# Kubeast — a self-hosted Kubernetes console with an AI assistant that can only see what you can see — retained record

> Retained reddit record from reddit:selfhosted.

Canonical: https://highsignal.app/data/records/bb50eac5ec56cfe7

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:selfhosted)  published 2026-10-23 14:25 ingested 2026-10-24 10:58

[original source ↗](https://www.reddit.com/r/selfhosted/comments/1xdfick/kubeast_a_selfhosted_kubernetes_console_with_an/)

## retained text



```
I've been building Kubeast, an open-source (MIT) web console for Kubernetes, and it's at a point where I'd love feedback.

- One console for many clusters: resources, YAML edits, live logs, a shell in the browser, Helm history and rollbacks.
- An AI assistant that reads the cluster with tools, using the signed-in user's permissions: their Kubeast role, enforced by the cluster through impersonation. It can't see or change more than they can.
- Passwords, keys and tokens are masked before anything reaches the model. It works with Ollama if you want nothing to leave your network.
- Every write and sensitive read goes to a hash-chained audit log, and access can be granted temporarily with approval.
- One Helm command (or Docker Compose) to install. No agent on the clusters it manages, only RBAC (the optional node shell aside).

Site with short clips of each feature: https://junginho0901.github.io/Kubeast/

GitHub: https://github.com/junginho0901/Kubeast

Full disclosure: I'm the author, and I used AI (Claude) heavily while building it. The permission model, masking and audit log are covered by about 500 end-to-end tests against real kind clusters. I'd especially like to hear what you'd need before trusting a tool like this on your own cluster.

```



document fetched 2026-10-24 10:32 · canonical https://reddit.com/r/selfhosted/comments/1xdfick/kubeast_a_selfhosted_kubernetes_console_with_an
