Open-source tool
Video loop.
Make a video with your agent.
Turn a voice recording and a folder of clips into a short, captioned video. Your agent does the editing. You review the words, the pictures and the finished cut.
Runs on your Mac with a coding agent. Free code, under the MIT licence.
A few videos I’ve made
Made for my emails, with the workflow this tool grew from.
Watch: Building for humans
Watch: Building a calm business
Watch: Building boring software
01 / The workflow
An edit you can keep changing.
I built this for my own videos. The useful part is being able to say: keep that line, use this shot, give the ending a little more room.
Bring your material.
A recorded voice-over or a video of you speaking, plus the clips you want to show. Tell your agent what the video is for.
Start with the words.
The agent proposes a cut of the narration. Read it, change it and check that it tells the story you want.
Then choose the pictures.
Open an HTML storyboard showing which clip goes with each part of the narration. Swap shots, adjust the pacing and choose your captions. Approve the plan before the first cut is rendered.
Watch. Change. Export.
Watch the first cut. Give feedback in your own words. Repeat until you’re happy, then export the finished video. You decide when and where to post it.
02 / A fictional example
A small repair, in three shots.
Imagine you’ve recorded a voice note about fixing a lamp, and filmed three short clips on your phone.
I nearly threw
this lamp away.
The lamp on the desk.
The switch
was the problem.
A close-up of the switch.
Ten minutes later, it was
back on my desk.
The lamp switched on.
“Hold the last shot longer. Keep the click of the switch.”
The agent revises the cut. You watch it again.
03 / Before you start
What you’ll need.
- A Mac and a coding agent.
- Use Claude Code or another agent that can run commands and inspect images. You work with it on your own computer.
- The video tools.
- FFmpeg with caption support and Node.js. The optional typed opening needs a browser. The repo’s setup check tells you what your project needs.
- API access, when needed.
- An ElevenLabs account and key with access to the features you choose: transcription, audio isolation or music generation. Generated music needs the Music Generation key permission. The included example runs without a key. Agent and API costs depend on your services and usage.
- Your recordings.
- Start with a voice recording and a few clips you have permission to use. A small first project makes it easier to learn the loop.
Rendering happens on your Mac. Your coding agent may use a cloud service. Transcription and voice cleanup send your selected recording to ElevenLabs; generated music sends your text brief and requested duration. Check the README before using sensitive footage.
04 / Try it yourself
Start with the example.
Open the repo in your coding agent. Ask it to follow the README, run the setup check and take you through the included example. Then start a new project with your own footage.
The README holds the setup instructions and the prompts, so there’s one place to keep them up to date.
Get it on GitHubShared as a working tool. I can’t promise support or regular updates.