Edit interviews from the transcript instead of scrolling waveforms
Interviews and podcast video have the worst ratio in editing: long recordings, short results, and every usable moment buried somewhere in between. AgenticCutter lets you edit interviews where they are easiest to read: in the transcript.
Why is interview editing such a grind?
Because an hour of conversation ends up as ten minutes, and you still have to listen through the whole hour once. Scrolling waveforms tells you nothing about what was said, so you scrub, listen, jump back, take notes, and slowly build a map of the conversation in your head. Meanwhile the actual editing question, which passages carry the story, has not even come up yet. For a podcast video or a customer interview, the finding costs more than the cutting.
How do you edit an interview with AgenticCutter?
You read instead of scrub. AgenticCutter transcribes the interview locally with mlx-whisper, and the transcript becomes your editing surface: you read the conversation, mark what carries the story, and say in plain language what stays and what goes. “Keep this answer, cut the small talk, tighten the middle”: the cutter turns that into a cut list and executes it once you approve. You direct via Claude Code or Codex, or work in the native macOS app with review UI and approve gate. Reading an hour of transcript is fast; listening to an hour of audio is not, and that difference is the whole point.
What does the cutter clean up automatically?
The mechanical part of the edit. Filler words and ums go, dead air between question and answer closes, overlong pauses get trimmed so the conversation keeps its pace without sounding rushed. Loud noises, a cough, a slammed door, a phone, are flagged by the audio noise guard and removed only with your confirmation, because in an interview the noise sometimes sits right on top of the best sentence.
How do you find the moments you want to keep?
Two ways. The transcript itself is searchable text: names, topics, and quotes are a text search away instead of a scrubbing session. And for everything the transcript cannot capture, on-device visual search takes a plain request like “find the part where she holds up the prototype” and searches your footage directly on your Mac.
What happens to confidential conversations?
Your recording is never uploaded: transcription, cutting, and rendering run on your Mac. If you pick a cloud cut brain, single still frames and transcript text go there, never the footage. If even the transcript text must stay on your machine, attach a local model via LM Studio and the whole pipeline runs fully local.
Does the cutter know who is speaking?
No. Speaker separation is not in the feature set, and neither is multicam: the cutter works on one recording and its transcript. Retake detection is in progress. What you get today is an honest transcript edit, not an automatic multi-camera production.
Frequently asked questions about interview editing
Can I pull Shorts clips straight from the interview?
9:16 export runs via the CLI, so vertical clips come out of the same cut. Automatic subject tracking is planned, which means today you set the frame yourself.
Do I get subtitles for the finished interview?
Yes, as SRT or VTT from the same transcript the edit was built on. The glossary keeps names stable, and optional translation preserves the timing.
How do I make sure an answer stays intact?
Say “keep the take” in plain language, or run the gentle sharpness level over that passage. And nothing renders before your approval, so a cut you dislike never reaches the finished video.
How do I handle two cameras?
Multicam is not in the feature set. The cutter works on the transcript of one recording, and you place the camera switches yourself. That is the honest answer today.
Will my guests' names be spelled correctly?
Yes, through the brand and name glossary: set a spelling once and it applies to every transcript and subtitle track, in every conversation after that.
Related features
- Transcript-driven editing
- Audio noise guard
- On-device visual search
Related use case: Create a rough cut