← Back to sitePrivacy by architecture

Privacy Policy

Last updated · July 26, 2026 · Applies to v0.5.0

The short version: OmniVox runs entirely on your computer. Your voice is captured, transcribed, and structured on your hardware. Nothing you say or type is transmitted to us or stored by us — there are no accounts, no API keys, and no telemetry. What the app does keep, including your transcription history, is a file on your own machine that you can read, export, or delete. We have no servers that ever see your data, because the data never leaves your machine.

The app

When you dictate, OmniVox captures audio from your microphone and processes it locally: Whisper transcribes the speech and Qwen structures it into a prompt, both running on your own CPU/GPU. Once the models are on disk this happens entirely offline — you can air-gap your laptop and OmniVox still works.

The audio itself is processed in memory and is not retained; no recordings are written to disk. The resulting transcripts are saved, deliberately — to a SQLite database on your own machine, so the History, Notes and Analytics pages have something to show you. That database is an ordinary local file: it is never uploaded, synced, or backed up to us, and you can delete individual transcripts or clear the whole history from the History page at any time.

OmniVox does not phone home. There is no analytics, crash reporting, or usage tracking of any kind. Settings and custom vocabulary are stored locally alongside your history and are never synced to us.

None of that has to be taken on trust. The app's webview ships with a Content-Security-Policy that restricts connect-src to 'self' plus three HuggingFace hosts, which exist so model weights can be fetched on first run:

src-tauri/tauri.conf.json
connect-src 'self' https://huggingface.co https://cdn-lfs.huggingface.co https://cdn-lfs-us-1.huggingface.co

Nothing outside that list is reachable from the app's interface, and none of those hosts is an inference or analytics endpoint — they serve model files. The source is published, so this is something you can check rather than something we ask you to believe.

This website

tryomnivox.comis a static marketing site. It sets no cookies and runs no analytics, tracking pixels, or third-party scripts. We don't build profiles, and there's nothing to opt out of.

The site is hosted on Vercel. As with any web host, Vercel may process standard server request data (such as your IP address and browser user-agent) to deliver pages and protect against abuse. That data is handled under Vercel's privacy policy; we don't collect or retain it ourselves.

Downloads & the installer

The install command (irm https://tryomnivox.com/install.ps1 | iex) and any download links fetch the OmniVox installer from GitHub Releases. Requesting those files is a normal connection to GitHub and is subject to GitHub's privacy statement. We don't receive any information about who downloads OmniVox.

Children

OmniVox is a general-purpose developer tool and is not directed at children. Because we collect no personal data, none is knowingly collected from anyone, including children.

Changes

If this policy ever changes, we'll update this page and the date above. Since OmniVox is local-first by design, any change here will still start from the same principle: your words stay on your machine.

Contact

Questions about privacy? Open an issue on GitHuband we'll respond there.