Install. Connect. Stay in control.

A real control panel on your computer.

Install Git, Python 3.12 or newer with Tk, and Docker Engine 28 or newer. Windows and Mac use Docker Desktop’s Linux engine. Copy the verified command from the installation page into PowerShell or Terminal.

Windows adds Start → Tbench → Tbench Actions. Mac adds ~/Applications/Tbench Actions.app. Linux uses the CLI; a graphical session with Tk can also open tbench-worker desktop. The app manages your connection and bounded runner session, rather than opening an installation script in a text editor.

Setup installs under your user profile. It does not request administrator access, modify certificate trust, add a startup service or silently change PATH. Use the printed launcher path if tbench-worker is not on PATH.

Authorize in the browser you already trust.

1. Choose Sign in with browser, or run tbench-worker auth login --name my-device.
2. Sign in to your approved Tbench organization using its existing account or configured company SSO.
3. An organization owner or admin matches the displayed code and reviews the device name, platform and scope before approving.
4. Return to the app. The scoped device connection is saved locally; no jobs start yet.

The ten-minute device request uses a proof challenge, bounded polling and a one-time approval. App keys and browser session cookies are never handed to the CLI. New credentials use Windows DPAPI or macOS Keychain; Linux keeps an owner-private file. This grants device execution only, not account administration or unrestricted API access.

If you are a developer without enrollment permission, ask your organization’s owner or admin to review your code. Never approve a device request somebody unexpectedly sends you.

Your push starts the session.

From an approved checkout, run tbench-worker push. It verifies your GitHub CLI identity, pushes the current branch and runs only the matching GitHub jobs on this computer. Add --ai-profile /path/to/review.json for an approved AI review and repair profile. The session is revoked after confirmed cleanup.

Desktop-managed sessions are a separate optional workflow.

Press Start runner and confirm participation. The session stops after eight hours or ten jobs. Stop runner asks the owned worker to stop and complete cleanup. Closing the app waits for its worker cleanup rather than killing it mid-job.

tbench-worker auth status
tbench-worker check
tbench-worker serve --max-hours 8 --max-jobs 10
tbench-worker stop

The GitHub workflow defines the job steps. Native Linux containers use x64 or ARM64 matching the Docker engine. macOS and Xcode builds use the separate native execution backend; the desktop app does not convert a Linux job into an Xcode job.

Revoke the device, preserve your history.

Stop active work and resolve recovery records, then choose Sign out device or run tbench-worker auth logout. The server confirms revocation before the app removes this device’s local credential. If revocation cannot be confirmed, credentials remain so you can retry safely. Your job history, journal and repositories are not deleted.

An organization manager can also revoke a lost device from the execution console. A browser sign-out and device revocation are separate operations.

One current release record.

Channel

v0.3.0-k21

Signed payload files

21

MacBook qualification

Client hardware test pending

The command pins the bootstrap hash and verifies the signed worker payload. These are locally generated desktop launchers around the Python runtime, not downloadable publisher-signed MSI/PKG installers. Existing releases remain immutable.