Last updated: October 5, 2026
ScribeFit turns your audio and video recordings into text and subtitles entirely on your device. Your recordings are never uploaded, transmitted, or stored anywhere. We collect no personal data.
All processing happens locally in your browser using JavaScript and WebAssembly. Your recordings never leave your computer. Once the speech engine is downloaded, you can disconnect from the internet and ScribeFit keeps working.
ScribeFit uses OpenAI's Whisper speech model. It needs a one-time download of a speech
engine (the model files): Standard is about 76 MB, and
High accuracy (Pro) is about 240 MB. The files come from
scribefit.mrrage.dev, and ScribeFit only fetches them when you press
Download the engine. The request asks for those files and sends nothing
about you or your recordings. Each file is checked against a fixed fingerprint (SHA-256) and
stored in your browser's own storage (Cache Storage). The engine runs on your device — your
recordings are still never uploaded. Delete this engine removes it again.
The ScribeFit browser extension collects nothing: no analytics, no telemetry, no cookies, no account, no ads. The extension requests only the "storage" permission, used to remember your theme choice and (if you buy Pro) your license key — both kept in your browser's local storage. It requests no access to the pages you visit. If we ever add privacy-friendly, cookieless analytics to the website version, they will never include your recordings or any personal data, and this policy will be updated first.
Purchases are processed by Lemon Squeezy (their privacy policy applies to checkout). Keys bought through Lemon Squeezy are checked once, with Lemon Squeezy, when you enter them; that request contains only the key. Older bundle keys (they start with FIT1) are checked on your device, with no request. After that, the key is kept only in your browser's local storage and isn't checked again.
ScribeFit's speech recognition is powered by OpenAI's Whisper (model weights under Apache-2.0; code under MIT), using the ONNX export published by onnx-community on Hugging Face. It runs through transformers.js by Hugging Face (Apache-2.0) and ONNX Runtime Web by Microsoft (MIT). ScribeFit's own code is MIT-licensed.