A five minute environment walkthrough

Use this page as a presentation in a browser, or print it to PDF. Each step links to the operational detail; the underlying diagrams are shared with the rest of the site.

1 Choose where work runs

The Mac remains your interactive workstation. Select Shipyard in Orca to use Linux compute. Closing the UI does not stop tmux processes or Docker services. Environment view.

2 Start an isolated API task

Orca creates a worktree and waits for its setup hook. Setup prepares isolated services, initializes data when new, reconciles authorization, starts the API and waits for health. Later restarts preserve data. Task workflow.

3 Add dashboard when the task needs it

The same agent creates a companion dashboard checkout. The preview targets that task’s API. Only one managed preview runs, so switching requires coordination. API and dashboard still have independent Git histories and reviews.

4 See the browser that Linux is running

Chromium renders on Shipyard. Playwright and the human viewer control the same browser session. Pause the agent before taking over; resume when finished. Browser downloads stay on Linux. Browser workflow.

5 Build appliance code independently

Search-engine tasks have isolated runtime and test dependencies. A shared Cargo cache makes iteration cheaper, while copied binaries prevent a neighboring build from replacing a running task. Real cameras and GPU inference are outside this setup. Appliance workflow.

6 Finish without losing data

Use Orca desktop Delete for the verified remote archive path. Hooks stop owned services, preserve volumes and retain companion edits. This is not automatic disk cleanup. Archive recovery.

7 Extend the map

A device-operating agent can contribute another machine using stable identifiers, evidence and the same templates. The architecture explorer and host reference then grow through the normal contribution workflow. Add a machine.