Every Kubernetes person has a kubectl history file that reads like a diary of regret. Six terminals, four contexts, one of them pointed at production when you thought it was staging, and a --context flag you’re typing from memory at 11pm. The desktop-GUI-for-Kubernetes genre has been trying to fix this for a decade, and most of it fails for the same reason: it wants to own your clusters, your auth, or both. Klotse is a native desktop manager that takes the opposite bet — it reads your existing kubeconfig and asks for nothing else.

The design decision that matters: kubeconfig-native

Klotse is a Python and PySide6 (Qt) desktop app built on the official Kubernetes Python client, for Windows, macOS and Linux. No server component, no login, no cluster agent, no vendor account. It uses the kubeconfig you already have and the auth helpers it already references. Nothing to sign in to, and — the part I care about — no new place where credentials live. For anyone whose threat model includes “yet another tool holding cluster tokens”, that single decision is most of the review.

It’s honest about maturity, too: the README leads with “Status: early” and a live-testing document describing what’s been verified against real clusters and what hasn’t. In a world of tools that ship “production-ready” in the launch tweet, that’s a green flag, not a red one.

What one window buys you

  • Multi-cluster without the mental gymnastics. Open contexts as tabs, group them, and — the clever bit — combine one resource kind across clusters in a single list. Select two objects and diff their YAML across clusters. The “why does staging work” question finally has a UI.
  • Read-only mode, per cluster, remembered. Flip your production context to read-only and every mutating action is refused until you say otherwise. This should be the default posture on any cluster that pays for something, and Klotse makes it a per-cluster setting instead of an RBAC project.
  • Logs, shells and files without leaving the window. Stream logs per container, per pod, per label selector with follow and regex filters; open shells and attach sessions; browse a pod’s filesystem; even peek into a PVC through a pod that already mounts it.
  • Filter expressions instead of grep. Lists take plain text plus expressions like Restarts > 3 or CPULim = 2*CPUReq — the “what needs attention” query becomes a saved filter instead of a kubectl incantation.
  • Dangerous verbs get previews. Cordon, uncordon and drain with a drain preview, bulk select with per-cluster confirmation, and scaling that accepts relative changes (*2, +1). Editor saves can go through a server-side dry run and a diff before anything is written.
  • Real usage charts. CPU and memory from metrics-server while the window is open, or genuine history from Prometheus/VictoriaMetrics — reached through the API server or an automatic port-forward when the API server can’t reach it.

There’s also a cluster-analysis menu that feels aimed squarely at homelab and small-team reality: capacity, health check, right-sizing, unused objects, upgrade preflight, RBAC lookup and Trivy Operator reports — all read-only inspections. Right-sizing from a single sample comes with an appropriate warning to compare against real history, which is the correct amount of humility.

Who it’s for

For me the sweet spot is the multi-cluster homelab and the consultant who inherits contexts: k3s here, EKS there, a customer’s cluster on Fridays. Klotse turns context-switching from a memory test into tabs. For a team, the change log — which records what you did through Klotse, without secret values — plus the “Can I…?” permissions lookup make it a decent junior-engineer cockpit: they can look at everything, and read-only plus dry-run review means looking is mostly what happens.

Getting it and the fine print

Windows and macOS have native packages (installer plus portable zip on Windows, dmgs for Apple silicon and Intel); they’re unsigned, so expect the Gatekeeper dance — on macOS that’s xattr -dr com.apple.quarantine /Applications/Klotse.app after moving it to Applications. Linux installs cleanly with pipx:

pipx install klotse
klotse --install-desktop-entry

You need kubectl on your PATH for drain, shells and port-forward; helm only to uninstall releases. Apache 2.0, plugins are off until enabled and their Python doesn’t load until you say so — the plugin manifest model is another small trust decision done right.

Verdict

Klotse won’t replace your terminal, and it isn’t trying to. What it replaces is the six-terminal context-switching tax and the clipboard-roulette of kubectl --context. The kubeconfig-native, no-server, read-only-by-choice design earns it a slot next to the terminal rather than a fight with it. Pin the version, keep prod read-only, and enjoy watching a young tool do the boring parts correctly.

Leave a Reply

Your email address will not be published. Required fields are marked *

WordPress Appliance - Powered by TurnKey Linux