Privacy policy

Last updated: 21 October 2026

Polyloqui is built so that as little as possible leaves your phone. Speech recognition, translation and voices run on your device whenever it can do them. There are no accounts, no advertising and no tracking. This policy explains exactly what happens to your data in the Polyloqui app for Android and in the Polyloqui web version, and who else is involved.

Who is responsible

Polyloqui is made by an independent developer. To contact us about privacy, use the developer email address shown on Polyloqui's page in the Google Play Store.

What stays on your device

Speech recognition and translation. The speech and translation models you choose to download run on your phone. When they are available, what you say is turned into text and translated on the phone itself.
The interpreter. Fully on the phone once its language packs are downloaded; nothing is sent anywhere.
Natural voices. Read-aloud voices you download speak on the phone.
Voice setup. The sentences you read during voice setup are kept in memory only while the screen is open, to measure accuracy and your voice level. They are never saved or sent.
Settings and saved items. Your name, languages, saved phrases, your permanent room code's secret, and a random device key (used only to put you back in your own place after a crash) are stored on your device.
Saved conversations. A conversation is only saved when you tap "Save the conversation", and then only where you choose to share or store it.

Voice rooms

Your voice goes to the other people in the room. In rooms of up to 8 people it goes directly from phone to phone, encrypted (WebRTC). When phones cannot reach each other directly, the encrypted audio passes through a relay (Cloudflare). In larger rooms and broadcasts it passes, encrypted in transit, through a media server (LiveKit Cloud). Neither we nor these services record it.
Everything said in a room is written down for everyone in that room. This is how Polyloqui works: each person's device turns their speech into text and the text is shown to all members of the room, so everyone can follow and translate. The room is told this when you join. The text passes through our room server to reach the others; it is not stored there.
The room server (hosted on Render) introduces the phones to each other and runs the room: it handles the name you type, the room code, who is muted or speaking, the language you speak and read, and the transcript lines on their way to the other members. It keeps this in memory only while the room is open. It also sees your IP address, which it keeps in memory for a short time to limit abuse. It writes nothing about you to disk.
Hosts can mute and remove people, lock the room and decide who comes in. Offensive words are hidden in captions unless the host turns that off; this happens on each device.
Blocking someone happens only on your device: you stop hearing them and seeing their lines.
Reports. If you report someone, we receive the reason you chose, your name and theirs as shown in the room, the room code, and up to their last 20 lines as your device showed them. This goes to our server's log and to our moderation channel, and is used only to deal with the report and prevent abuse. The reported person is not told who reported them.
Feedback. If you send feedback from the app, we receive what you wrote, whether it is an idea or a problem, and the app version. It goes to our server's log and our moderation channel. Nothing else is attached; add how to reach you only if you want an answer.

When a cloud service helps

If a device cannot do a task itself (mostly web browsers without downloaded models), the room server can ask a cloud service to do it. Only the text or audio needed for that task is sent. Each service handles it under its own privacy policy (DeepL, Microsoft, Deepgram, Google, Cloudflare, Anthropic); where a service offers a setting not to use our requests to improve its products, it is switched on:

Translation of transcript lines: DeepL or Cloudflare Workers AI, or another phone in the room that has the Polyloqui app.
Natural read-aloud voices in the web version: Microsoft Azure Speech (the text to be spoken is sent).
Speech recognition as a last resort in the web version: a speech service such as Deepgram or Google Cloud or Cloudflare Workers AI, if enabled (short audio clips of your own speech are sent). The web version can also use your browser's own speech recognition; in Chrome that sends audio to Google.
Summaries, only when someone in the room taps "Summary": made on that phone when it has the smarter-translation model; otherwise, if the server offers summaries, the room's transcript is sent to Cloudflare Workers AI or Anthropic (Claude) to write it.

On the phone only

Translating a sign or menu: the photo is read on the phone (Google ML Kit text recognition, which runs on the device) and deleted straight after; it is never uploaded.
Live captions and name alerts run on the phone; nothing is recorded or sent. When the phone hears your name it buzzes, and so does your watch if one is connected.
Your phrasebook (lines you save from the interpreter) stays on the phone.
NFC tags: writing a room link to a tag happens on the phone; the tag holds only the link.
Ratings: the app may ask Google Play to show its rating card once; we do not see your rating.

The Android app also uses

Firebase Crashlytics (Google) to receive crash reports: what failed and where in the app, the app version, the phone model and Android version. Crash reports never contain what anyone said.
Firebase Cloud Messaging (Google), only if you turn on "Tell me when this room opens": your phone gets a notification token and subscribes to that room's notifications.
Google Play Billing, only if you buy Polyloqui Pro: Google handles the payment under its own privacy policy; we never receive your card or payment details. The app asks Google Play whether you have Pro and remembers the answer on your phone. To check that a purchase is real, the app sends its purchase token to our server, which asks Google Play about it and gives the app a signed pass; we do not keep the token.
Google ML Kit for on-device translation and language identification. Its language models are downloaded from Google.

Downloads

Speech models and voices are downloaded, when you choose them, from Hugging Face and GitHub (and for the web version possibly through our server). Those services see your IP address as with any download.

Error reports from the web version

If the web version hits an error, it sends our server the error message, the file and line, and your browser type, so we can fix it. It never sends what anyone said.

Permissions the app asks for

Microphone: to carry your voice in rooms and to hear you in the interpreter and in "Hear better". Used only while you use those features.
Camera: only to scan a room's QR code. Images are not saved or sent.
Notifications: to show that a room is active (so it keeps working with the screen off) and, if you ask for it, when a room opens.
Bluetooth: to use your Bluetooth earbuds.

Children

Polyloqui is not directed at children under 13, and we do not knowingly collect data from them.

How long data is kept

Room data exists only while the room is open. Nothing about you is stored on our server, except reports, which we keep for up to 90 days after the report is dealt with. Crash reports are kept by Firebase for up to 90 days. Data on your device stays until you delete it or uninstall the app.

Your rights

Because we do not store data about you, there is normally nothing for us to give you or delete. You can delete everything on your device by clearing the app's data or uninstalling it. If you are in the European Union or another region with data protection laws, you have rights to access, correct, delete and object to processing of your personal data, and to complain to your data protection authority. Contact us through the developer email on Polyloqui's Google Play page.

Changes

If this policy changes, the new version will be published here with a new date.