Last updated 17 August 2026 · Applies to the Dialect mobile app for iOS and Android
Your voice never leaves your phone. Recordings are transcribed on the device itself and the audio file is deleted immediately afterwards. No audio is ever uploaded anywhere.
The words do leave. The text of what you write is stored on our servers, and sent to Google's Gemini model so the app can find Scripture that meets it and give the entry a name. That is how the app works, and there is no version of it where the text stays only on your phone.
Nothing you write is used to train any AI model, sold, or used for advertising. There is no advertising in Dialect and no analytics tracking of any kind.
You can export everything and delete everything, from inside the app.
Dialect is an independent app. It is currently in closed beta with a small number of invited testers. Questions about this policy, or requests about your data, go to unityaiapps@gmail.com.
Because Dialect is in beta, this policy will change as the app does. Material changes will be noted in the app and the date at the top of this page will be updated.
This is the most important thing in this document, so it comes first.
When you record a prayer, the audio is captured on your device and transcribed on your device, by a speech-recognition model that ships inside the app. It does not call a transcription service. The recording is written to a temporary file only long enough to be read, and is then deleted — on success and on failure alike. Dialect keeps no copy of it, and there is no code path in the app that uploads audio anywhere.
The only thing kept about the recording itself is how many seconds it lasted.
The text — your transcript, and things you write by hand — is sent to our backend and, for some features, on to Google.
Dialect stores your journal on Supabase, a hosted database provider, in a database located in the United States (Oregon). Everything is sealed to your account at the database level, so one account cannot read another's rows.
To find Scripture, name an entry, recognise recurring images, and write the longer look-backs, the text is sent to Google's Gemini models through Google Cloud's Vertex AI, in the United States. Specifically:
| When | What is sent |
|---|---|
| Right after you record, to propose spelling and mis-hearing fixes | The transcript |
| When an entry is saved or edited | The transcript, to turn it into a numeric representation used for search |
| When an entry is saved, to suggest Scripture and give it a name | The transcript, plus your own lexicon terms and season labels as context |
| When you search | Your search text only |
| Weekly look-backs and the long-form Portrait | Up to 60 recent entries, shortened, plus your own written meanings |
Under Google Cloud's terms, Google does not use this content to train or fine-tune its models. Google does cache inputs and outputs for up to 24 hours to speed up responses, and may retain prompts for up to 90 days for abuse monitoring if its automated safety systems flag something. Those are Google's platform behaviours, not ours, and we describe them here so you know they exist.
When the app breaks, a crash report is sent to Sentry, an error-tracking service, so the fault can be found and fixed. This is the only other place any data goes, and it is deliberately kept as thin as possible:
Crash reporting is only active in builds configured for it. If it is switched off, no report is generated at all.
There is no other service. Dialect contains no analytics SDK, no advertising SDK, and no social-media SDK, and nothing tracks what you do in the app. It does not register a push token with Apple, Google, or Expo — the daily reminder, if you turn it on, is scheduled by your phone's own operating system and never touches a server.
We do not collect location, advertising identifiers, browsing history, health data, or your device's contact list (see below).
If you use the "invite from contacts" screen, Dialect reads your address book on your device to show you a list. That list is not uploaded. Only the email addresses you explicitly tick are sent to our server, and only to check whether those people already have a Dialect account so we can offer to connect you. Names are never sent. The addresses you tick are used for that check and are not stored.
You can decline the contacts permission entirely and still use every other part of the app.
Dialect is private by default. Nothing you write is visible to anyone else unless you deliberately share it. There is no public feed and no "make public" setting.
When you do share, please understand what that means, because it is not the same as a link you can un-publish:
This mirrors ordinary life: you cannot un-tell someone something. Share accordingly.
Export. Settings offers a full export of your journal as a readable document, and a complete structured export of your data. Both are produced on your device and handed to whatever you choose — nothing is sent to us.
Deleting one entry removes it and everything derived from it.
Deleting your account. Settings has an account-deletion option. It works in two stages, on purpose, because it is the most irreversible thing in the app:
What deletion does not remove, deliberately. Things you gave to other people stay with them: a dream or intercession they accepted into their journal, and replies you wrote on someone else's shared entry. Those have become part of another person's record, and taking them back would delete from their journal, not yours. Where this happens your name is removed and you appear only as someone who has since left. Text you shared for comment is cleared once its 14-day window has closed.
If you owned a group, the group is handed to its longest-standing member rather than destroyed, so it does not disappear for everyone else.
Your journal is kept until you delete it or delete your account. There is no automatic expiry — the point of the app is that a prayer from two years ago is still there.
The exceptions: an entry shared for comment stops being readable by its recipients after 14 days; invitation links expire; the temporary audio file is deleted within seconds of being transcribed.
Traffic is encrypted in transit and data is encrypted at rest by our hosting provider. Every table is protected by row-level security, so the database itself refuses to return one account's rows to another. Your login session is held in your device's secure keychain. Passwords must be at least 10 characters.
No system is perfect, and Dialect is a beta built by one person. If you find a security problem, please report it to the address in section 1 rather than disclosing it publicly, and it will be taken seriously.
Wherever you live, you can see your data (it is all in the app), export it, correct it, and delete it, without asking anyone. If you are in the EEA, the UK, or a US state with a privacy statute, you also have the right to ask what is held about you and to object to processing; write to the address in section 1 and you will get a real answer. Our lawful basis for processing, where that concept applies, is the performance of the service you asked for.
Data is processed in the United States, by us and by the three services named above (Supabase, Google Cloud, and Sentry). If you are outside the United States, using Dialect means your data is handled there.
Dialect is not directed at children under 13 and we do not knowingly collect their data. If you believe a child has created an account, contact the address in section 1 and it will be removed.
Dialect is currently a closed beta. Beta software can lose data. Please do not treat it as the only copy of anything precious — the export in Settings exists for exactly this reason, and it is worth using occasionally.
Questions, requests, corrections, or concerns: unityaiapps@gmail.com