What CutGPT downloads
The web app downloads a speech model and the files that run it. Your video and audio are never sent.
Updated
To transcribe on your computer, the web app needs a speech model and the code that runs it. Your browser downloads both as normal web requests.
What your browser downloads
- The speech model, from Hugging Face. It's OpenAI's open-source Whisper model: base.en by default, or tiny if you choose Fast in Settings. It's about 200 MB on the GPU or about 80 MB on the CPU, and the Fast model is smaller. Your browser downloads it once and caches it.
- ONNX Runtime files, from the jsDelivr CDN. With Transformers.js, they run the model in your browser: on the GPU through WebGPU when it's available, otherwise on the CPU.
What's never sent
No video or audio is sent. Like any web request, these downloads let Hugging Face and jsDelivr see your IP address and browser.
Other things to know
- There are no advertising trackers on the site. Visits are counted with Vercel Web Analytics, which uses no cookies and can't identify you.
- Vercel, which hosts the site, keeps standard request logs.
Read more in our Privacy policy.
Still need help?
Tell us what happened and which browser or iPhone you use.
Or email us at support@example.com