dsh-computer-use

Most computer-use plugins target macOS or Windows. If your agent lives on a Linux box with a real X11 graphical session, finding a lightweight option without Node native modules, PowerShell, or an accessibility framework has been surprisingly hard. dsh-computer-use fills that gap: it drives the real pointer and keyboard through Python ctypes calling libX11 and libXtst (the XTest extension), giving your agent full desktop control with zero heavy dependencies.

The plugin ships nine tools — full or region screenshots with scale metadata for accurate coordinate mapping, mouse move and click, drag, wheel scroll, text typing, key combos (ctrl+s, alt+F4), cursor position readout, and screen geometry lookup. Every input action reports the native pixel coordinates the OS actually registered, and can optionally attach a fresh post-action screenshot so the model can verify the result immediately.

If your agent runs on a Linux machine with DISPLAY set and you want to let the model see and operate the real desktop without pulling in a pile of build-time dependencies or system-level frameworks, this is the lightest option available — all you need is python3, libX11, libXtst, and ImageMagick.