Computer use
Current status: implemented in the source, but not exposed to regular users in the current build.
Computer use means observing and operating a real desktop window or application. It is not a catch-all name for the built-in browser, plugins, or image input. The latest Desktop source retains real desktop control, platform-readiness checks, action confirmation, stop control, and an MCP bridge, but the current build disables the capability by default and provides no normal user entry to enable it.
The short answer
- The implementation exists: the source contains local-desktop and isolated-environment execution paths, plus Windows and macOS readiness checks.
- It is not currently exposed:
computerUsedefaults to off, no user-facing settings row offers it, and the Computer Use MCP server is removed from the regular plugin list. - The browser remains available: the built-in browser can operate pages inside its own pane; that is not control of other windows on the computer.
- Attach an app remains available: users can explicitly select one window for EvoX to view or read. The window-confined Operate tier appears only when the current connection supports it.
- Plugins remain available: a plugin can call only the tools it declares and is authorised to use. Installing one does not grant desktop-control access.
How it differs from other capabilities
| Capability | Current status | Scope |
|---|---|---|
| Built-in browser | Available | Open and operate web pages inside the EvoX browser pane. |
| Attach an app | View and Read available; Operate depends on the connection | Work only with one window explicitly selected by the user; recapture it at send time and refuse actions outside it. |
| Plugins and MCP | Depends on connection and permissions | Call the external-service tools explicitly exposed by an integration. |
| Image input | Available | Analyse images that the user actively adds to the conversation. |
| Computer use | Implemented in source, not exposed | Intended to observe and operate a real desktop window or application. |
Browser-task prompts do not belong on this page. The browser page already explains how to open, inspect, and authorise pages, so this article does not repeat that workflow or present browser actions as real desktop control.
What the source confirms
These facts are confirmed in the current source, but they do not mean regular users can use the capability today:
- A local-desktop execution path and an isolated-environment execution path exist.
- Windows and macOS readiness checks exist; macOS also checks Screen Recording and Accessibility permissions.
- First-use acknowledgement, stop handling, and MCP request-bridge logic exist.
- A browser automation path exists, but the source explicitly treats browser automation separately from real desktop control.
What must not be documented as available yet
Until the capability is reopened, the documentation should not instruct users to look for or promise:
- Reading or controlling desktop application windows that the user did not explicitly select;
- Moving the real mouse, sending cross-window keyboard input, or operating freely across the desktop;
- Completing tasks through a picture-in-picture view, virtual pointer, or “using your computer” indicator;
- Installing an ordinary plugin to gain control of the whole computer.
These may be future product directions, but they should not currently be presented as an EvoX entry point, procedure, or product screenshot.
Related pages
- For web pages, read Browser.
- To provide one open window as bounded context, read Attach an app.
- For connected external services, read Plugins.
- For screenshots and design files, read Image input.
EvoX Docs · Features · Capabilities