Privacy Policy

Last updated September 7, 2026

Unwatch is operated by Elowin Labs LLC ("we"). This page describes what happens to data when you use the site at unwatch.video. The short version: the summary is written on your device, the server only sees the id of the video whose captions it fetches, and nothing you summarize is stored on our side.

What the server receives

When you submit a video, your browser asks our server for that video's public caption track by its 11-character video id. The server fetches the captions (from YouTube, from a caption service, or from a machine we operate for that purpose), returns them to your browser, and keeps a copy of the caption text in a cache keyed by video id for up to 30 days so later requests for the same video are faster. The cache contains the public captions only; it is not linked to you.

Requests are subject to per-network rate limits, which use your IP address for the duration of the limit window (one minute) and are not stored beyond it. Our hosting provider, Cloudflare, keeps sampled request logs (URL, status, timing, country, user agent) for operational debugging for a limited period.

What stays in your browser

The transcript is split up and summarized by a language model that runs inside your browser: Chrome's built-in Gemini Nano, or an open model executed with WebGPU. The transcript text and the summary are not sent to us. Your browser stores the following under this site's origin: unwatch.theme (light or dark), unwatch.engine (the engine you picked), unwatch.webllmModel (the WebGPU model you picked) and unwatch.history (your last 20 summaries, so you can reopen them). Clearing the site's data in your browser removes all of it; the Clear button under Recent summaries removes the history.

The first use of a WebGPU model downloads its weights from Hugging Face and the WebLLM library from the jsDelivr CDN; your browser caches both. Chrome downloads Gemini Nano from Google under Chrome's own terms. Those downloads are requests your browser makes to those providers, subject to their privacy policies.

Cloud fallback

If your browser can run none of the on-device engines, the page may offer a cloud fallback when we have enabled it. In that case, and only then, the parts of the transcript being summarized are sent through our server to a hosted model provider (OpenAI) for that request; the page labels the result "cloud fallback". We do not store the text, and the fallback is rate-limited per network. You can always choose an on-device engine instead when one is available.

Video metadata

Your browser fetches the video's title, channel name and thumbnail directly from YouTube's public oEmbed and image endpoints, which means YouTube sees those requests from your browser, as it would if you opened the video.

Analytics

We use Plausible, a privacy-focused analytics service hosted in the EU, to count visits and a few named events (a summary started, which engine was used, whether captions could be fetched). Plausible does not use cookies, does not collect personal data and does not track you across sites. Events never include the video id, title or any text.

Cookies

Unwatch sets no cookies. See the cookie page for the full list of browser storage used.

Your rights

Because we do not keep personal data about you, there is nothing for us to export or delete on request beyond what your browser holds. If you have a question about this policy, write to hello@elowinlabs.com.

Changes

We will update this page when the way the site handles data changes and bump the date above.