Why the script is the timeline

Most of the videos we make start life as words. A script, a talk, a walkthrough, a rant into a webcam. Then we drag that recording into an editor and spend the evening hunting for those words again, by ear, on a waveform.

Cutgraph Studio starts from the other end. It transcribes your recording on your own machine, then hands you the transcript as the timeline. Strike a word and the picture and sound cut with it. Drag a scene and the video reorders. Drop a clip on a word and it lands there.

v0.1 is our upcoming free developer preview. The source release is about to open. This post covers what it is, why we built it the way we did, and what is still rough.

The idea: read your video, don't scrub it

Text-based editing isn't new, and we didn't invent it. What we wanted was a version that feels like editing a document, runs entirely on our own computers, and doesn't stop at cutting.

So the Story view gives you four lenses on one video:

• Script: the words. Click to seek, drag across words to cut or restore. The gap closes on its own.
• Board: the scenes as cards. Drag to reorder. A coloured strip shows each scene's job in the story: Hook, Problem, Example, Payoff, Wrap.
• Timeline: Picture, B-roll, Graphics, Music and Captions lanes, in sync with the script, for when you do want to see the shape of the thing.
• Source: the whole video as a readable script.mdx you can copy. Read-only for now.

Change it in Script, Board or Timeline and the other lenses follow. You can try the core gesture without installing anything on our homepage: the toy there strikes words, re-times a little timeline and updates the captions.

Why local

We edit a lot of footage we can't upload: client calls, unreleased product, family. Running WhisperX on your own GPU (or CPU) is fast enough now that a cloud round trip buys you very little for this job, and it removes a whole category of questions about who can see your media.

So in Cutgraph, transcription, editing and rendering happen on your machine. The app only accepts connections from your own computer. There's no account, no telemetry and no upload. A few optional things can reach the network, like the co-pilot or stock search, and we list every one of them in Your footage, your machine.

Why there's a cartoon camera in it

Talking-head videos get better with a little motion: a title card, a diagram landing on a word, a character reacting. Most of us can't animate, and we didn't want to learn a second tool for it.

Cutgraph has a drawing engine built in. Any scene can be your footage, a full-frame animation over your audio, or a character cameo that lands on a word you pick. The sample character is Lens, a small pixel-art camera with eleven expressions, nine poses and nine moves. Preview and export use the same engine, so what you see is what renders.

Why the co-pilot only proposes

There's an optional Ask bar backed by Claude. It can plan scenes, find filler words, rank B-roll for a phrase and set up an animated shot. Every answer arrives as a card with Accept and Reject, and nothing changes until you accept. An accepted batch is a single undo step.

It uses your own Anthropic API key, billed by Anthropic. We don't sit in the middle with a credit meter, and if you never add a key, nothing is sent anywhere. We think an editor should treat AI like a sharp junior editor: useful, fast, and never allowed to touch the cut without asking.

What's rough

This is a developer preview, and we'd rather you hear it from us:

• You install it from source. Python 3.12+, Node 24 and ffmpeg. A packaged installer exists as a recipe but isn't the supported path yet.
• Windows 11 with an NVIDIA GPU is what we test. Linux is best-effort. macOS isn't supported.
• Transcription is English only, and the first run downloads about 3.5 GB of models.
• It's single-user, single-machine. No sync, no collaboration, no recording studio.
• Some things exist in the pipeline without a button yet: silence trimming, karaoke captions, face-tracked vertical reframing.

The full, tagged list is on the features page, and the changelog says what landed in each version.

Fair Source, and what that means

The source preview will open under the Functional Source License (FSL-1.1-ALv2). You can read it, run it, change it and share it, including for client work. The one thing you can't do is turn it into a competing commercial product. Each version becomes Apache 2.0 two years after it's released. That's why we say "Fair Source" rather than "open source". Pricing explains it in more detail.

Try it

Join the waitlist.

The source preview is about to open. Join the waitlist to hear when it is ready, and see Get started for a preview of the install.

FAQ

Is Cutgraph Studio free? Yes. The upcoming v0.1 developer preview will be free to download and run on your own machine. A paid Cloud option and Pro packs may come later; the local editor is the product today.

Is Cutgraph open source? It is Fair Source, not open source. The source preview will open under FSL-1.1-ALv2: you can read, run, modify and redistribute it for permitted purposes, but not ship a competing commercial product. Each version becomes Apache-2.0 two years after release.

What do I need to run it? Windows 11 is the tested platform and an NVIDIA GPU helps. Linux works on a best-effort basis; macOS is not supported yet. You need Python 3.12+, Node 24 and ffmpeg.