Last updated: 27 September 2026
SayKeys is a keyboard. It can see what you type and hear what you dictate, so it owes you a plain account of what it does with that. This document is written from the app's actual data flows rather than from a template.
Who is responsible. SayKeys is operated by Jonathan David Grey, the data controller for the purposes of the UK and EU GDPR.
- Email: [email protected]
A postal address for written correspondence is available on request by email.
Everything in this document has been checked against what the app actually does, rather than written from a template.
1. The short version
- Dictated audio is sent to speech-recognition companies to be turned into text. We keep no copy. Those companies apply their own retention policies to what we send them — see §4. The one exception is voice samples, which you opt into and which we do store (§2.2).
- What you type is not sent anywhere as raw text. A vocabulary summary is, and only if you switch on AI learning, which is off by default.
- Your clipboard, notes and settings are stored only on your device unless you sign in and enable sync.
- We do not sell your data, do not use it for advertising, and do not build advertising profiles.
- Everything the app stores about you can be deleted on request.
2. What SayKeys collects, and why
2.1 Voice and dictation
| What | When | Where it goes | Kept? |
|---|---|---|---|
| Microphone audio | Only while you are actively dictating | Speech-recognition provider (see §4) | Not by us. The provider's own policy applies to their copy. |
| The transcript | Every dictation | Returned to your device; sent to a language model when polish is on | Not by us. Same for the language model provider. |
| Word and second counts | Every dictation | Your account, if signed in | Yes — usage totals |
The microphone is used only while a dictation is running. SayKeys does not listen in the background and has no wake word.
If polish is set to Natural, Formal or Code, the transcript is sent to a language model to be cleaned up or reformatted. In Raw mode it is not sent to any language model.
The AI tools in the keyboard — rewriting or improving selected text, writing a message from a description, translating between your chosen languages, and turning a spoken request into a prompt for a coding assistant — all work by sending the text you selected or dictated to a language model and putting the result back. That text is processed to answer the request and is not retained by us. If you do not use those tools, no text is sent for them.
2.2 Voice samples (optional, opt-in)
If you choose to build a personal voice model, the passages you read aloud are uploaded and stored so the model can be trained. This is the one case where audio is kept.
- Stored in our private
calibrationsstorage bucket and indexed invoice_samples. - Used only to train an adapter for your own account. There is no shared or pooled voice dataset, and your recordings are never used to improve anyone else's experience.
- Deletable at any time — see §7.
2.3 Typing and personalisation
- On your device, always: words you type, corrections you accept or reject, swipe paths, and a personal vocabulary derived from them. This drives autocorrect and suggestions and stays local.
- Sent off device only if you enable AI learning (off by default): a summary of your vocabulary and phrasing — words and their frequencies, phrases you repeat, corrections you undid — is sent to a language model to refine your personal profile, and the resulting profile is stored on your account. By default this summary contains no sentences you wrote, only counted words and phrases.
- Raw text samples are a separate setting, also off by default. If you turn it on, short excerpts of text you actually typed are included in that summary so the model can read your style directly rather than infer it from word counts. This is the only setting under which sentences you wrote leave your device for personalisation. Turning it off stops it immediately.
- Your personal vocabulary may also be sent with a dictation to bias speech recognition toward names and terms you actually use.
Unless you switch on both AI learning and raw text samples, the raw text of what you type never leaves your device.
2.4 Clipboard, notes, and calendar
- Clipboard: SayKeys keeps a local history of what you copy so you can paste it again. It is stored on your device. It is uploaded to your account only if you turn on clipboard sync, which is off by default and applies to every device signed in to your account. With it on, your pinned text clips are kept for up to 90 days, and the text you most recently copied is shared so your other signed-in devices (for example the desktop app) can paste it; recently copied text is deleted from our servers after 15 minutes. Only text is synced. Text that the app you copied it from marks as sensitive (for example a password manager) is never uploaded. Turning clipboard sync off deletes everything it uploaded.
- Screenshots: if you grant image access, SayKeys notices new screenshots so it can offer them in the clipboard panel. Images stay on your device and are not uploaded.
- Notes: stored on your device, synced to your account only if you enable notes sync. Notes may be sent to a language model to be filed or organised when you use that feature.
- Calendar: if you grant calendar access and use the calendar feature, SayKeys reads and writes events on your device. Calendar contents are not uploaded to our servers.
2.5 Account and billing
If you sign in: your email address, an account identifier, your plan, and — for team plans — the team you belong to and the email addresses you invite.
2.6 Usage and diagnostics
- Usage counters: words dictated, seconds of audio, and which features you used. These enforce plan limits and populate your stats screen.
- Crash reports: if the app crashes, a report containing the error, your app version, device model and account identifier is uploaded so the fault can be fixed. Crash reports do not contain your text or audio.
