Primate Vision

Primate Vision gives Claude the ability to actually watch and understand video.

Point it at a file or URL and ask questions in plain language: "when does the forklift enter the aisle?", "count the people wearing hard hats", "what happens after the door opens?" Claude gets back structured, grounded results — what was detected, where in the frame, and at which timestamp — instead of a vague summary.

It runs on Primate Intelligence's own scene-understanding models, built around predictive world models rather than frame-by-frame captioning. That means it reasons about objects persisting, moving, and interacting over time, so it holds up on questions that require tracking something across a clip rather than describing a single still.

Typical uses: reviewing security and safety footage, auditing operational or manufacturing video, sports and performance breakdowns, media logging and search, and QA on anything captured on camera.

Getting started takes about a minute. Sign up, create an API key, and connect — new accounts include free processing time, and no card is required for your first successful run. Usage-based pricing after that, with a free tier that stays free.