Turn Any Video Into a Queryable Brief

Video is the worst format to retrieve from - you cannot skim it, and the one line you need is buried somewhere in an hour. So you watch once, forget most of it, and never go back. The fix is to get the words out of the video and into a form you can search and question.

You do not need a paid service. yt-dlp pulls the audio; Whisper (OpenAI's open transcription model, which runs locally) turns it into text. Both are free and run on your machine, so long or private videos cost nothing and stay yours.

bash# 1. grab the audio
yt-dlp -x --audio-format mp3 -o audio.mp3 "VIDEO_URL"

# 2. transcribe it locally
whisper audio.mp3 --model small --output_format txt

NotebookLM answers only from the sources you give it and cites back to them, so it does not wander off into made-up facts. Drop the transcript in - alongside any related docs - and it becomes a grounded knowledge base for that material.

• Ask direct questions - 'what did they say about pricing?' - and get an answer with a citation to the exact spot.
• Ask for a brief - the key points, the steps, the claims - so an hour of video becomes a page you actually keep.
• Load several transcripts at once and question them together, so a playlist becomes one searchable body of knowledge.

Because every answer is grounded in your transcript and cited, you can trust it and jump straight to the source when you want the full context. That is the whole win: the knowledge stops rotting and starts answering.