Qt 6 · C++17 · Windows & Linux

Let your AI use your computer.

M-Cowork is a desktop app that lets people use their AI like OpenCode. It sees the screen, moves the mouse and types on the keyboard.

scroll to run the agent ↓

01 / 04

screenshot {"format":"png"}

It sees your screen.

One call captures the desktop and hands it back to the model as an image. The agent always knows what you would see.

02 / 04

mouse_move {"x":1104,"y":412}

It moves the mouse.

The model reads the screenshot, picks coordinates and the tool moves the real cursor there. Press Esc at any moment to abort.

03 / 04

mouse_click + keyboard_type

It clicks. It types.

Clicks and keystrokes travel through the same JSON-lines protocol over standard I/O, so the UI and the engine stay cleanly separated.

04 / 04

wait → end_session

It waits. Then it stops.

The agent lets the page settle, finishes the job and ends the session. An idle timeout is the safety net if anything hangs.

MODELS

Bring your own model.

Pick a provider in the settings panel, add your key, model and a custom base URL. Everything is stored locally.

GeminiOpenAIClaudeGroqOllama

BUILD

Build it.

Windows (MSYS2 UCRT64)

.\scripts\install-packages.ps1
.\scripts\build.ps1
.\scripts\run.ps1

Linux

./scripts/build.sh
./scripts/run.sh

STATUS

Where it stands.

  • Qt 6 UI, frameless window, native behavior
  • Computer-control tool v0.4.0 + client wrapper
  • AI provider settings panel
  • Agent loop with tool calls and screenshots
  • Chat history, streaming, markdown
  • Voice input and permission prompt
  • Installer / packaged release