Turn your raw footage into a rough cut automatically
The first pass is the long one: before any story takes shape, someone has to clear out the ums, the silence, and the false starts. AgenticCutter does that pass for you and creates a rough cut automatically.
Why does the rough cut take longer than the shoot?
Because the first pass is almost all cleanup: ums out, pauses trimmed, false starts hunted down. None of it is creative work, but all of it has to happen before the creative work can start. So you scrub through the whole recording, ear on every sentence, hand on the razor, and by the time the timeline is finally clean, the energy you wanted for the actual edit is gone. The rough cut is the least loved part of editing precisely because it is pure labor without a single decision that needed you.
How does AgenticCutter build your first cut?
From the text, not the waveform. AgenticCutter transcribes your raw footage locally with mlx-whisper and proposes a complete cut list on the transcript: which words go, which pauses close, which spots need your attention. You read the proposal the way you would read a document, adjust it in plain language via Claude Code or Codex, or work through it in the native macOS app with its review UI. Only after you approve does ffmpeg execute the cut and render locally. The result is a rough cut you can refine instead of a raw file you still have to fight through.
What actually happens in the first pass?
Three kinds of cleanup. Filler words and ums are cut, so nobody hears you thinking. Gaps and overlong pauses are trimmed to remove silence without flattening your rhythm: visual pause preservation keeps the pauses where something happens on screen. And loud noises, a door slam, a cough, a phone, are flagged by the audio noise guard so you can confirm each one instead of discovering it in the published video.
How hard does this first pass cut?
As hard as you set it. Cutting sharpness comes in four levels, from gentle to strict, and content profiles for talking head, screencast, and show apply sensible defaults for each format automatically. A screencast keeps its demo pauses, a talking head gets tightened harder. Whatever the profile proposes, the decision stays yours: nothing renders before your approval.
Where does your raw footage live during all this?
On your Mac, the whole time. mlx-whisper transcribes on-device, ffmpeg cuts locally, and the render lands on your disk as mp4 or ProRes. Your video is never uploaded. If you pick a cloud cut brain, single still frames and transcript text go there. Fully local works with LM Studio.
What doesn't the rough cut do yet?
Retake and stumble detection is in progress: recognizing that you said a sentence twice and keeping the better take is not done yet. Video transitions are coming; audio crossfades at cut points already run via the CLI, so cuts sound smooth even before the picture gets its polish.
Frequently asked questions about the automatic rough cut
Do I have to approve every single cut?
In the app, yes: the approve gate is deliberately mandatory. You can confirm whole blocks at once instead of clicking through each cut. Full-auto with a safety net via the CLI is in progress.
Does it work for screencasts?
Yes. Content profiles cover talking head, screencast, and show, and visual pause preservation keeps the pauses where something happens on screen, so a demo does not get cut mid-click.
What about door slams, coughs, or a ringing phone?
The audio noise guard flags them in the material. Door slams, phones, coughs: nothing gets removed without your confirmation, so an important sentence never disappears with the noise.
How do I find a specific moment in the raw footage?
With on-device visual search: describe it as “find the part where …” and the cutter searches your footage on your Mac, without uploading a single frame for the search.
How do I get the rough cut out?
You export from the native macOS app as mp4 or ProRes: the rough cut renders locally with ffmpeg and lands on your disk, ready for your fine cut.
Related features
- Cut gaps and pauses
- Adjustable cut aggressiveness
- Content profiles: talking head, screencast, show
Related use case: Edit interviews