Your employee controls your actual desktop, not just a browser tab. It clicks, types, manages files, and runs terminal commands in real applications on your machine, the same way you would sitting at the keyboard. It works through the free Desktop Companion app and covers native software, file management, and shell tasks in one capability.
Your employee controls your actual desktop, not just a browser tab. It clicks, types, manages files, and runs terminal commands in real applications on your machine, the same way you would sitting at the keyboard. It works through the free Desktop Companion app and covers native software, file management, and shell tasks in one capability, backed by a single tool the employee calls Computer Controller.
It works by taking a screenshot of everything connected to your machine, stitched into one image, then reasoning over what it sees the way you would. If you run more than one monitor, the screenshot includes every display in that stitch, with the exact pixel offset of each one, so a click lands on the right screen even when your setup is multi-monitor. Before a risky click, it will move the mouse there first so you can see where it's aiming.
This is different from the employee's browser connection, which reads the actual DOM of a web page: form fields, buttons, structure. Computer Controller has no such map to read. It only has pixels, so it is the tool for anything without a page to parse: Figma, Excel, Logic, Final Cut, Photoshop, or any other native app on your machine. For ordinary website work the employee reaches for the browser tool instead, because it's faster and more reliable there.
Everything runs through a single companion connection shared across your employee roster, but you switch it on per employee. Available on the Founder plan and above, it needs the Desktop Companion app installed and paired to your account, plus Screen Recording and Accessibility permission granted in System Settings on first use.
On the desktop side: take a screenshot, click (with modifier keys like shift or command), move the mouse, type text, press a key or shortcut combo, and scroll in any direction. On the file side: list a directory, read a file, write or append to a file, pull file details, and search by filename. On the terminal side: run a shell command in a working directory you choose, with a timeout you can set. Each of those maps to one action the employee calls on the same Computer Controller tool, so there's nothing extra to enable per capability once the companion is connected.
Multi-step work, like opening an app then filling in a form, gets broken into a sequence of these individual actions by the employee itself. There's no separate 'run a macro' mode. Every click and keystroke is deliberate and logged, which is also why a long, unattended desktop job can be slower than the equivalent browser task.
File access is scoped to your home directory and enforced at the companion, not just suggested to the employee: reads are capped at 10 MB, writes at 10 MB, and a directory listing returns at most 200 entries. Terminal commands run as your own user, never with sudo, and time out after 30 seconds by default (you can raise that up to 120 seconds for a slow build or install). A hard-coded blocklist rejects destructive patterns before they ever run: rm -rf against your root, home, or a wildcard path, disk-format commands, writing directly to a raw device, and shutdown or reboot.
The employee is also instructed never to type a password or interact with a security dialog on your behalf, and never to read SSH keys, credentials, or .env files unless you've explicitly asked for that file by name. None of that is optional per task: it's baked into the tool's own rules, so it applies the same way whether you asked for a quick file rename or a multi-step app workflow.
First, pair the Desktop Companion app to your account (it's a separate free download, distinct from the platform itself). Then grant it Screen Recording and Accessibility in System Settings > Privacy & Security on macOS. The app has to be quit and reopened once after that first grant before it takes effect, and macOS may prompt once more the first time the employee actually clicks or types. If a call fails, the tool is told not to retry blindly. It reports what happened and asks whether a permission prompt is sitting on your screen or the companion has gone offline, rather than repeating a doomed action.
If the companion isn't connected at all, the action fails immediately with a message asking you to open the app. If Accessibility permission specifically is missing, a click or keypress hangs for 5 seconds and then returns a clear explanation instead of hanging forever.
The companion app gives the employee real mouse and keyboard on your desktop.
Install the lightweight companion app and pair it with your workforce. When you delegate desktop work, the employee operates your applications with real input: moving the cursor, clicking, typing, and navigating menus on screen.
Combined with screen vision, the employee runs the same observe-and-act loop a person does: look at the interface, decide the next step, act, and check the result. Destructive actions stay blocked at the companion level, and you can watch or interrupt at any moment.
The employee opens Figma, navigates layers and frames by clicking what it sees in the screenshot, and renames or reorganizes components the same way you would with a mouse, since Figma has no page structure a browser tool can read.
For a native spreadsheet app rather than a web-based one, the employee clicks into cells, types formulas and values, and scrolls through sheets by seeing and acting on the actual window.
It runs git status, npm install, or a build command in a folder you specify, reads the output, and reports back, with commands timing out automatically if something hangs.
Ask it to find every .env.example in a project or open a specific config under your home directory, and it lists, reads, and edits the file directly instead of asking you to paste the contents into chat.
| Before | After |
|---|---|
| Automation ends where APIs end. | Any clickable app is automatable. |
| RPA scripts break when the UI shifts. | The employee sees and adapts to the interface. |
| Desktop work cannot be delegated to AI. | Describe the outcome and the employee drives the apps. |
| Integrations exist only for popular tools. | The interface itself is the integration. |
Destructive patterns (rm -rf /, sudo, dd, reboot, fork bombs) are hard-blocked at the companion app before they can run. File access is bounded to your home directory, reads and writes are capped at 10 MB, and terminal commands time out after 30 seconds by default (up to 120 seconds if you raise it for one task).
Yes, the free Desktop Companion app, paired once to your account, plus Screen Recording and Accessibility permission granted to it on first use. Computer Controller is off by default and you enable it per employee once the companion is connected.
Anything in software without an API: design tools like Photoshop, spreadsheets and native office apps, video editors like Final Cut, internal desktop software, or a task that genuinely needs to see and click what's on your screen. The employee's browser connection reads the page's structure directly and is used instead for anything that lives inside a website.
A screenshot captures every connected monitor stitched into one image, with the pixel position of each display included, only when a task actually calls for it, the same way it would reach for any other tool.
Founder plan or above. Computer Controller, file access, terminal commands, and screen vision are one connection and one plan gate, not four separate unlocks.
No. The employee is instructed to never type a password or interact with a security dialog on your behalf, and to never open SSH keys, credentials, or .env files unless you've named that exact file yourself.
Let Your Employee Use Any Desktop App is part of Where their work lives.
Your AI agents manage their own workspace. Scheduled tasks run daily, weekly, or on custom cron cadences. A built-in kanban board tracks what is in progress. Every document lands in a personal Drive. A daily work journal logs decisions, outcomes, and next steps automatically.