Privacy Mode & Data Retention

Content you do not save is not retained. Content you do save is stored as history encrypted on your device, which we cannot read. There is no path on which we hold your translation content in a form we can read on our own initiative, and we never use your content to train AI models.

Two features are the exception, and both need your action: group polls, and voice messages you ask us to translate. Section 5 sets them out.

To translate or transcribe what you say or type, we and the speech, translation, and AI services behind the result have to process it. That is unavoidable - it is how the feature works.

Retention is the separate question of what survives afterwards. The controls described here are about retention. They are not a claim that cloud processing never happens.

When something becomes part of your history - a saved translation, a transcript, a recording, or an AI conversation - your device encrypts it before it leaves, and we store ciphertext. Documents and saved audio work the same way. Messages are end-to-end encrypted between participants.

By default your encrypted history syncs across the devices you sign in to, so your work follows you. You can change that at any time in Data Controls: Cloud sync, Manual, or Off, where the server rejects cloud history writes so nothing new leaves your device.

One point worth being precise about: when you ask us to re-transcribe or re-translate a recording you already saved, your app supplies the key so that recording can be processed for that request only. We cannot read your stored content on our own - only when you ask us to act on it.

Live conversation translation, dictation you do not save, and AI turns you discard are processed to produce your result and then dropped. Live speech exists briefly in memory during processing and is not written to storage.

We never sell your data, and we never use your content to train our models.

There are exactly two exceptions, and both require you to use the feature:

• Group polls. The server has to read a poll to tally votes and show results, so polls are not end-to-end encrypted. They are disabled in healthcare conversations.
• Voice-message translation. Translating a voice message requires decrypting that one message on our servers. We ask for your consent first, record the consent, and audit the access. It never happens unless you choose it.

Your content is processed by third-party speech, translation, and AI services solely to produce your result. We configure them to keep your content out of their model training and to minimize what they retain, wherever the service supports it.

Those speech, translation, and AI services receive only the content needed to produce your result - never your name, account, or identity. We make the calls on your behalf, so a provider cannot link a request back to you. Payment and app-store providers are different: they necessarily receive identifying and transaction data, because a payment cannot be processed without it.

Running an account requires records that are not your translation content: account and authentication details, billing and subscription history, security and abuse-prevention logs, support conversations, and, where the law requires it, audit and compliance metadata.

Saved files, documents, generated images, messages, and projects are retained while you use those features, until you delete them or close your account. Healthcare audit logs hold metadata only - never PHI, never translation content. The retention period for each category is listed in Section 7 of the Privacy Policy.

Healthcare accounts operate under an executed Business Associate Agreement, with stricter session limits, audit logging, and PHI-handling controls, and use only services covered by a BAA. Features that cannot be end-to-end encrypted, such as group polls, are disabled in healthcare conversations. These protections apply only to accounts approved for healthcare use under a signed BAA.

Enterprise customers can request organization-level review of history and storage policy, data residency, a dedicated managed cloud deployment, or an on-premises deployment.

You can delete individual history entries or your whole history, switch sync off, export your data, or delete your account from the product controls - see the Data Deletion page. Billing records, fraud-prevention logs, and required audit and compliance records may be retained where law or a signed agreement requires it.

Live translation, dictation, and AI content is processed to produce your result and is not kept as cloud history unless you save or sync it. Anything you do keep is stored only as encrypted history that Vavus cannot read.

Content you do not keep is not retained at all, and content you keep is stored only as history encrypted on your device before it reaches us. The two exceptions are group polls and voice messages you ask us to translate, both of which require you to use that feature.

No. We never use your content to train our models, and we contract with the third-party speech, translation, and AI services we use to exclude your content from training theirs and to minimize retention where supported.

No. Saved history is encrypted on your device, so the server stores only ciphertext. The exception is when you ask us to re-transcribe or re-translate a saved recording - then your app supplies the key so that recording can be processed for that request only.