For experts, media buyers and small businesses who shoot their own Reels
The editor runs on your Mac: it transcribes speech, trims pauses and adds captions with word highlighting. An AI agent in chat picks the takes, the hook and the B-roll — and you can fix anything by chatting or by hand on the timeline.
Sound familiar?
You recorded a tip on your phone. There are pauses, some “uhm”, and the opening line twice. Then comes the editor: cut the pieces, type the captions, sync every word to the voice, find a cutaway for “keeps scrolling”.
One Reel a week is fine. When you need several a week, editing takes longer than shooting.
So I built an editor where software assembles the draft and an AI agent in chat makes the content calls.
How it works
The heavy lifting stays local: transcription, audio, frames, rendering. The AI agent reads the transcript and contact sheets and decides what stays and what goes. Every decision lands on the timeline, where you can adjust it.
With the “Media” button in the editor, or just drop the file into the input folder.
Whisper small transcribes speech, the app trims long pauses, levels the voice and lays out captions. Local and free.
You type “Edit this reel” in chat. The agent fixes words, picks the best takes, removes filler, drafts 5 hooks, places B-roll and writes the post caption.
Contact sheets: the first 15 s at 4 frames per second plus 12 frames across the clip. Technical errors block export.
MP4 1080×1920, a clean cut or transparent ProRes overlays. Next to it — a folder with the project, SRT, transcript and post caption.
The draft took ≈18 s and the final MP4 export ≈2 min (on an Apple M5, 55-second clip).
The output
This clip was edited in ReelsMaker end to end: 54.9 s of raw footage became 40.8 s. The hook sits on the first 3.6 s, captions run 3–5 words with the spoken word highlighted, and four B-roll inserts match what’s being said.
Auto-edit
The “Auto-edit” button runs a local Whisper small model. The app transcribes speech, trims long pauses, levels the voice and lays out captions. No cloud, no per-minute fees.
Agent editing
You type “Edit this reel”. The agent reads the whole transcript, fixes misheard words, picks the best take and removes the filler. Every cut in the plan comes with a reason.
Hook
The agent drafts up to five hooks of up to 7 words. Each one is tied to real words in the transcript, so the hook never promises what the clip doesn’t deliver. The chosen one sits on the first 3.6 s with a soft fade-in.
Captions & manual edits
Captions are grouped 3–5 words at a time, respecting sentences and cuts, and the spoken word turns bold. Every caption is its own element: click it to edit text, font, background box, shadow and timing on the right.
B-roll
Every cutaway is linked to its phrase. If a B-roll insert in the agent’s plan doesn’t line up with its words, the app rejects it. Diagrams are kept as SVG plus PNG.
Export
Before export there’s a technical check: words without captions, text outside the safe zone, overlapping titles, voice out of sync. Any error blocks the export.
Details
Untick a sentence and it disappears from the video along with its captions. Repeats are flagged: “Repeats the line from 11.8 s”.
Words the model isn’t sure about are listed separately with timecodes. The transcript is a draft, and the app doesn’t pretend otherwise.
The first 15 s at 4 frames per second and 12 frames across the whole clip. The agent reviews them before export.
A local MediaPipe model finds faces twice a second and smoothly follows them in the 9:16 frame. With several people or fast movement, check the result.
Every change is saved within half a second, and previous versions stay on disk. Manual edits survive when the agent re-applies its plan.
The agent writes the caption: the first line continues the hook, then 2–3 lines of substance and one call to action. In the author’s words, no made-up promises.
Who it’s for
Records a tip on the phone and gets an edited Reel with a hook and captions. Changes are a chat message away.
Makes videos for clients. The clean cut and transparent ProRes overlays can be finished in another editor.
The owner films the product and the process. Everything runs on one Mac, files aren’t uploaded anywhere.
Numbers and limits
FAQ
No. It builds a draft: transcript, trimmed pauses, leveled voice and captions. Takes, repeats, the hook and B-roll are chosen by the AI agent in chat.
Claude, Codex or Content OS connected to the editor. The agent drives the editor through its built-in tools, and you see the result on the timeline.
Nowhere. Transcription, audio processing, framing and rendering run on your Mac. To make its calls, the agent in chat gets the transcript text and contact sheets, not the video itself.
It’s a draft. The small model gets some words wrong — this clip had two misheard words. Uncertain words are listed separately, but a confident word can still be wrong, so the agent reads the whole text and fixes it before editing.
Yes. Text, size, position, color and timing of every element, video framing, clip edges on the timeline. Or just type the change in chat.
There’s no direct export yet. You can carry over the media, SRT captions, the clean cut and transparent ProRes 4444 overlays.
No, macOS only.
We’ll go from raw video to MP4. I’ll show what the button does, what the agent does, and what it takes to run the editor on your Mac.
Discuss on TelegramOther products