Transcribe a video locally — no API key, no cloud

Steps

1. Install peepshow + whisper.cpp

   Both are stand-alone CLIs.

   npm install -g peepshow
   brew install whisper-cpp

2. Run peepshow on the video

   Audio extraction + transcription happen automatically when whisper.cpp is found on PATH.

   peepshow ./meeting.mp4

3. Read the transcript

   The JSON manifest contains `transcript.text` and per-segment timings.

   peepshow ./meeting.mp4 --emit json | jq '.transcript.text'

4. Pick a model size

   Default = `base`. Use `large-v3` for accuracy or `tiny.en` for speed.

   PEEPSHOW_WHISPER_MODEL=large-v3 peepshow ./meeting.mp4

Why it works

Whisper-on-cloud (OpenAI Whisper API, AssemblyAI, Deepgram) sends every byte of audio to a third party. For meetings, medical content, legal interviews — that's a privacy non-starter. peepshow + whisper.cpp runs the same model class entirely offline. The cloud providers stay one flag away if you want them.

When it helps

• Meetings, interviews, depositions where audio shouldn't leave the box.
• Air-gapped workflows (research labs, regulated industries).
• High-volume transcription where API costs add up fast.

Pitfalls

• `whisper.cpp` doesn't ship via npm — install via brew / apt / source.
• Larger models (`large-v3`) need 4-8GB RAM. `base` works on a laptop.
• Disable telemetry with `PEEPSHOW_TELEMETRY=0` for a fully offline pipeline.

Works with these LLMs

Pairs with these sinks