Privacy Policy — Ownword

Last updated: 8 September 2026

Ownword is a voice-first keyboard for Android. This policy describes exactly what the app does with your data. It is written from the source code rather than from intent, and the claims in it are checked automatically on every build by audit_privacy.py, which fails if the code stops matching.


The short version


What leaves your device

The app contacts exactly one server: the Exchange your account is on, which is where the speech model comes from, where your uploads go, and where the language models you ask answer from. Every request is sent over HTTPS, and every one is authenticated with your API key except the two that ask for a key in the first place - a phone that has not been given one has none to send.

WhatWhenWhy
A photo or video you takeyou tap or hold the camera keyto store it in your account
A file you chooseyou choose to upload itto store it in your account
Usage countsperiodically, if enabledto improve the app
One word you reportonly when you hold it and choose toto fix a correction that failed
What you copied, or the text you selected or typed - or the words of the web page you are on, or the conversation in an AI chat apponly when you ask for a summary: Summarise in the paste key's list, Summarise what I copied or Summarise this page on the AI key's hold, or the speaker key in its Summarise mode - never from a password field. The page's words are read off the screen through Read the screen, only when you have turned it on and only on that tapto summarise it. From the speaker key, the summary is then read aloud, like anything else read aloud
What you dictated, and the project you choseonly when you tap the refine keyto reword it to match that project
A question you dictatedonly when you choose Ask a question and speakto answer it
Words you ask to hear read aloudonly when read aloud is on and you ask for something to be read: an answer in the chat you tap Read aloud under, or every answer if you turned on reading them as they arrive; text you select in another app and tap Read aloud on; the selection or the text in the field you are typing in, when you tap the speaker key - never from a password field; what you dictate with the echo key, as soon as it is written into the field, and the last thing you dictated when you pick a voice or Hear the last dictation again on its hold - never in a password field; text or a web page you share with Read aloud with Ownword - a shared page is fetched from its site on that tap, as Ask Ownword does, and its words are read. The sentence under Try it and the voice samples carry none of your wordsto read them to you in the voice you chose. They are streamed back as sound, and nothing is kept on the server
What is on your screen, when you ask for itonly when you have turned Read the screen on in Android's Accessibility settings, and only when you then ask for a reading - the accessibility button, or the speaker key's holdto read that screen aloud - or, when you tap Summarise this page or the speaker key over your browser, to read the page open in it so it can be summarised or read aloud; or, over an AI chat app such as ChatGPT or Gemini, Summarise this chat or the speaker key, to read the conversation. For a chat it scrolls the conversation up to find where it starts and back down to where you were - the only time it moves anything on your screen. It reads when you ask it to: it never reads by itself, never watches what you do, and keeps nothing about the screen. The words go to your own voice server to be read, like anything else read aloud. Turn it off in Android's Accessibility settings at any time - or from Settings, Read aloud, Read the screen, which opens the steps to turn it on or off
Long text you ask to hear read aloudonly when it is longer than the limit you set (5,000 characters unless you change it) and Prepare long text on the server is onto read it on your server in the background. This one is kept there: the words and their sound stay in the server's voice library, where anyone with a key and the reading's id could play it. Turn the setting off and long text is streamed like the rest, and nothing is kept
A request for a key: your email address, your name, your reasononly when you fill the form in and press Sendso the person who issues keys can decide, set the account up, and reach you about it

Everything above is started by you, except usage reporting. Audio you record is not on this list. It is worked out on the phone by the model you chose, and the route that once sent it to a server was closed in 1.0.362.

A request for a key carries five things and no more: the email address you typed, the name you typed, the one line you typed, the app's version, and a random id for this installation. The address is the only personal detail the app ever sends, it is sent only on that one screen and only when you press Send, and it is used to set your account up and to reach you about it. The install id is made by the app, is not the Android id, and says nothing about the phone; it exists so that asking twice replaces your request instead of queueing behind it. The app says all of this above the button, before you send anything.

About text you type

Nothing you type is sent anywhere in the background, and there is no keystroke log. Four buttons send text, each only when you press it, and each is described exactly below.

Reporting a word that was not fixed

When the keyboard fails to fix a word — you meant and if and a stray key gave you andnif — you can hold that word and choose Report a missed fix. That sends that one word, so the fix can be worked out. Nothing else goes with it, nothing is recorded by typing, it does not work in password fields, and it does nothing at all if diagnostics are switched off. The button says where the word goes before you press it.

An earlier version had one - an AI Text Fix key that sent the field to the server to be rewritten. It was off by default and it has now been removed outright, so the question no longer needs an answer that begins "only if".

