Send off ten tasks. Go do something else.
A few on one machine, or one on each. Pick any of them up from your phone or your desk.
Cloud coding agents, on your own server.
Kellix is a control plane for AI. Chat with an agent, add your machines, and the work happens on them around the clock. You own all of it, end to end.
self-hosted · your machines · any model, cloud or local
Let's see an example.
An invoice dashboard, from the first message to a change the agent checked itself.
Hand off
Describe the change and walk away. The agent builds it, runs it, and tells you when it is done.
Machines
Choose where it runs: your server, a worktree of its own, or a fresh virtual machine.
Computer use
The agent opens your app on a real desktop and checks its own work. Take control whenever you want to.
Secrets
Credentials stay in your vault. The agent can use them, and it never sees them.
One install, all of it.
Everything runs on hardware you control, with the models and keys you bring.
what you get
self-hosted- models
- OpenAI, Anthropic, Google, Ollama, vLLM, or any OpenAI-compatible API
- sandboxes
- Virtual machines on your own hypervisor, with a desktop when needed
- secrets
- An encrypted vault, injected at the network edge
- agents
- Orchestrators and subagents, each with its own model
- people
- One install, a private workspace for everyone
- phone
- An installable app with push notifications
- voice
- Dictation through any Whisper-compatible endpoint
- network
- Machines only dial out, so they work behind NAT
your hardware · your models · your keys
Questions people actually ask.
Hardware, models, privacy, and how it stays up to date.
What is Kellix?
A self-hosted control plane for AI coding agents. You chat with an agent, add machines to the control plane, and start sessions that do the work on those machines, with git, a shell, a browser, and a full desktop when the agent needs to check its own work. You reach the same live session from any device.
How is it different from Cursor cloud agents or Devin?
The experience is the same idea: hand off a task, let it run in the background, and come back to a finished change. The difference is ownership. Kellix runs on your own server, the agents work on machines you control, and you bring the models and keys. Nothing about it is rented.
What hardware do I need?
A Linux host for Kellix itself (a small VM is plenty), and at least one machine for sessions to run on. That can be any computer you enroll with a one-line install, or a virtual machine Kellix builds for you on a paired hypervisor.
Which models can I use?
Any OpenAI-compatible API, or a local runtime like Ollama or vLLM. You bring the keys, and you can pair a strong model for planning with a fast one for the busywork.
Is my code sent anywhere?
Only to the model provider you pick, and nothing else leaves your machines. Point it at a local model and nothing leaves at all.
Do my machines need open ports?
No. Every enrolled machine dials out to Kellix, so it can sit behind NAT or a firewall and still run sessions.
How do updates work?
From the dashboard. Open Settings, check for updates, and choose Update now. Your sessions, data, and vault carry over, and every enrolled machine updates itself right after.
Can multiple people use it?
Yes. One install serves everyone, and each person gets their own machines, projects, and a private workspace with their own secrets.
Install it on any Linux host.
One command adds Kellix as a normal service. Open the address it prints, create your profile, and you're in.
curl -fsSL https://downloads.kellix.ai/main/install.sh | sudo bash
- 1 Run the command on a Linux host.
- 2 Open the address it prints and create your profile.
- 3 Enroll your first machine and start building.