Production-grade Python desktop automation engine driving legacy GUIs using Vision LLMs without APIs.

What’s included:

• Vision-driven element localization supporting Claude 3.5 Computer Use, GPT-4o, and local VLM fallbacks
• Self-healing visual retry loop with state verification before and after every dispatched OS event
• Native OS event driver bypassing hardware emulation blocks with human-like bezier curves and jitter
• Configurable headless virtual display and multi-monitor offset handling for Windows and Linux environments