Press kit
Everything here may be quoted. The name and the icon may be used to write about OpenCaptions, as the trademark policy says.
In one line
Section titled “In one line”OpenCaptions adds animated, word-by-word captions to videos: on your iPhone or Mac with nothing uploaded, from Claude and other agents, or on your own server with one Docker file.
The facts
Section titled “The facts”| Maker | Léo Gaudin |
| Licence | AGPL-3.0 |
| First release | 3 November 2026 on the App Store, version 1.0.0 of the self-hosted stack on 4 November 2026, the Mac on 9 November 2026 |
| iPhone and iPad | App Store, iOS 18 or later |
| Mac | Mac App Store, macOS 15 or later on Apple silicon, the same purchase |
| Price | The app is free, up to 1080p at 30 fps with a small mark. Pro is one purchase with no subscription, 14,99 € at launch, and adds 4K, 60 fps, HDR and every look. Self-hosting is free. |
| Agents and scripts | The opencaptions npm package: an MCP server for Claude Code, Claude Desktop, Cursor and other clients, and a command line. It runs on the computer with no server |
| Try it | opencaptions.app, in the browser, with nothing to install |
| Source | github.com/leogaudin/opencaptions |
| Interface | 13 languages |
| Speech models | Whisper (about 100 languages), Parakeet (25 European languages), Apple Speech on iOS 26 |
What is different about it
Section titled “What is different about it”- All of it is open source: the app, the web editor, the server and the engine that draws the captions.
- One engine draws everything. A Rust engine runs in the browser as WebAssembly, in the app, on the render server and in the npm package, so the preview and the export are the same code.
- Nothing leaves the phone. The app transcribes on the device, has no account, no analytics and no tracking. See the privacy policy.
- You can run it yourself, on a server or a Raspberry Pi, with amd64 and arm64 images.
What it does not have
Section titled “What it does not have”No Android app. No emoji, B-roll or virality scoring. Fewer templates than hosted tools. The first transcription on a phone is slow, while the model is prepared for the device.
Images
Section titled “Images”- The icon
- The web editor, on Sprite Fright by Blender Studio, CC BY 4.0
