Computer use
Control a graphical desktop from the CLI or SDK, and share its viewer with a person.
Desktop-enabled large computers run Chromium, a terminal and a file manager. They use the same isolated VM and files as your commands. Small computers keep Lightpanda; graphics start only when requested.
New large computers include the desktop. Use CLI 0.1.9 or newer and JavaScript SDK 0.4.0 or newer. Python 0.3.0 is prepared as a source build; its PyPI release is pending. Existing computers keep their original image.
workpaper run research --size large --json
workpaper desktop research start --json
workpaper desktop research launch browser --json
workpaper desktop research screenshot /tmp/research.png --json
workpaper desktop research press ctrl l --json
workpaper desktop research type --json -- "https://example.com"
workpaper desktop research press enter --jsonHuman takeover
workpaper desktop research url --json
workpaper desktop research openThe hosted viewer at browser.workpaper.co checks your sign-in and computer ownership. No local port is needed. Each VM has a separate URL; tabs for the same VM share its screen and input. Closing a viewer does not stop the VM.
Files and stopping
Use /workspace; /home/agent points to the same directory. workpaper desktop research stop stops graphics only. workpaper stop research pauses the whole computer and keeps disk and memory. Forks are independent copies.
Teach your agent
Run workpaper agents --json, or give the agent the complete agent guide. It covers screenshot coordinates, literal typing, keyboard shortcuts, CLI errors, JS and sync/async Python SDKs, files, pause/resume, and human takeover.