Hey HN! We’ve been working on this project for almost 3 years now, and recently decided to finally make it open source.
It’s a local-first video editor that runs in the browser and is mostly built from scratch. We’re using WebCodecs for all the encoding and decoding work, which lets us get performance pretty close to native editors.
Since day one the goal of the project was to solve LLM-based video editing, but the models have only recently become capable enough to make it genuinely useful. Instead of trying to force our own chat interface onto users, we built a CLI that is designed to work with existing coding agents like Claude Code.
The idea is that agents can operate the editor directly, while you can still open the same project in a visual interface to review their work and change things yourself.
A few things that set it apart from other NLEs: - The editor is an infinite canvas (kind of like Figma, but for video editing) - You can build compositions as SolidJS modules and mount them through the CLI - Editing and rendering happen locally. The editor can render both HTML and GPU native entities - Everything that runs on your device is completely free, including 4K/60 fps exports, unlimited projects, and no footage duration limit - Your footage stays on your device unless you explicitly use one of our analysis services. For transcription, we only upload the audio track
So when it’s open source, how do we intend to finance future development?
Coding agents don’t include every backend service they need to understand footage and make accurate editing decisions. We provide optional services like transcription, multimodal understanding, and object tracking, which we bill with a small margin.
What can agents currently do with the editor? - A lot of common post-production work: social media edits, captions, color grading, zoom effects, and animations - We recently added support for the html-in-canvas API, which turned out to be really useful for creating serious motion graphics - Video and audio understanding. For example, you can ask an agent to find certain scenes, summarize a YouTube video, or identify the relevant parts of a longer recording.
There’s still quite a lot we want to build. A few of the bigger things on our backlog: - Agentic edits based on beats and other audio characteristics - Native 3D capabilities - A full WebGPU renderer - A native color-grading UI
Would love to get some feedback & happy to answer any questions!
Comments URL: [https://news.ycombinator.com/item?id=49257843](https://news.ycombinator.com/item?id=49257843)
Points: 1