Piranha

You'd reach for this when you need to break down Brazilian Jiu-Jitsu footage frame by frame. It runs YOLO pose detection on your videos to track body positions, then feeds that skeletal data to Claude for technique analysis. The setup uses streamable HTTP transport, so you can point it at video URLs and get back position tracking plus AI commentary on guard passes, submissions, or whatever grappling sequence you're studying. Useful if you're coaching, reviewing comp footage, or just trying to figure out why that sweep keeps failing. The server sits at pi.phos.nz and handles the computer vision pipeline so Claude can focus on the actual technique critique.