Computer use

Most AI stops at generating text. A Busy Bee agent gets its own real computer and uses it — a live browser, a terminal, and a file system — the same way a person at a keyboard would.

Its own isolated machine

Every project runs on its own machine, separate from every other project. No shared state, no cross-contamination. On that machine an agent can:

• Use the web — navigate sites, log in, fill forms, and click through real web apps in a live Chromium browser.
• Run commands — open a terminal to install packages, run scripts, and execute test suites.
• Work with files — create, read, and edit files on a persistent file system.

How it operates the screen

The agent works the desktop through a tight loop of real actions: take a screenshot to see what's on screen, then click, type text, press a key (with modifiers like Ctrl or Shift), or move the mouse to hover. It looks before each move, checks the result, and adjusts — instead of just describing steps it can't take.

Watch exactly what it did

Every computer-use session is recorded end to end. Open the Replay tab on any task to scrub through a video of the whole session, with a timeline of screenshots the agent captured along the way. Nothing happens off-screen — you can review each click and keystroke after the fact.

Recording finalizes even when a task fails partway, so you can usually review what happened right up to the point it stopped.

Availability

Computer use is an opt-in capability. It has to be enabled for your organization, and each agent type has to be set to run in the visual (GUI) machine before its agents get a browser and terminal. If an agent isn't using a computer, that toggle is the first thing to check.

Where it fits

Computer use powers the hands-on work behind many playbooks and features:

• Web research — driving real sites and pulling data from them.
• Code and GitHub — building, testing, and committing code.
• Data and analysis — gathering and crunching data on the machine.