Skip to content
AndonDocs
GitHub
Run machines and deploy / Machines

Remove a machine

View as MarkdownThe same text, at this page's address plus .mdOpen in ClaudeAsk questions about this pageOpen in ChatGPTAsk questions about this page

Take a machine off your board when you retire it, give it away, or enrolled it by mistake. Revoking it on the board cuts its access at once; uninstalling the agent on the computer stops it and clears its files. Do both.

1. Revoke it on the board

  1. Open Settings → Machines.
  2. On the machine’s card, open the ⋯ menu and choose Revoke…. Only owners and admins see this menu.
  3. Confirm with Revoke.

The machine loses all access at once and takes no more work. Sessions already running on it keep running there until you stop them. A revoked machine is hidden from the list; turn on Show revoked to see it.

Its cards move to another of your machines that has their team’s repositories: a card pinned to it, its queued runs and its pull request’s checks and reviews go there, and a card linked to one of its worktrees moves only to a machine with the same worktree. A machine on the same computer is picked first. A card no other machine can take waits, its queued run saying the machine it was sent to is gone, until one can.

2. Stop the agent on the computer

  1. Take the use-andon skill out of your coding agents first, while the agent can still do it:

    shell
    andon policy cli-skill off
  2. Stop the service:

    shell
    andon uninstall-service

    On a Mac this takes the LaunchAgent out of launchd; on Linux it disables and stops the systemd user unit. The service file itself stays.

  3. End any run sessions still open: tmux -L andon ls lists them, and tmux -L andon kill-server ends them all. Commit or copy anything you want from their worktrees first.

For a second agent on the same computer, put --profile <name> before each command (One machine, several boards).

3. Remove its files

Nothing above deletes files. To clear the agent off the computer, move these to the Trash:

What Mac Linux
Service file ~/Library/LaunchAgents/dev.andon.agent.plist ~/.config/systemd/user/andon-agent.service
The andon command ~/.local/bin/andon ~/.local/bin/andon
The agent, its terminal helper and skill ~/.local/share/andon/ ~/.local/share/andon/
Config, credential file and host id ~/.config/andon/ ~/.config/andon/
Run state, accounts, Codex and Gemini homes ~/.local/state/andon/ ~/.local/state/andon/
Logs ~/Library/Logs/andon/ the journal (journalctl --user -u andon-agent.service)

Before you move ~/.config/andon/ to the Trash, note the machineId in its config.json, and remove the machine credential from the system store with it:

  • Mac: security delete-generic-password -s andon-agent -a <machineId> (or delete that andon-agent entry in Keychain Access).
  • Linux desktop: secret-tool clear service andon-agent account <machineId>. On a server it was the credentials.json file in ~/.config/andon/.

Every agent on the computer stores its credential under the service andon-agent, so always name the account: without it, these commands can remove another board’s agent’s credential.

Card worktrees live in your projects folder (~/Projects/worktrees/ unless you changed projectsRoot). Move them to the Trash once you no longer need their branches.

If the machine runs agents for other boards too, keep the folders in the table and ~/.local/bin/andon: in each folder, remove only the profile’s profiles/<name>/, and its service file (dev.andon.agent.<name>.plist on a Mac, andon-agent-<name>.service on Linux). Take the profile’s machineId from ~/.config/andon/profiles/<name>/config.json.

Put a machine back

To enroll a revoked machine again, run setup on it and approve it on the board:

shell
andon setup --url https://<your board>

The board installer refuses to run over an existing install, so use andon setup while the agent’s files are still there, and the installer once you have removed them.

The machine comes back as a new machine. When it first syncs its repositories, the cards still on the revoked one move to it: the ones whose team’s repositories it has, with the worktrees it finds again at the same path or on the same branch. Their queued runs go to it, so a review round or CI fix waiting for the old machine starts there.