Summarising a page

If you copy a page and choose Summarise, what you copied is sent to your own account's language models to be summarised. Only what is on the clipboard at that moment goes, and only because you chose it. The screen says so before the request is made.

The speaker key can do the same when you switch it to Summarise - from its hold, or in Settings → Read aloud. A tap then sends the text you selected, or what you copied, or the text in the field you are typing in, in that order, to your own account's language models; the summary comes back to a screen that shows it and sends it to your own voice server to be read aloud. Its hold names each choice with its first words before anything is sent. It refuses password fields before anything is sent, and nothing about it is automatic.

Refining what you dictated

If you describe a project and tap the refine key, two things are sent to your own account's language models: the last thing you dictated, and the description you wrote for that project. They come back as one reworded version, which replaces what was on the screen.

Asking a question

If you hold the refine key, choose Ask a question and speak, the transcription of that question is sent to your own account's language models to be answered. It is a question you chose to ask, and nothing else goes with it - not the field you were typing in, not your project descriptions.

Text you select in another app

Select text in any app and its menu offers Ask Ownword; the share sheet offers it too. Only when you tap it, the text you selected goes to your own account's language models, and their answer is kept in your AI history on the phone. Nothing else on the page or the screen is read: Android hands the keyboard the selection and nothing more, and the keyboard asks for nothing more.

Share a page from your browser to Ask Ownword and the keyboard fetches that page itself - the one you shared, once, from its own site, over https - and sends its readable words to your own account's language models to be summarised. This is the only time the keyboard reaches any site other than your own server, and it happens only on that tap. The page's words are kept in your AI history on the phone, under the page's title and address, so the chat can be reopened; the page itself is not kept, and nothing is sent to the page's site but the request for it. A web address you type or paste into a chat is fetched the same way, up to two in one message, only when you send that message. And when you tap Summarise this page and the page on screen shows too few words to summarise, the page's address is fetched the same way, once, on that tap.

The keyboard does keep one draft on the device: if you are part-way through writing something and the keyboard closes, the text is held so it can be put back when you return - on its own if you come straight back, otherwise offered with one tap. It is stored encrypted, only ever on your phone, only for an hour, never from a password field, and it is deleted the moment it is restored or you send or clear the message.

About usage reporting

If left enabled, the app periodically sends a summary of how it was used. That summary contains counts, durations and error codes only:

It contains no text of any kind — not words, not transcripts, not even the lengths of individual words. Nothing in it names you, your device, or which apps you were typing in. It carries a random identifier created when the app is first run, which is gone when the app is removed — so it cannot be tied to you, and it cannot even be tied to an earlier install of the app on the same phone.

One thing worth stating exactly. The upload is authenticated with your API key, the same as everything else the app sends, and it arrives in your own storage. So the server can tell which account it came from, in the same way it can for a file you upload deliberately. Nothing inside the summary says who you are, and nothing joins the two — but we would rather write that down than let "it does not identify you" carry more weight than it can bear.

What it is used for is counting: how many installs are active, which version they are on, and whether a feature is being used at all. Never who.

You can turn it off in Settings at any time — it is on the Your data tab, beside the list of what the keyboard stores.


What stays on your device

Stored in the app's private storage, which other apps cannot read:

Automatic backup is switched off. Android would otherwise copy app data to your Google Drive; this app opts out of both cloud backup and phone-to-phone transfer, so none of the above leaves your device that way.

Uninstalling the app deletes all of it.


What is never collected


Permissions

PermissionUsed forIf you refuse
Microphonedictationthe keyboard works; dictation does not
Camerathe camera keythe keyboard works; that key does not
Internetfetching a speech model, uploads, and what you ask the language modelstyping works offline; so does dictation once a model is on the phone
Vibratekey press feedbacktyping is silent
Notificationsthe player - pause, back, stop - while something is read aloud, asked for when you turn read aloud onreading still works; the player does not show at the top of the screen or on the lock screen
Accessibility (Read the screen)reading the words on a screen aloud, including text you cannot selectoff unless you turn it on in Android's Accessibility settings, and everything else works without it
Foreground service (media playback)keeping a reading going with the keyboard closed or the screen lockedgranted at install; nothing is read unless you ask

There is no storage permission. Files you attach are chosen through Android's own picker, which grants access to that one file.


Your data on the server

Files you upload, and what you ask the language models, go to the account you hold with your provider. Audio you dictate does not: it is worked out on the phone. Retention, deletion and access on the server are governed by that account, not by this app, and are configurable there.

This app does not have a separate account of its own, and does not create one.


Children

Ownword is not directed at children and is not designed for use by anyone under 13.


Changes

Material changes to this policy will be published here with an updated date above, and noted in the app's release notes.


Contact

Questions about this policy or your data: me@rnsu.net