play.talk
privacy
This translation is here to help you read the page. The English text is the one that applies.
Draft, September 2026 — to be read through by a grown-up who knows the law before launch. Plain words, on purpose.
who we are
play.talk makes a talker for a child who is learning to talk: a single-screen app that turns typed words into pictures and speaks them. It is run from the United Kingdom. Questions and requests go to the address at the foot of every grown-ups' page.
what we keep, and why
- A grown-up's email address, to sign you in and to write to you about your talker. No passwords: we send a link and a code instead.
- The child's first name, to name their talker.
- The lines the child says — the words, and when they were said. They are the child's own record, shown as the scrollable history, and they teach the app which word to suggest next. The server keeps the most recent six hundred lines; each of the family's devices keeps its own copy.
- Pictures the family adds — photos from a device, pictures drawn to a family's own description, and the meanings a family gives to words. They belong to that talker and are never shown to another family.
- Play pages a grown-up asks for, the words that describe them, and the fixes asked for afterwards.
- The devices paired to a talker — a name like "iPad", when it was paired and last seen — so a grown-up can see them and forget one.
- How much a talker has drawn, built and spoken each day and month, to keep to its plan.
- Payment is taken by Stripe. We keep a reference to the subscription and its standing; we never see or hold card numbers.
We keep nothing else. There is no advertising, no tracking, no analytics, and no selling or sharing of anything for anyone else's purposes.
who does work for us
- OpenAI draws the word pictures, guesses the next word, writes the play pages, and checks what a grown-up types before it is drawn or built. A photo a family chooses to turn into a sticker is sent there once to be redrawn. Their API does not train on what it is sent.
- Microsoft Azure speaks the child's lines in the chosen voice.
- ElevenLabs makes sound effects for play pages.
- Stripe handles subscriptions and payments.
- Render hosts the service; the email service named in our settings sends the sign-in links.
Pictures of everyday things drawn from a word alone, and phrases spoken in a voice, are shared between families, since they are the same picture and the same phrase for everyone. Nothing a family added or asked for is ever shared.
children
The account belongs to the grown-up, who is responsible for the talker and the child's use of it. The child's words are personal data and are treated as such: kept for the family alone, used only to run the talker, and gone when the talker is deleted. The app shows no advertising and asks the child for nothing.
how long
A talker's data is kept while it is subscribed and for twelve months after the subscription ends, in case the family comes back. A grown-up can delete a talker at any time from the grown-ups' page: it is taken out of service at once and its files are removed within thirty days. Nightly backups roll off within fourteen days.
your rights
You can download everything a talker holds from the grown-ups' page, and delete it there too. You can ask us to correct or erase anything about you, or to explain what we hold. If you think we have got something wrong, tell us first; you can also complain to the Information Commissioner's Office at ico.org.uk.
keeping it safe
Every connection is encrypted. Sign-in links and codes work once and expire within minutes. A device's key is stored only in a form that cannot be reversed, and can be taken back from the grown-ups' page. One cookie signs a grown-up in; there are no others.
changes
If this page changes in a way that matters, we will say so on the grown-ups' page before it takes effect.