Screen perception for desktop cowork agents must process full 1080p and 4K display buffers without dropping frames. Neo v0.1 splits perception across dual Tesla T4 GPUs:
- GPU 0 (Perception & OCR): Extracts UI bounding boxes and text anchors using parallelized visual embeddings.
- GPU 1 (Action Reasoner): Translates bounding box coordinates into deterministic mouse and keyboard action vectors.