Chapters

User manual

The Dozer Sandbox manual

Every chapter of the Dozer Sandbox manual, for the doz command and its dashboard.

25 chapters. Built from dozer-sandbox 33193c5.

Dozer Sandbox runs AI coding agents in small Linux virtual machines on your Mac: each agent gets its own disk, a network that only reaches what you allow, and your project folder if you share it. A sandbox pauses in a millisecond and wakes from disk in a third of a second, with everything where you left it.

You drive it with the doz command or its dashboard in your browser (doz ui). Every page here shows both.

Reading order#

New to Dozer? Read these first, in order:

  1. Getting started — install, set up, and talk to Claude in a sandbox in 15 minutes.
  2. Sandboxes and their lifecycle — pause, sleep, hibernate, wake; what's kept where.
  3. Terminals and sessions — attaching, the Ctrl-] menu, what the agent is doing, the clipboard, tmux.
  4. What the agent can do — permissions and the network.

Then, as you need them:

Conventions#

  • Commands are shown as you type them: doz up. NAME is a sandbox's name; my-app is an example.
  • Every command takes --json (machine-readable output), --store DIR, -v (every step) and -q (no progress). doz help COMMAND shows a command's options.
  • Settings are named section.name, like sandbox.clipboard; see the Settings reference.
  • Screenshots use the light theme; the dashboard follows your Mac's appearance.

This manual is checked against doz itself on every build: every command, flag and setting it names exists, and the settings reference is generated from the program.

The command-line user guide and the walkthrough cover the same ground for command-line users, more tersely.

Edit this page on GitHub