Privacy · last updated 2 November 2026

What happens to your voice

You are about to speak private things into someone's software. You should know exactly where that goes. This page is written to be read, not to be survived.

The short version

Your recording is sent to OpenAI, in the United States, to be transcribed and translated. There is no other option today.
OpenAI may keep it for up to 30 days to detect abuse, then deletes it. They do not train their models on it.
Our server never writes your audio, transcripts or translations to disk. They pass through memory and are gone.
The iPhone and Android apps keep your past conversations on your phone — the words, never the audio — so you can look something up later. Not on our server, not in iCloud or Google backups.
There is no account, so we hold no name, email, phone number or password. Passes are paid through Apple; we never see your card or your Apple ID.
No cookies. No advertising. No analytics service and no third-party scripts of any kind.
Because of the first two points, this app is not appropriate for privileged, clinical or legally sensitive conversations. See below — we would rather lose the sale.

What happens, step by step

You hold the button and speak. Your phone records compressed audio, about 20 KB for a typical sentence. Nothing has left your phone yet.
You let go. The recording is sent over an encrypted connection to our server.
Our server forwards it to OpenAI to be turned into text, and then sends that text — along with the recent conversation, so pronouns and short answers still make sense — to be translated.
The translation comes back and is shown and spoken on the phones in the conversation. If you chose the natural voice, the translated text is sent to OpenAI once more to be read aloud.
Our server keeps none of it. The audio exists in memory for the seconds it takes to process, and is never written to disk.

What we store

On our server, in one database, only this:

A random identifier your phone generated the first time you opened the app (or the website, if you joined from a browser). It is not derived from your phone, your number or you. It is a random string.
How many free translations that identifier has left and has used, and when a pass it bought runs until.
The dates it was first and last seen, and whether it is the iPhone app, the Android app or the website.
For each translation, and each group conversation started: when, and in which languages. Never what was said.
For each purchase: which pass, the price, when, and Apple's transaction number for it, which is how a refund finds the right pass. Apple handles the payment; we never see your name, card or Apple ID.
Promo codes, what they are worth, and whether they have been redeemed.
Messages you send us with the contact form: what you wrote, the email address if you gave one, and when. Deleted once we have dealt with them.

That is the entire contents. No audio, no transcripts, no translations, no location — and no names or contact details, except what you choose to put in a message to us.

What we don't store

Your recordings. Held in memory during processing, never saved.
Your transcripts or translations. Same.
Your conversation. Our server never keeps it. The iPhone and Android apps keep its words on your phone, in Past conversations (see what is kept on your phone); on the website, it disappears when you close the page.
Group conversations. A shared room exists only in server memory. It is deleted when the host leaves, and automatically after four hours.

Our server logs record errors, purchases and code redemptions. They never contain what anyone said.

What OpenAI does with it

Transcription, translation and the natural voice are all performed by OpenAI on servers in the United States. Exactly what is sent to OpenAI: the recording of what is said, the text transcribed from it, the recent sentences of the conversation (so pronouns and short answers still make sense), and the translated text when it is read aloud. Nothing that identifies you goes with it: no name, no account, no device identifier.

The app asks for your permission before the first conversation, and sends nothing until you agree. OpenAI processes this data only to provide these services to us, under the commitments below, which give it the same or equal protection as everything else on this page. As of their published policy:

They do not train their models on data sent through the API by default.
They may retain inputs and outputs for up to 30 days to provide the service and identify abuse, after which it is removed unless they are legally required to keep it.
Access is limited to authorised employees and specialist contractors bound by confidentiality, for abuse investigation and legal compliance.
Data is encrypted in transit (TLS 1.2+) and at rest (AES-256). They have completed a SOC 2 audit.

Source: OpenAI's enterprise privacy commitments. We have quoted their terms rather than paraphrasing them into something more comfortable.

What is kept on your phone

Stored on your phone, never sent to us except where noted:

Your random identifier, in the iPhone's keychain (on Android, in the app's own storage), so your remaining translations and your pass survive reinstalling the app. This is the one item sent with each request.
Your chosen languages and voice preference.
Your past conversations, in the iPhone and Android apps: what was said and its translation, who said it, the languages and the time — never the audio. The newest 100, so you can search them later for an address or the name of a medicine. They are never sent to us and are left out of iCloud and Google backups, so they stay on that phone. Delete one or all of them in Past conversations; they also go when the app is deleted.

A pass is also on record with Apple, so Restore purchases in the app brings it back on a new iPhone signed in to the same Apple ID. Someone who joins from a browser keeps their identifier in that browser instead.

Things we do not do

No cookies.
No analytics service and no tracking. We count translations — when, and between which languages, against the random identifier — to run the free translations, passes and fair use, and to see how much the app is used. Never what was said.
No advertising, and nothing sold or shared with anyone.
No third-party scripts, fonts or embeds. Every page loads only from this domain.
No profile of you, because we have nothing to build one from.

One honest footnote: whoever hosts this site sees connection records, including IP addresses, in the ordinary way that any web server does.

Please read this part

Where this app is not appropriate

Because recordings leave the country, may be held for up to 30 days, and we offer no data processing agreement, Mama Translate should not be used for conversations that carry a professional duty of confidentiality.

Clinical care. Use your organisation's approved interpreting service. Health information is special category data and this app is not assessed or approved for it.
Legal advice or anything privileged. Use a certified interpreter. We cannot give you a processing agreement to put on file.
Asylum, immigration or safeguarding. The people in these conversations often have the most to lose, and the least ability to consent meaningfully.
Anything safety-critical. Not a privacy point, but the same principle: there is a delay and speech can be misheard.

If you need this app to be usable in those settings, the honest answer today is that it is not. We would rather say so than sell you something you would have to defend later.

Telling the other person

The other people in a conversation are speaking into your phone, and they may not know where their words are going. We think you should tell them, in plain words: "this app sends what we say to a translation service abroad." If they are uncomfortable, that is a reasonable thing to be.

If you are using the natural voice, tell people the voice they hear is AI-generated. OpenAI's usage policies require that disclosure, and it is the decent thing regardless.

Children

This app is not designed for children to use on their own. An adult may of course use it to talk with a child, which is one of its better uses.

Deleting your data

Deleting the app removes your languages and voice choice. The record on our side is a random identifier and a few numbers, with no way to connect it to you by name. To have it removed as well, write to us and we will find it with you and delete it.

When this changes

If we ever begin storing conversations, add analytics, or move processing somewhere new, this page will say so and the date at the top will change. We will not quietly broaden what we do and leave the page as it was.

Who we are

Mama Translate is made by Dario Presezzi, an independent developer, who is responsible for the data described on this page.

Contact

Write to us — for deletion requests, questions, or to tell us this page is wrong about something. Leave an email address if you would like an answer.