peepshow/ how-to/ screen-recording-to-llm

Reel #H-11 ◆ QuickTime / OBS / Screen Studio → LLM ◆

peepshow how-to / screen-recording-to-llm

Send a screen recording to Claude, GPT, Gemini for instant review

Screen recordings are the highest-signal video LLMs see: an entire workflow captured in 30 seconds. peepshow extracts scene-change frames (which usually means UI state changes) so the LLM walks through exactly what the user saw.

Steps

  1. Install peepshow
    npm install -g peepshow
  2. Run on the recording

    Scene detection catches UI state changes — frames land where pixels shift.

    peepshow ./bug-repro.mov
  3. Hand to your LLM

    Drag into Claude / ChatGPT — paste the frames as a conversation turn.

    peepshow ./bug-repro.mov --emit json | claude-cli -p 'What changed between frames?'
  4. Auto-post to your bug tracker

    GitHub Issues or Linear sinks file a ticket with the frames attached.

    peepshow ./bug-repro.mov --sink github-issues

Why it works

QA teams have used screen-recordings since QuickTime existed. The cost has always been: someone has to watch them. peepshow flips the workflow — every recording becomes a structured frame timeline + transcript that an LLM can triage, file, and link to a tracker automatically. Cycle time drops from minutes to seconds.

When it helps

  • Bug reports where the user attaches a .mov.
  • Design reviews captured in Screen Studio / Loom / OBS.
  • Customer-success calls where one frame explains the issue better than the transcript.

Pitfalls

  • Scene threshold defaults to text content — for video games / continuous-motion recordings, switch to `--strategy fps`.
  • Audio narration is usually quiet — `whisper.cpp --gain 1.5` (set via `PEEPSHOW_WHISPER_GAIN`) helps.

Works with these LLMs

peepshow + Claude Claude reads images. peepshow turns a video into a timeline Claude already knows how to reason about peepshow + Gemini Native video + peepshow = control plane for long footage, animated formats, and cost ceilings. peepshow + GPT-4o / GPT-5 GPT vision is image input only. peepshow turns video + animated formats into the frame timeline GPT

Pairs with these sinks

GitHub Issues sink File an issue with the frames uploaded inline. Linear sink Open an issue with the video evidence attached. Slack sink Channel post with Block Kit formatting.

Related how-to

Loom → LLM Send Loom recordings to Claude / GPT / Gemini. peepshow extracts scene frames + transcript from the Video → Slack Post video highlights to Slack automatically. peepshow's Slack sink emits a webhook message per run

Other how-to workflows

Pairs with