- Contribution to improving SayKeys (opt-in, "Help improve SayKeys"): anonymised swipe paths and correction examples used to train our models. Off unless you agree to it, and turning it off stops every contribution path at once.
2.7 Notifications and updates
If you allow notifications, SayKeys uses them for reminders you set on your own notes, to tell you about a new feature or a usage milestone, and — on directly-installed builds — to offer an app update. Nothing about you is transmitted in order to show a notification, and notifications can be turned off in Android settings without affecting anything else.
Builds installed outside Google Play can download and install their own updates. Your app version is sent when checking for one, so we can tell whether a newer build exists.
2.8 Support email
If you email support, your message and address are processed — including by an automated assistant that drafts replies — so that we can answer you.
3. What we never do
- We do not sell or rent your data.
- We do not use it for advertising or share it with ad networks.
- We do not read your clipboard, notes or calendar except to provide the features above.
- We do not use the accessibility service to read the contents of your screen. It reports only whether a text field exists and whether one is focused, so the floating microphone knows when to appear.
- We do not record audio outside an active dictation.
4. Who else processes your data
| Processor | What they receive | Purpose |
|---|---|---|
| Supabase | Account, settings, stats, opted-in synced content, voice samples | Sign-in, database, storage |
| Cloudflare | Dictation audio in transit, transcripts | Routing, transcript cleanup |
| Groq | Dictation audio, transcripts | Speech recognition, cleanup |
| OpenAI | Transcripts, vocabulary summaries | Polish, personalisation, vocabulary correction |
| Modal | Dictation audio | Speech recognition on GPU |
| Soniox | Live dictation audio stream | Real-time transcription |
| FUTO | Nothing about you | Source of the downloadable swipe model |
| Google Play | Purchase and subscription records | Billing |
These providers process data on our instructions. They may process it outside your country, including in the United States; where required we rely on standard contractual clauses for those transfers.
Two different things, and it matters which is which. We do not store your dictation audio or transcripts — there is no copy on our servers to hand over, sell, or lose. But once a request reaches one of the companies above, what happens to it is governed by their policy, not ours. Providers of this kind commonly hold requests for a short period for abuse monitoring before deleting them. We do not ask any of them to retain your data, we cannot read it back, and we do not use it for anything after your result comes back — but we cannot promise on their behalf, and we would rather say so than imply a deletion we do not control. Their current terms are the authoritative answer, and they are listed above so you can read them.
5. Legal bases (UK/EU users)
- Contract — running the keyboard, transcribing your dictation, operating your account and plan.
- Legitimate interests — security, abuse prevention, crash diagnostics, and keeping usage counts accurate.
- Consent — voice-model recordings, AI learning, contributing data to improve SayKeys, clipboard/notes sync, calendar and image access. Each is off until you turn it on, and each can be withdrawn at any time without losing the rest of the app.
6. How long we keep things
| Data | Retention |
|---|---|
| Dictation audio | Not retained by us — discarded once transcribed. See the note below on processors. |
| Transcripts | Not retained on our servers; the provider's policy applies to their copy |
| Voice samples (only if you opt in to a personal voice model) | Kept until you delete them or close your account. This is audio of your voice, stored deliberately, and it is the one exception to everything above. |
| Account, settings, stats | Until you close your account |
| Synced pinned clips and notes | Until you delete them or disable sync (pinned clips at most 90 days) |
| Synced recently copied text | 15 minutes |
| Crash reports | Deleted once the fault is understood |
| Diagnostic data collected to answer a specific question | Deleted as soon as it is answered |
About the processors. "Not retained" describes our systems. The speech and language providers in §4 operate their own retention — typically a short period during which requests may be held for abuse monitoring before deletion — and that is governed by their terms, not ours. We do not ask them to keep your audio or text, and we do not retrieve it from them.
7. Your rights and how to use them
You can ask us to give you a copy of your data, correct it, delete it, or stop a particular use of it. Email [email protected] and we will respond within 30 days.
In the app you can, at any time: delete your voice recordings, turn off AI learning, turn off contribution, turn off each sync, revoke microphone, calendar, image, overlay and accessibility permissions, and sign out.
UK and EU users may also complain to their local data protection authority.
8. Children
SayKeys is not directed at children under 13 and we do not knowingly collect their data. If you believe a child has given us data, contact us and we will delete it.
9. Security
Traffic is encrypted in transit. Your sign-in tokens are held in Android's encrypted storage. Voice samples are in a private bucket reachable only with your own credentials, and database rows are protected by row-level security so one account cannot read another's.
No system is perfect; if a breach affects you we will tell you and the relevant regulator as the law requires.
10. Changes
If this policy changes materially we will say so in the app before the change takes effect. The date at the top always reflects the current version.
Questions: [email protected]