Give agents practiced, local-first video perception: capability discovery, coarse-to-fine evidence gathering, and decentralized provider and knowledge-pack authoring.
