Twelve Labs launched Pegasus 1.6 for understanding first-person video. It structures hand-tool interactions, action sequences and outcomes by segment, offering five capabilities: action segmentation and labeling, detailed captions, quality assessment, search and selection, and consent- and privacy-related detection. The aim is to reduce manual preparation of robot-training data, not to demonstrate robot-control performance itself. In actual use, the accuracy of automatically selected scenes and labels must be validated separately.