The local runner

The runner starts the official claude or codex CLI that is already installed and signed in on your computer. It runs only work that you own, in your organization. No one else's work runs on it, and the platform never bills it.

An owner or admin must allow runners for the organization and accept the vendor terms. See Subscriptions and vendor terms.

Run the install command from setup or from Runners. It needs Node.js 22 or newer, on macOS or Linux. On Windows, use WSL. The installer:

1. downloads the runner from your Agent server and checks it;
2. puts it in ~/.agent-runner and tells you how to add it to your PATH. It never edits your shell profile;
3. offers to install and sign in to Claude Code and Codex, on the vendors' own pages;
4. signs the runner in to Agent. A browser page opens; check the code and select Approve. Approve a runner only when you just started it yourself;
5. on macOS, starts the runner now and at every login. On Linux, run the start command it prints.

Run the install command again to update.

• agent-runner login: sign the runner in with a code in your browser.
• agent-runner start: start taking work.
• agent-runner doctor: show each CLI, its sign-in state and its sandbox.
• agent-runner repo connect and repo list: see Local repositories.
• agent-runner atlas scan: profile repositories for the Repository Atlas.
• agent-runner audit: a local look at where your own agent time and tokens go. It sends nothing.

Online, Offline, and Approved, not started. A runner can show Takes platform work or Your own tasks only.

Select Revoke on Runners. Its token stops working at once, and a task it runs is asked to stop.

Each person can have 5 runners and 50 queued tasks. One runner per computer.