What happens to your video

Why this is structural, not a policy

Most privacy statements are promises about what a company does with data it has. This one is different in kind: CaptionSuite has no server that could receive your file. There is no upload endpoint, no processing queue, and no storage bucket, because the work happens in the tab.

That is not generosity. It is the same decision that makes the tool free and unlimited: running on your hardware costs us nothing per video, so there is nothing to ration and no reason to build the infrastructure that would create a privacy question in the first place.

Exactly what crosses the network

Four things, and nothing else:

• The page itself — HTML, JavaScript and CSS, from our host.
• The speech model — 45MB or 80MB depending on which you pick, from the Hugging Face CDN, the first time you transcribe. Your browser then caches it. This is a download to you.
• Anonymous page analytics — which pages were viewed, from Vercel. No file names, no file contents, no transcripts, no account.
• How the page is used — Microsoft Clarity records clicks, scrolling and pointer movement, which is how we find out that people reach the style panel and stop there. The editor is masked: the whole tool carries Clarity’s mask attribute, so your file name, your transcript and the video frame are never part of that recording. It is set in the page itself rather than in a Clarity setting, so changing a setting cannot turn it off. Clarity also sets two cookies to tell one visit from the next.

Your video, its audio, and the captions you produce are not in that list. They exist in the tab’s memory while you work and are gone when you close it.

Where the model is stored, and how to remove it

The speech model is downloaded once and kept by your browser in its Cache Storage — the same mechanism a site uses to work offline. It is stored per site, counts towards this site’s storage quota, and stays there until you clear it, which is why the second video you caption starts transcribing immediately.

Nothing of yours is stored alongside it. The cache holds the model files and nothing else: no video, no audio, no transcripts. Those live only in the tab’s memory while you work.

You do not have to go through browser settings to get it back. Below is what is cached on this device right now, read from the cache itself rather than assumed, with a button that deletes it. On disk it sits inside your browser’s own profile folder, in its Cache Storage directory, not anywhere you would normally keep files, so this is the easier way to reach it.

Clearing site data for this domain does the same job and a bit more: in Chrome or Edge it is the padlock in the address bar, then Site settings, then Delete data; in Firefox and Safari, clear website data for this site. Either way the model is downloaded again the next time you transcribe.

How to verify it rather than believe it

A privacy claim you cannot check is a marketing claim. Check this one:

• Open your browser’s developer tools and choose the Network tab
• Drop in a video and transcribe it
• Sort the requests by size. The only large ones should be model files from a Hugging Face domain
• Look for any request with a large upload payload — there should be none
• You will see small requests to clarity.ms. That is the usage recording described above, and the editor is masked out of it. Expand one and you will not find your transcript in it
• The strongest test: once the model has downloaded, disconnect from the network entirely. Transcription and burn-in both still work

If you cannot allow the model download either

Some environments cannot fetch from a third-party CDN at all. The model download is a normal HTTPS request for static files and contains nothing of yours, but if your policy forbids it, the honest answer is that automatic transcription will not work — you can still import an existing .srt or .vtt file, edit it, and burn it in, all of which run with no network at all.

Files you download

The MP4, .srt, .vtt and transcript files the tool produces are created in your browser and saved by your browser to wherever you keep downloads. They do not pass through us.