A clearer view of AI.
THE INDEPENDENT AI.

WORTH TESTING · 1 NOVEMBER 2026

agent-desktop: AI agents using desktop apps, the 20-second demo

A silent showreel posted by Chris Tate of Vercel Labs, plausibly genuine. Verdict: developer level, intermediate to advanced; you need a terminal and a coding agent. It shows three labelled cursors working in Notes, Reminders and Calculator on a Mac, an agent opening Event Viewer on Windows 11, and a terminal agent ('fx' with a DeepSeek model) reading an app's controls and clicking them by id. It's edited, so it can't prove that real input runs in parallel (the repository says one agent at a time), and Linux isn't shown.

Try this

  1. Run more than one agent and see whether each gets its own labelled cursor. 0:00
  2. Check that the cursor labels tell you which agent is acting. 0:00
  3. Try two agents typing into different apps at once, and note whether input really runs in parallel. 0:02
  4. Try it on Windows 11. 0:08
  5. Ask an agent to find its way through nested system menus (the demo uses Event Viewer). 0:11
  6. Run 'agent-desktop snapshot --app <App> -i' and read the list of controls it finds. 0:13
  7. Click by element id ('agent-desktop click @<id>') instead of screen coordinates. 0:14
  8. Drive it from a terminal and watch the cursor overlay follow along. 0:15
  9. Have the agent loop over several controls (the demo ticks four checkboxes) and check the result. 0:16
  10. Use it from a different agent or model (the demo uses 'fx' with a DeepSeek model). 0:19

Timestamps point to the moment in the source video.

Cost

Free and open source (Apache-2.0); you need a coding agent.

Source

Hype or Fact?

Partly fact agent-desktop: 'agents use your computer in parallel, each with its own cursor'

Related AI tools

How we knowSteps to try, not a claim that we independently tested every result. Follow the linked sources and check current prices. Our methods & sources ↗