Official implementation of CVPR'26 paper A Multi-Agent Perception-Action Alliance for Efficient Long Video Reasoning - View it on GitHub