Most video tools describe what is on screen. We care about the production decision behind it — filmed or generated, which light, which caption animation, which sound cue, which code layer. That difference is what makes a breakdown useful for remaking rather than just summarising.
The analysis core is inspired by the open-source rea-video project: local signal analysis for shot boundaries, OCR, frame rate, sensor noise and metadata, with an optional multimodal model for the semantic layer. It is a command-line engine, so bringing it to the web means running real compute somewhere else — that is the part we are building now.