Privacy Policy

This tool does not require an account and does not use advertising or analytics trackers. Limited data is processed only to provide and protect the transcription service. Audio handling depends on the device you use.

• On desktop: Transcription happens entirely inside your browser using Whisper Small or Moonshine AI models. No audio, text, or personal information is sent to a server.
• On mobile: Audio is sent to temporary cloud processing (Vercel Blob storage, or directly through our server on iPhone Safari), transcribed by Groq Cloud using Whisper Large V3, and deleted after processing.

Desktop: Audio is decoded and processed entirely within your browser. English audio uses the Moonshine Base model; other languages use the Whisper Small model. Both run through WebGPU with a WASM fallback. The model files are downloaded once from Hugging Face and cached in your browser. After that, transcription can work offline.

Mobile: Audio is split into chunks and compressed on your device. On most browsers each chunk is uploaded into a temporary Vercel Blob object under a random, unguessable address, and Groq fetches it from that address for transcription. On iPhone Safari, chunks are sent through our server straight to Groq without being stored. Successful chunks are deleted immediately. Failed or abandoned uploads are removed by a daily cleanup job once they are older than five hours, so they remain for at most about 29 hours.

Mobile API requests are checked by Vercel BotID and rate limited. The application converts the request IP address into a keyed HMAC value before placing it in the shared rate-limit store; the raw IP address is not stored by the application. Short-term rate-limit entries expire within 30 minutes, and the daily cloud usage counter (the amount of audio processed, without any audio content) expires within 24 hours.

Hosting and API providers may process standard network logs under their own privacy and retention policies. A separate random secret is used for transcription grants, and no rate-limit identifier is included in transcript output.

You can export your transcript as plain text, timestamped text, JSON, Word, or PDF. All document generation happens entirely in your browser; no data is sent to a server for export.

The "Continue with AI" buttons copy your transcript to the clipboard and open ChatGPT, Claude, Gemini, or Grok in a new tab. We do not send your transcript to these services. You choose whether to paste it, and each service's own privacy policy applies once you interact with it.

AI model files, approximately 250-410 MB depending on the model and your device, are cached in your browser's native cache after the first download. You can clear this cache at any time through your browser settings.

We use a single localStorage flag to remember whether the model was previously loaded and skip redundant loading UI. We do not store audio, transcripts, personal data, or tracking identifiers in local storage.

• Hugging Face on desktop: Used only to download open-source AI model files on first use. The Hugging Face Privacy Policy applies.
• Vercel (hosting, Blob and BotID) on mobile: Provides hosting, temporary storage, and abuse prevention. Uploads are deleted after processing or by the daily stale-file cleanup.
• Upstash Redis on mobile: Stores only keyed, pseudonymous rate-limit and daily usage counters with a maximum 24-hour lifetime. Audio and transcripts are never stored there.
• Groq Cloud on mobile: Processes audio using Whisper Large V3. Groq's terms and retention policy apply while it processes the request. See the Groq Privacy Policy.
• AI chat services when requested: ChatGPT, Claude, Gemini, and Grok are opened only after you choose a "Continue with AI" action. We do not send transcript data to them automatically.

This service is not directed at children and does not knowingly create profiles about children. The same limited processing described above applies to every request.

If this policy changes, the updated version will be published at this URL with a revised "Last updated" date.

Questions about this policy can be sent through LinkedIn or by opening an issue on GitHub. The source code is open (MIT License), so you can verify how your data is handled.