What happens to your voice (almost nothing)

Putting a microphone in someone's home is a big thing to ask. We knew that from the start. A device that's always listening only works if people trust where their voice goes.

So here's the honest answer to the question we get most: what happens to your voice when you talk to Sonora? Almost nothing. And that's by design.

The short version

Your voice is processed on the device, in your home, and is not stored or sent anywhere. There's no recording sitting on a server. There's no transcript with your name on it. When Sonora finishes helping with something, the audio is already gone.

That's the whole policy. The rest of this is how we made it true.

Always listening isn't always recording

These two phrases get mixed up constantly, so it's worth being precise.

Sonora is always listening for one thing: its wake word. That detection runs locally, on a small dedicated part of the chip. Until it hears the wake word, nothing is processed, nothing is stored, and nothing leaves the device.

When you do speak to it:

• The audio is processed on the device, by the on-device AI chip

• Sonora works out what you asked and does it

• The audio is discarded, not saved

• Nothing about the exchange is uploaded by default

The microphones are on. The recorder, in the way people fear it, simply isn't.

Why we process on the device

The easy path is the cloud. Send the audio to a big server, let it do the work, send the answer back. It's cheaper to build and easier to scale.

It also means your voice leaves your home. We weren't willing to do that, so we did the harder thing and put the intelligence inside the device. That decision shaped the whole product:

• We needed a dedicated AI chip powerful enough to understand speech locally

• We had to make the models small enough to run on it, without making them dumb

• We gave up some convenience to keep your voice where you are

It cost us time and complexity. We think privacy you can't see is worth more than features you can.

The button that actually works

Software promises are easy to make and hard to trust. So we added something physical: a mute button on top of the device that cuts power to the microphones at the hardware level.

Not a setting. Not a toggle in an app that you have to believe. A switch. When the mics are muted, they aren't quietly listening in some lower-power mode. They're off, and you can see that they're off.

We wanted there to be one thing about privacy you don't have to take on faith.

What we do keep

We want to be straight about this part too. Some things we do collect, because the product needs them:

• Your name and email, for your reservation and updates

• Basic diagnostics, like a crash report, so we can fix bugs

• Settings you choose, so your device stays set up the way you like

If you ever turn on a feature that genuinely needs the cloud, we'll tell you clearly, before you use it, never quietly in the background.

The point

A device in your home should earn its place there. For us, that meant building so that the answer to "where does my voice go" is short, true, and a little boring: nowhere. It stays in the room, does its job, and disappears.

That's the version of smart we wanted to live with. We hope it's the one you want too.