Local Python engine using Vision LLMs to automate legacy desktop software via screen perception. Zero seat fees.

What’s included:

• Multimodal visual coordinate grounding converting visual display buffers into precise hardware actions
• Self-healing execution cycle with automatic visual state verification and fallback recovery routines
• Completely selector-free architecture operating natively across Windows, macOS, and Linux X11/Wayland
• Local audit trail with structured JSONL action logging, coordinate heatmaps, and step-level screenshots