Privacy
Wolfcaps turns speech into subtitles on your own computer. This page says exactly what does and doesn't leave it.
Your videos and subtitles stay on your device
- Videos and audio are read by your browser and processed by AI models running on your GPU. They are never uploaded.
- Subtitles, edits and translations are saved only in this browser (so you can continue later). When you start a new video you choose whether to keep the project. Remove saved projects with the × in “Recent projects”, with the button below, or by clearing this site's data in your browser. Videos you send from your phone are kept in this browser too, until you remove them.
- Exported files are written straight to your disk.
Sending a video from your phone
- With “Send from your phone”, the video goes directly from your phone to your computer over an encrypted connection (WebRTC). It never passes through our server and is never stored anywhere else.
- Only the connection setup passes through our server: a few kilobytes that tell the two devices how to reach each other (their network addresses and encryption keys). It is relayed between your two devices, not stored, and forgotten when the code expires (10 minutes without activity). The code in the QR link is random and is never sent in a web address our server logs.
- To find a direct route, each device asks a public STUN server (Cloudflare's and Google's) for its own internet address. Those servers see your IP address, nothing else.
- The received video is kept in this browser's private storage so you can continue its project later. Remove it with the button below or on the home page.
What we download to your device
The AI models (1–4 GB each) are downloaded once from Hugging Face and cached in your browser. Like any website you download from, Hugging Face sees your IP address for those downloads; it never sees your videos.
Remove what's stored in this browser
Translating with a chatbot
If you choose “Chatbot” translation, you copy the subtitle text yourself and paste it into a chatbot of your choice (ChatGPT, Claude, Gemini…). That text then goes to that service under its own privacy policy. Wolfcaps sends nothing anywhere for this; it only prepares the text and reads back the answer you paste.
Anonymous usage statistics
To learn which computers and browsers people use and what goes wrong, the app sends a few anonymous events to our own server (no cookies, no third-party scripts, nothing that identifies you):
| Visit | browser and version, operating system, GPU type, number of CPU cores, whether this device can run the app |
| Subtitles made | finished / cancelled / failed, or taken from subtitles already in the file (with their format, e.g. SRT), spoken language, quality setting, video length, processing speed, number of lines |
| Translation | method (Fast / Best / Chatbot), language pair, number of lines, time taken |
| Export | format (SRT, VTT, ASS, TXT, MP4, MKV), time taken, the subtitle style preset (e.g. “Netflix”); for a Reels & Shorts video also the caption style and font names (e.g. “Bold”, “Rubik”), the video height and length |
| Phone transfer | which way (from your phone, or an export sent to it), finished / cancelled / failed (and why, e.g. no direct connection), connection type (same network or across the internet; IPv4 or IPv6), file size range (e.g. 100 MB–1 GB), time taken, speed |
Never: file names, video or audio, subtitle text, or your IP address. The server adds only your country (from the connection) to each event.
Credits and licenses
- Speech recognition: OpenAI Whisper (MIT); Hebrew: ivrit.ai Whisper models (Apache-2.0), converted for the browser by us.
- Translation: Meta NLLB-200 (CC-BY-NC 4.0, non-commercial) and Google Gemma (Gemma Terms of Use).
- Speech detection: Silero VAD (MIT). Runtime: Transformers.js (Apache-2.0), ONNX Runtime Web (MIT). Media: Mediabunny (MPL-2.0).
- Caption fonts (Reels & Shorts), served from this site: Rubik, Heebo, Assistant, Secular One, Karantina, Suez One, Varela Round, Fredoka, Montserrat, Poppins, Anton and Bebas Neue, under the SIL Open Font License 1.1 (copyright notices and license).
- H.265 video preview (in browsers that can't play it): the HEVC decoder of FFmpeg 7.1.2 (LGPL 2.1 or later; FFmpeg's license notes), unmodified, compiled to WebAssembly by us. Its source: ffmpeg-7.1.2.tar.xz.