Network Transparency
What leaves your machine, and how to check for yourself.
Audited: September 2026 · MurmurTone 1.0.x
The guarantee
- Your voice audio is never transmitted. Speech recognition runs entirely on your machine.
- Your transcripts are never sent by MurmurTone to us or any third party.
- No analytics, telemetry, or crash reporting in the desktop app.
MurmurTone runs speech recognition on your own computer. This page lists every network connection the app makes, why it makes it, and how to verify it yourself in about five minutes. The app's network behavior can only change in a signed update that you choose to install.
Every connection the app makes
| When | Host | What is sent |
|---|---|---|
| Startup and update check (about daily) | downloads.murmurtone.com | A request for a small version manifest (a generic "MurmurTone" user-agent and your IP address, as with any web request) |
| You click "Update now" | downloads.murmurtone.com | Downloads the signed installer, which is verified before it runs |
| You activate a license | api.polar.sh | Your license key and a device label (your computer's hostname and operating-system name) |
| You click "Upgrade" in the app | our Cloudflare Worker checkout broker | Creates a Polar checkout session and retrieves your license key |
| You choose a model or enable GPU | downloads.murmurtone.com | Downloads the speech model or GPU libraries you selected |
None of these ever carry your audio or your transcripts. The checkout host is our own Cloudflare Worker: a stateless broker that creates your Polar checkout and hands back your license key. It carries no audio, no transcripts, and no payment-card data (Polar processes payment). We are moving it to a checkout.murmurtone.com address.
Verify it yourself (Wireshark, about 5 minutes)
- Install Wireshark and start a capture on your network adapter.
- Use MurmurTone normally: launch it, dictate a few sentences, and click "Check for updates" on the About tab.
- Watch the capture while you dictate: transcription produces zero network packets. It runs entirely on your machine.
- To see which servers the app contacts, apply the display filter
tls.handshake.extensions_server_name. You will see only the hosts listed in the table above.
What a capture also shows (and it is not us)
Windows makes its own background connections that appear in any packet capture. These are the operating system, not MurmurTone sending your data:
- Microsoft SmartScreen and certificate-revocation checks (OCSP/CRL) when Windows validates our signed app.
- The Microsoft Edge WebView2 runtime (which powers the settings window) checking for its own updates.
- The Windows connectivity probe (msftconnecttest.com).
None of these carry your voice or your text. Separately, the first time you run a downloaded installer, Windows may show a SmartScreen "unknown publisher" prompt: that is a Windows reputation feature for new downloads, not a network upload. See our FAQ for more on the install-time prompt.
Optional local-AI text features
An optional AI text-cleanup feature is coming. When you enable and configure it, it sends the transcript text (never your audio) to a local AI model on a host you choose, by default your own machine (localhost). If you point it at a different host, the transcript text goes only where you send it. Your voice audio still never leaves your device.
Questions about any of this? Email support@murmurtone.com. See also our Privacy Policy.