The guy in the video says "We use OpenAI Step API to extract the locations of the detections".

I am unaware of that API and could not find it.