REFERENCE

Computer use on Linux.

Which coding agents can use a desktop’s screen, mouse and keyboard on Linux by themselves, taken from each project’s own documentation and checked on 28 September 2026. And how ibara gives computer use to any of them on Omarchy.

Updated

Built-in computer use, by agent#

“Desktop computer use” means the agent itself can see the screen and work the mouse and keyboard in any app. Browser control is narrower: the agent drives a web browser only. Every agent below can also use MCP servers.

AgentDesktop computer useOn LinuxBrowser controlSource
Claude CodeA research preview on macOS in the CLI; macOS and Windows in the desktop appNoClaude in Chrome, which works on LinuxCLI, Linux app, Chrome
CodexComputer Use in the ChatGPT desktop app, on macOS and WindowsNo: “not yet in the Linux preview”In the desktop app; none in the CLIComputer Use, Linux app
Gemini CLINoNoAn optional browser agent, off by defaultSubagents
opencodeNoNoNo built-in toolTools
omp (oh-my-pi)A computer tool for the host desktopYes. Wayland screen capture is off by default in its Nix packageYesRepository
GooseThe computer controller, on macOS onlyNo: its UI automation is for macOS onlyNo built-in toolComputer controller
Hermes AgentYes, through Cua’s driverYes. Pure Wayland needs an XWayland bridge for screen captureYesComputer use
OpenClawBuilt in on macOSThrough an experimental plugin, tested on X11YesComputer use

Omarchy and Wayland#

Omarchy runs Hyprland, a Wayland compositor. Several of the Linux options above are built for X11 or need extra setup on Wayland: omp’s Nix package leaves Wayland screen capture off by default, Hermes Agent needs an XWayland bridge on pure Wayland, and OpenClaw’s Linux plugin is tested on X11. ibara is built for Omarchy and Hyprland: it uses Cua’s Hyprland input plugin, adapted for Omarchy, and every click and keystroke goes through Cua.

One screen, one cursor#

Built-in computer use drives the computer you’re sitting at. On Hyprland there is one seat, one cursor and one keyboard, so an agent’s click or keypress takes your focus, as an Omarchy discussion describes. ibara puts the agent on a separate Omarchy computer you own instead. Your agent works there. You keep working here.

Add computer use with ibara#

ibara is an MCP server, so any agent in the table that can add an MCP server can use it. Claude Code, Codex, opencode and omp have been tested. In the ibara console, choose Connect an Agent in the fleet’s toolbar, then Copy Prompt, and paste it to your agent; it adds ibara to its own settings and tells you which computers it can use.