AI Lessons
Computer Use

Canine

Computer Use

Letting Claude drive an agent computer

How Canine lets Claude see and operate its Omarchy agent computers: a Python server in each VM that speaks Anthropic's computer-use actions and reads the accessibility tree, the Wayland input and coordinate details behind it, and the MCP tools that connect Claude Code to it. Nine chapters and five labs run against a real machine.

15 chapters · 61 min read · updated 2026-09-28

Start reading

Chapters

  1. 1What computer use is, and what we built5 min
  2. 2Seeing the screen5 min
  3. 3Pressing keys and clicking on Wayland6 min
  4. 4The accessibility tree6 min
  5. 5The server as a Python package6 min
  6. 6The Canine side: from Claude to the VM5 min
  7. 7War stories5 min
  8. 8Current state and open work2 min
  9. 9Glossary and cheat sheet3 min

Labs

  1. 10Labs: setup2 min
  2. 11Lab 1: The raw API3 min
  3. 12Lab 2: Drive the desktop3 min
  4. 13Lab 3: The accessibility tree4 min
  5. 14Lab 4: uinput from scratch3 min
  6. 15Lab 5: Through Canine's MCP tools3 min

Reading settings

Theme
Text size
100%
Your progress

Progress is stored in this browser only. Copy it to move to another device.