Get started
What Aetox is
A harness for AI agents: one window, two desks, four rooms, and a model you bring yourself. Read this before installing and the rest of the setup makes sense.
Aetox is a harness for AI agents. An agent is the model plus the hands, the memory and the rules it works under — and the harness is everything around the model: a real terminal, real files, a real browser, a language server, memory it is allowed to keep, skills it opens when a job calls for one, and the checks that must pass before it says the work is done.
You bring the model — an account you already pay for, or one running on your own GPU. The model thinks; the harness is what lets it act on your machine instead of only talking about it.
Two desks in one app
| Desk | For | What it gets |
|---|---|---|
| Assistant | Documents, meetings, research, pictures, day-to-day work | Files, browser, terminal, the office and its teams |
| Code | A project folder on this machine | The project's files, its language server, its tests and git |
Switch between them from the bar at the top. They are two desks, not two apps: each has its own instructions and its own memory, so a conversation about code does not fill up the desk you use for documents.
Four rooms in the same window
The conversation sits next to the work: a slide, a browser, files and a terminal. The Code desk adds git and a language server. You watch a step happen instead of reading that it happened, and you can stop it at any step.
What the harness gives an agent
- Hands. Every step is a real tool call: write the file, run the command, open the page, click it.
- A context of its own. What it knows is grouped and linked, and a run sends only the points the job touches — not everything that exists.
- Skills and MCP on a shelf. The model sees names and opens one when it needs it, so fifty skills cost what five cost.
- A check before it hands back. It runs the project's tests, opens the file it made, and says what it could not prove.