Alpha software. Expect bugs. noHuman Team runs AI workers on your own machine with real file, browser and desktop access — start on work you can afford to have go wrong, and leave the approval queue on. What is rough right now

Back to home

Desktop Workers

Some tasks need more than text — they need a browser, a running application, or a visual check. Desktop workers give your noHumans a screen to work with: your real desktop.

7.1 What Changed

Earlier versions of noHuman Team gave each noHuman a simulated Linux desktop viewed over VNC. Those virtual desktops have been removed. noHuman Team is now a native app, and desktop-capable noHumans (the Automator in the Startup Team, for example) drive the same computer you are sitting at.

Old links still work. This page keeps its address so bookmarks and search results don't break, but the VNC viewer, the Desktops page and the "virtual display" no longer exist.

7.2 How Desktop Workers See and Act

A desktop worker takes a screenshot of your display, reasons about what it sees, and then clicks and types using the operating system's own input layer. Support differs per OS:

  • macOS — full support. Screenshots and input via cliclick, gated by the two permissions below.
  • Windows — screenshots are supported; desktop click and type is experimental. Prefer browser tools for web work.

On macOS, two permissions make this possible, and both are granted to the noHuman Team app itself, not to a terminal or to node:

  • Screen Recording — lets the worker capture what is on screen. System Settings → Privacy & Security → Screen Recording → enable noHuman Team.
  • Accessibility — lets the worker move the mouse and press keys. System Settings → Privacy & Security → Accessibility → enable noHuman Team.

macOS prompts for both the first time a worker needs them. If a worker reports that it cannot see the screen, check that the grant landed on noHuman Team and restart the app.

7.3 The Browser Panel: Watch and Take Control

Every noHuman with browser tools has its own Chrome browser with a persistent profile. It is a real window on your machine, and it also shows up live inside the dashboard: open a noHuman and use Browser in the conversation toolbar to open the right panel. There is no remote-desktop viewer and nothing to install.

The browser starts on demand. Press Open browser and the panel fills with a live picture of what the noHuman's browser is showing. It looks like a browser: a row of tabs on top, then back, forward, reload and the address.

  • Looking around is always allowed. Scroll the page and click any tab in the tab row to look at it. Neither one touches the real Chrome window, so you never pull Chrome in front of what you are doing, and the noHuman is not interrupted.
  • The picture follows the noHuman. The tab its browser tools are working in carries a small robot icon, and the view goes there on its own, even when that tab is in the background of the real window. Pick a different tab and the view stays with you; press Follow noHuman to go back.
  • Take control when you want to help: click, drag, type and paste straight into the picture, use back, forward, reload and the address bar, and close tabs. Press Back to watching to hand it back.
  • Show window brings the real Chrome window to the front. Use it to sign in to a site with your password manager or a passkey, which only work in the real window. The noHuman keeps the session afterwards.
  • You share one browser. While you are in control the noHuman can still act in it, so take control when it is idle or tell it to wait.

The address bar accepts ordinary web addresses only. The live view is available to your dashboard session alone, and taking control is written to the audit log. If the picture stays blank, the Chrome window is minimised: Chrome stops drawing a minimised window, and Show window brings it back.

7.4 Use Cases

Web Research

"Research the top 5 project management tools and summarize their pricing."

The noHuman opens its browser, visits each site, reads the pricing pages, and compiles a summary.

Form Filling

"Fill out the vendor application form at example.com/apply with our company details."

The noHuman navigates to the form, fills in each field, and can even submit it (if you tell it to).

Website Testing

"Open the landing page you just built and check if everything looks right."

The Developer opens the page in its browser, inspects the layout, and reports any issues.

Native Apps

"Open Numbers, add this week's figures to the sales sheet, and export it as PDF."

Because the worker drives your real desktop, it can operate any application you can — with the same access you have, so be deliberate about what you ask.

7.5 Limitations & Tips

What desktop workers can do:

  • Browse websites in their own Chromium
  • Fill out forms, click buttons, navigate
  • Read and extract text from pages
  • Take screenshots of your desktop and act on them
  • Operate native desktop applications on Mac; Windows desktop input is experimental

What they can't do:

  • Work on a separate screen — they share your desktop, so a worker clicking while you type will collide
  • Handle CAPTCHAs reliably
  • Log into sites that require 2FA (unless you provide the code, or sign in yourself from the Browser panel or the real window)

Tips:

  • Prefer the browser tools for web work. They act on the page structure directly and don't need the screen at all.
  • Watch in the Browser panel. It is quicker than hunting for the right Chrome window, and it works for every noHuman from one place.
  • Be specific about URLs. Tell the noHuman exactly where to go: "Go to example.com/pricing" not just "check their pricing."
  • Step away during desktop tasks. A desktop worker uses your mouse and keyboard; use the autonomy dial (ask-first or guarded) if you want to approve each action.
  • Check the browser window if a task seems stuck. The noHuman might be waiting on a page load or an unexpected popup.