Privacy

Last updated September 24, 2026 · Draft pending legal review

Pardon? shows live captions of a conversation so someone who is hard of hearing can follow along. That means it handles sensitive things: what people say, and what their voices sound like. Here is exactly what happens to them.

The short version

• Audio is never saved. While a table is running, sound goes to our speech-to-text provider to become words, and is then discarded.
• Transcripts are deleted after 30 days.
• Voice fingerprints are kept only for people who said OK, and you can delete them at any time.
• No ads, no selling data, no training AI on your conversations.

What we collect

Names you enter. Your name, your circle's name, and names of people you add (for example "Grandma"). These show on captions.

Transcripts. The words spoken while a table is running, who said them, and when. They're shown to members of your circle and used for "Catch me up." We delete them 30 days after the table.

Voice fingerprints (only with consent). If someone agrees, Pardon? learns what their voice sounds like so it can put their name on their words. A voice fingerprint is a list of numbers computed from a few seconds of speech. It isn't a recording and can't be played back. We keep one only after that person agrees, which they give on their own phone or which you confirm on their behalf (for example, after asking Grandma). Anyone can delete a fingerprint, or withdraw consent, in the app; withdrawing deletes it.

A device key. When you join a circle, your phone gets a random key so we know it's you. Joining from an invite needs no account.

Your account (optional). If you sign in with Apple, Google, or an email link, we store your email address, the name your sign-in provider shares, and which circles belong to you. With Sign in with Apple you can choose to hide your email.

Limited technical data. A one-way hash of your network address when you start a circle, to prevent abuse. Standard server logs kept by our hosting providers.

Audio

While a table is running, the table phone (and any phone lending its mic) streams audio to our speech-to-text provider, Deepgram, to turn it into words. The table phone also keeps the last three minutes of audio in its memory to recognize voices; it is never written to disk or uploaded as a recording. Short clips are sent to our server to compute voice fingerprints and are discarded immediately after.

Who helps us run Pardon?

• Deepgram: turns speech into text.
• Anthropic: writes "Catch me up" summaries from recent transcript text (no audio).
• Supabase: our database, sign-in, and live sync between phones.
• Apple and Google: only if you choose to sign in with them.
• Vercel: hosts our website and server.

They process data only to provide these services to us.

Recording and consent

Some places require everyone in a conversation to agree before it's recorded or transcribed. Pardon? shows a clear listening indicator, and asks the person starting a table to let everyone know it's on. Please do.

Children

Pardon? isn't directed to children under 13, and we don't knowingly collect their information. If children are at the table, their words may be captioned like anyone else's. Don't save a child's voice fingerprint unless you are their parent or guardian and agree to it.

Your choices

• Forget a voice: press and hold a person's name, or turn off "Recognize my voice" in Settings.
• Delete my data: Settings → Delete my data removes you and your voice from the circle.
• Delete a circle: the person who started it can delete the whole circle, including everyone's transcripts and voices.
• See and download your data: the Account page shows everything we store about you and lets you download it, clear it, or delete your account.

Questions or requests: berk@itsmanmade.studio.

Not a medical device

Pardon? is a captioning tool. It isn't a hearing aid and isn't intended to diagnose or treat hearing loss. Captions can be wrong. Don't rely on them where accuracy is critical.