Is live call translation private? What happens to your audio

Is the call recorded?

No. We never record your call: nothing we run writes its audio to disk, on the server or in the database. Speech is translated as it passes through three services — one turns it into text, one translates the text, one speaks the translation — and they keep little or nothing: the audio only while it is being transcribed, the text not at all, and each spoken translation is deleted from the voice service's history within seconds.

There is no recording to download later, which also means we have none to leak, hand over, or accidentally leave in a storage bucket.

Is it end-to-end encrypted?

No, and it should not be described as though it were.

Every connection is encrypted in transit: your browser to the call infrastructure, and from there to the translation services. What breaks the end-to-end property is unavoidable and worth stating plainly — a translator has to hear a conversation to translate it. The translator joins the call as a participant and receives the audio.

This is the honest shape of every live machine-translation product, whoever sells it. A service that offers real-time translated speech and also claims end-to-end encryption is describing two things that cannot both be true at once. Treat that claim as a reason to read the rest of the page more carefully.

The practical consequence: a translated call is private in the ordinary sense — invite-only, and never recorded by us — and is not the right tool for a conversation that must be cryptographically private from the provider.

What is kept after the call?

Coarse diagnostics, and no content.

Each call writes a summary row used to tell whether the service is working: how long it ran, whether it reconnected, quality measurements, and a two-letter country code. No transcript text, no audio, no email address, no IP address, and no room name goes into it.

Those rows are kept for 180 days and then deleted automatically by a nightly job.

Who can join a call?

Only somebody with the link. Rooms are invite-only, and a room holds exactly two people plus the translator.

That two-person cap is a product decision rather than a security boundary, but it has a security consequence worth knowing: a room cannot quietly accumulate extra listeners, because there is no seat for one.

Where to check this

The privacy page states the same things in more detail, including which third parties are involved and what each one receives. It was written from the code rather than from a template, and it is candid that no lawyer has reviewed it.