Analyse a dashcam clip with an LLM in seconds

Steps

1. Install peepshow

   npm install -g peepshow

2. Run with motion-only preset

   peepshow ./dashcam.mp4 --strategy scene --max 30 --dedup perceptual

3. Push to archive + alert chain

   peepshow ./dashcam.mp4 --sink s3 --sink slack

4. Hand the highlight frames to an LLM

   peepshow ./dashcam.mp4 --emit json | claude-cli -p 'Was there an incident?'

Why it works

Dashcam reels are like CCTV — long stretches of nothing punctuated by sudden events. Scene detect spots the entropy spikes; perceptual dedup drops the duplicate frames of an empty road. The LLM gets the 30 frames that actually matter.

When it helps

• Insurance / accident reporting workflows.
• Fleet ops reviewing incidents across many vehicles.
• Personal driving logs — keep only what matters, archive everything to S3.

Pitfalls

• Night footage has lower scene-detect signal — raise sensitivity with `--scene-threshold 0.2`.
• Audio rarely matters for dashcam; `--no-transcript` saves time.

Works with these LLMs

Pairs with these sinks