Don't watch the video — have it summarized.

Video Debrief turns any video — webinar, ad spot, tutorial, your own recording — into a summary plus a searchable timeline built from transcript and visual description, in minutes. You skip the watching and still keep everything: every number, every on-screen text, every cut.

The best part: everything runs locally. No cloud upload, no forced API, no third-party servers. A local vision model describes the frames, Whisper transcribes the audio, your agent writes the summary. Your material stays on your machine.

Video Debrief is a skill for Claude Code (it also runs in Cursor and other agentic systems) — and the Python pipeline works just as well standalone in the terminal.

What it does (4 local stages)

1. Download / Locate — yt-dlp pulls the video (URL) or takes your local file.
2. Scene Detection — ffmpeg finds every hard cut plus heartbeat frames across long static passages.
3. Transcript — auto-captions, whisper.cpp or faster-whisper (your choice).
4. Vision — a local vision model describes every scene frame in 5 sentences, OCR included.