On this page
Kepler is a desktop app for the people who keep systems running: SREs, platform engineers and whoever is on call tonight. You ask what is wrong in plain words. Kepler runs read-only commands, reads the dashboards you already use, recalls what happened last time, and comes back with the cause and a suggested fix. It shows every command it ran, and it asks before it changes anything.
What Kepler does
Ask, and see the evidence
Kepler runs kubectl, shell and cloud reads, shows each one as a card, and answers with the output as evidence. Nothing is a guess you cannot check.
A terminal and browser you share
The terminal and browser panels are the same ones Kepler uses. Sign in to Grafana once, then ask what p99 did in the last hour.
It remembers
Your services, your incidents, your runbooks and how your team works. Every session makes the next one shorter.
It keeps watch
Ask Kepler to watch a rollout or a fix. The check costs nothing while it is quiet, and Kepler looks into it when something changes.
Automations
A morning brief, a triage of every new alert, a deploy guard. Describe it in plain words and Kepler builds the steps, or start from a template.
Reach it from your phone
Connect Slack, Telegram, Discord or WhatsApp and keep working in the same session. Approvals work there too.
You decide what runs
Every conversation starts in Observe, which is read-only. Switch posture when you want Kepler to do more, one conversation at a time.
- Observe: Kepler reads, investigates and tells you what it would do.
- Assist: Kepler asks before every change and shows you the exact command.
- Yolo: Kepler acts on its own, within the limits you set.
- Delete, scale, restart and rm -rf are blocked in every posture. No setting unlocks them.
A war room for big incidents
Open a war room and describe the symptom. Kepler leads, and starts a member for each angle of the incident, each working in its own pane.
- Members pin what they find, so one finding reaches the whole room.
- Every change any member wants waits in one queue for your approval.
- Closing the room stops every member.
It knows your infrastructure
Kepler builds a map from what is already on your machine: your kubeconfig, SSH config, cloud CLIs and repositories. Ask about checkout and it already knows which cluster you mean.
- What deployed a workload, from which repository.
- Which workloads talk to a database.
- Where your Grafana, Prometheus and logs live.
Memory that grows
After each investigation Kepler records what broke, what caused it and what fixed it. Ask what you did the last time checkout crashlooped and it pulls the answer back.
Your rules live in kepler.md, plain Markdown you can edit. Everything stays on your machine.
Work on any machine
Turn on Remote access and point a conversation at a server, a cloud VM or a sandbox. Commands run there and every tool row says where. Kepler connects the way you do and never holds a password or a key.
It runs on your machine
Conversations, memory and settings stay on your computer. The only traffic out is to the model you choose and to your own tools. Kepler is not a monitoring service and installs nothing in your clusters: it uses the kubectl, cloud CLIs and dashboards you already have.
What you need
- macOS 13 or later on Apple Silicon, 64-bit Windows, or Linux (.deb or .rpm).
- A free Kepler sign-in with Google, GitHub or email.
- A model: the hosted Kepler models, your own key for Anthropic, OpenAI, Google, xAI, AWS Bedrock or OpenRouter, or a local model through Ollama.
- The tools already on your machine: kubectl, your cloud CLIs, your SSH config.
Kepler and RubixKube
Kepler works on its own. You do not need a RubixKube account to use it. With RubixKube, Kepler can also read the platform: ask what is broken in production right now and get the open incidents, the health of each environment and the RCAs in the same chat.
Questions
Can Kepler change production?
Only if you let it. New conversations start in Observe, which is read-only. In Assist, Kepler asks before every change. Destructive commands such as delete, scale and restart are blocked in every posture.
Does my data leave my machine?
Your conversations, memory and settings stay on your computer. Kepler sends prompts to the model you choose and calls the tools you connect. Nothing goes to RubixKube.
Which models can I use?
The hosted Kepler models with your sign-in, your own key for Anthropic, OpenAI, Google, xAI, AWS Bedrock, OpenRouter or any OpenAI-compatible endpoint, or a local model through Ollama.
Does it work on Windows and Linux?
Yes. Windows support is early: the terminal is PowerShell and Kepler does not run in the background there yet. Linux ships as a .deb and an .rpm.
Do I need RubixKube?
No. Kepler needs a free sign-in and a model. If you use RubixKube, Kepler can read its incidents and RCAs too.
I have spent most of my working life keeping other people's systems alive. Coding agents got a perfect world model for free. The rest of us got dashboards at 3am. Kepler is my attempt at the other half: an agent that learns your infrastructure the way a good colleague does, and asks before it touches anything.
Fifty operators used the closed beta and shaped what it is today.
Try Kepler on your own machine.
Download Kepler for macOS, Windows or Linux. Read the docs to get started.