v0.6.0 — First Public Release

copilot-computer-use v0.6.0

The world's first free Computer Use agent powered by GitHub Copilot API.

Highlights

• 🆓 Zero cost — GPT-4o is 0x multiplier on Copilot (completely free, no premium quota consumed)
• 👁️ Vision via Copilot — Base64 screenshots analyzed through copilot-vision-request: true header
• 🔄 SSE Streaming — Real-time token-by-token AI response display
• 🖥️ Cross-platform — macOS + Linux executors (pyautogui + optional xdotool)
• 🛡️ Robust — 429 retry with exponential backoff, screenshot change detection, anti-loop prompts

Features

• GitHub OAuth Device Flow authentication + JWT auto-refresh
• VS Code version spoofing (mimics Copilot Chat headers)
• 10 action types: click, double_click, right_click, type, key, scroll, move, wait, done, fail
• Coordinate grid overlay (--grid) for better AI positioning
• Screenshot saving (--save-screenshots DIR) for debugging
• Task planner for breaking complex tasks into steps
• Model cost awareness (shows FREE/premium at startup)
• 40 unit tests
• English + Chinese README

Quick Start

git clone https://github.com/Zey413/copilot-computer-use.git
cd copilot-computer-use
pip install -e .
python -m src.copilot.auth          # One-time GitHub auth
python -m src.main "Open Chrome and search for weather"

Requirements

• Python 3.9+
• GitHub Copilot subscription ($10/mo or free for students)
• macOS or Linux

Research Background

Based on reverse engineering of Raven (Copilot API proxy) and architecture from self-operating-computer.