In almost every company there is a whole library of knowledge lying around that nobody uses: the recorded training, the workshop capture, the webinar, the two-hour strategy meeting. Recorded, filed away — and then never watched again. Studies on learning videos make it clear: with recordings of around 80 to 90 minutes, fewer than 30 percent make it to the end. With short clips it is almost 80. The knowledge is there. It only rots because nobody invests two hours to get to the one relevant passage.
There is a better way, and it is surprisingly simple: you do not watch the recording — you ask it. "What was said about the budget, and at which minute?" And you get the answer with a reference to the passage, in seconds. That way every long video becomes a one-page summary for the team and a searchable question-and-answer archive. The best part: it runs entirely on your own machine — no upload, nothing leaves the building.
Ask the video instead of watching it
Behind the scenes there is little magic happening, and that is exactly the point. You pull the pure audio track out of the video and have it transcribed locally — with whisper.cpp, the offline port of OpenAI's Whisper; the fast Turbo model handles 90 minutes on a modern laptop in a few minutes. The result is a transcript with timestamps. This transcript then sits as a file next to you, and you simply put your questions to Claude Code about it. You do not need to keep any of it in your head — you ask in plain language, and the rest happens by itself.
What you get out of it:
- A one-page summary for the team — clear, without filler.
- The key decisions from a meeting, each with a timestamp.
- A task list from everything that was discussed as a to-do.
- A searchable archive: every recording becomes a knowledge source you can query.
And along the way you sidestep a data protection problem
The obvious reflex would be to quickly upload the video to YouTube for the automatic subtitles and set it to "unlisted". For internal content, that is an own goal. "Unlisted" is not data protection, just hiding: the video physically sits on the servers of a US corporation, and anyone with the link can watch, forward and embed it. On top of that comes the legal side — a recording contains voices, names, often faces, so personal data. The basis for such US transfers, the EU-US Data Privacy Framework, was challenged before the European Court of Justice once again in October 2025. Anyone working locally does not have this problem in the first place.
Unlisted is not data protection. It is hiding.
From a trick on the laptop to a system for the company
Making a single video searchable on your own machine is the quick win — and that is exactly what the recipe below is for. But the real difference in a company comes from something else: that every new recording is automatically transcribed and summarised, that all of it ends up searchable on the intranet so the whole team can ask instead of just you at the laptop, and that it is run reliably, maintained and GDPR-clean. That is the last mile: turning a trick into a system. The trick I am giving you here. The system I build.
The complete recipe for a single video — one file that you drop into Claude Code yourself, with every step and command — you can grab for free below. It is built so that the assistant guides you through every point of friction. And if you want to turn it into a system for the whole team, my number is in there too.