Privacy policy
Last updated October 28, 2026
OpenSubs is a browser tool that transcribes speech into subtitles, translates them, styles them, and burns them into a video. This page describes exactly what data the tool collects, what it never collects, and why. The OpenSubs app is open source (AGPL-3.0-or-later), so what it does with your files can be checked in its code.
The OpenSubs desktop app for Windows has its own privacy policy.
The short version: your video never leaves your machine. Transcribing, styling, previewing and burning happen 100% in your browser, and there is no endpoint to upload a video to. Your content leaves your device only when you choose a cloud service: subtitle text if you pick a cloud translator, and the audio span you trimmed if you pick cloud transcription. Signing in, to burn a video or to pay for translation, also uses your sign-in session.
What OpenSubs never collects
- Your video or audio. The file is read through an object URL and decoded by your browser. There is no upload endpoint in this product, so a footage leak is not something we have to promise against — it is not possible.
- What you subtitle. We have no record of which files you open, how long they are, or what is in them.
- Your API keys. If you bring your own key for a translation or transcription service, it is used from your browser directly to that provider. It is never sent to us and never written to disk.
What's stored locally, on your device only
- The speech model. The first time you transcribe, ONNX Whisper weights are downloaded from
huggingface.co, or in mainland China from ModelScope,modelscope.cn, and cached by your browser. Each is the other's fallback, and you can pick either under “Download from”. That is a download, not an upload — no audio is in the request. - Your preferences. Chosen style, word effect, output size and similar settings live in your browser's local storage.
- Your subtitles, briefly. Signing in navigates the page away and back, which would otherwise throw away a transcription you just waited for. The cues are kept in local storage for 24 hours so they survive that, and you can discard them at any time.
- A pointer to your video — never the video. For the same reason, your browser is asked to remember a reference to the file you chose, so it can be offered back after sign-in. No copy of the footage is made, and your browser asks your permission before opening it again, and this link to your video is deleted together with your subtitles once they are 24 hours old.
- Your session token, if you sign in. The page also remembers when our account server last confirmed you are signed in, checked each time you open the page while signed in. Used to let a video export go ahead, to check your credit balance and to authorise paid translation.
Clearing your browser's site data for opensubs.app removes all of it.
Data that leaves your device, only if you choose it
- Subtitle text, for cloud translation. Selecting a cloud translator instead of the on-device one sends the subtitle lines to it. Never the video, never the audio — just the text. The on-device option sends no text anywhere, and it is the default.
- Audio, for cloud transcription. Selecting an OpenAI-compatible transcription service and supplying your own key uploads the trimmed audio span to that service. Only the span you trimmed to, not the whole file.
Data used only if you sign in
You can try OpenSubs without an account. One is needed to burn subtitles into a video, which takes a free sign-in, and to pay for translation on our own backend.
If you create one, we hold what an account needs: an identifier for how you signed in, your credit balance, and a ledger of what credits were spent on — the product and the feature, for example “OpenSubs · translate”. The ledger records that a translation happened and what it cost. It does not record the subtitle text.
Paid translation
When you translate on our backend, the subtitle lines are sent to gateway.opensubs.app, which passes them to the translation model we hold a key for and returns the result. The lines are used to produce your translation. They are not retained by us afterwards, and they are not used to train anything.
You are shown the price before the job runs and charged exactly that. A job that fails costs nothing.
Payments
Payments are processed by our payment providers. We never see or store your card details. Credit purchases and their reversals are recorded against your account so your balance is correct.
Messages you send us
“Ask us on Telegram”, which OpenSubs offers where something has gone wrong and at the foot of the app, opens the OpenSubs topic in our public group on Telegram, in a new tab. Opening it sends us nothing. Where something has failed, OpenSubs shows a short error code for you to include in your message; it names the type of error and never your video, its audio, your subtitles or a file name. What you post in the group is public, can be read by anyone in it, and is governed by Telegram’s terms and privacy policy, not by this page, so please don’t post passwords, card details or private files.
Messages sent through the message box OpenSubs used to have are used only to look into the problem and reply, and are deleted 180 days after they were sent, or sooner if you ask at hello@opensubs.app.
This website
Page views on this site, and which buttons are clicked, are counted by a self-hosted, privacy-preserving analytics service. It sets no cookie, records no personal data, and does not follow you to other sites. It never sees your video or your subtitles. To find out where burning a video goes wrong, so we can fix it, the web app also keeps anonymous counts of how far a task gets, for example a burn started, finished or failed, or a subtitle file downloaded. A count carries only the step and details such as the file format, the type of device, how you signed in and the kind of error — never your video, its audio, your subtitles, file names or your account. The browser extension and the desktop and phone apps send none of this.
Third parties
- Hugging Face and ModelScope — host the speech model files: ModelScope (run by Alibaba) for mainland China, Hugging Face for everywhere else, each the other's fallback. Contacted only to download them, once.
- A translation or transcription provider of your choosing — only if you select a cloud option, and under your own key unless you use ours.
- Our payment providers — only if you buy credits.
There are no other third parties. No CDN, no analytics vendor, no advertising network.
Data deletion
Local data: clear site data for opensubs.app in your browser.
Account data: email hello@opensubs.app and we will delete your account and its ledger. Since the app itself holds nothing about you, that, and any message you sent us, is the whole of what we have.
Children's privacy
OpenSubs is not directed at children under 13, and we do not knowingly collect information from them.
Changes to this policy
If this policy changes materially, the date at the top changes with it.
Contact
Questions about this policy: hello@opensubs.app.