Legal

Privacy Policy

Last updated: September 8, 2026

Verbatype is in public beta on iPhone (TestFlight) and available as a direct download on Mac. This policy describes what the shipping apps do today. We update it when the product changes.

Overview

Verbatype turns speech into text. Recognition can run entirely on your device, with your audio staying there, or through an optional cloud pass that sends audio and text to named third-party providers for a more accurate result. On iPhone the cloud pass is off until you turn it on; on Mac it runs automatically whenever you're online. This policy says what data exists, who receives it, how long we keep it, and how to delete it.

We do not use your dictations to train AI models. See "We do not train on your dictations" below.

Who we are

Verbatype is made by NMC Apps. Questions about this policy or your data: [email protected].

Your account: no sign-in, one account per device

There is no sign-in, no email address, and no password. On first launch the app generates a random identifier and stores it in your device's Keychain. It registers that identifier with our server, and that identifier is your account: it is what your purchases, word credits, usage meters, custom dictionary, dictation history, and weekly reports are attached to.

The identifier is not synced between devices. Your iPhone and your Mac are two separate accounts, with separate history, separate dictionaries, and separate purchases — unless you link them, described below. The identifier survives reinstalling the app on the same device, because it lives in the Keychain rather than in the app's own storage.

We do not ask for your name or email, and the identifier is a random value we cannot connect to your real-world identity. If you email us, we will know your email address.

A second, separate device identifier travels with history sync batches, so we can tell one device's dictation records apart from another's. Per-call usage telemetry (described below) carries your account identifier itself as the device value, not a separate identifier.

Linking devices to share a subscription

You can link a second device to your account so it shares your subscription (or other active entitlement) and your word-credit balance — for example, linking a Mac to an iPhone that already has an active subscription. There are two ways a link is created. The deliberate way: you generate a short code on one device and enter it on the other. The automatic way: while your iPhone is on a paying plan — an active subscription, a positive word-credit balance, the BYOK unlock, or an active trial — it publishes a link through Apple's iCloud Keychain — the same private mechanism this app already uses to sync your API keys (see "Bring your own keys" below) — and a Mac signed into that same iCloud Keychain redeems it silently the next time you launch the app there, with no prompt. The automatic path only ever reaches a device signed into your own iCloud Keychain, and it stops working once you're no longer on a paying plan, because the app deletes the automatic-link data at that point. Either way it happens, you can see every device linked to your account in the app at any time, and remove any of them — see the next paragraphs for what a link shares and how to remove one.

A link shares billing only. It shares which plan you are on and your word-credit balance, so the linked device is not treated as free. It does not share your dictation history, transcripts, audio, or custom dictionary. A linked device cannot read anything stored under the account it is linked to, and never learns that account's identifier.

Up to three devices can be linked to one account. Either side can remove a link at any time — from the device that is linked, or from the account it is linked to — and the change takes effect immediately. Deleting an account (see "Delete account" below) also removes any links to or from it.

What we hold

• Audio you dictate. Recorded so it can be transcribed. On the on-device path it never leaves your device. On the cloud path it is sent to Groq. Audio recordings are kept on your device; we do not store your audio on our servers. The app also sweeps its own local recordings automatically, roughly a day after they were made, once they've been processed.
• Transcribed text. The result of your dictation, including corrections you make. Stored on our server while history sync is on.
• Your account identifier and a device identifier, as described above.
• A weekly usage meter. Seconds of audio processed in the cloud, number of text calls, and number of words. Used to apply the free-tier limits and to draw down word credits.
• Per-call usage telemetry. For each cloud call: the provider, the model, your tier, token counts, audio seconds, and latency. This is sent even when history sync is turned off, because it is how usage is metered and billed. It does not include the text or audio of the call.
• Custom dictionary terms. Names, jargon, and spellings you add. Stored on our server, attached to your account, so the app can restore them. On the cloud path they are sent to Groq as a recognition hint.
• Weekly speech reports. If you generate one, the report itself (the text of the analysis and its suggestions) is stored on our server.
• A push notification token. Only after you grant notification permission. Used to send notifications to your device. It is not linked to your account identifier in our database — it lives on its own and is removed independently, not as part of account deletion.
• Keyboard diagnostics. Not collected. A developer-only diagnostic mode exists in internal builds — where it can record where you tapped (coordinates and key geometry) and memory samples to debug layout and touch problems, never the characters you type — but it is disabled in every released version, on every platform.
• A hashed IP address for failed device-link attempts. If a device-link code is entered incorrectly, we store a SHA-256 hash of the attempt's IP address for a short time, solely to rate-limit repeated guessing. A successful link stores nothing here.
• Purchase records from Apple. See "Payments" below.

We do not collect your contacts, photos, or location, and we do not collect anything from other apps.

How your dictation is processed

On-device recognition (the default)

Recognition runs on your iPhone or Mac using WhisperKit on Apple's Core ML and Neural Engine. On this path your audio is turned into text on the device and is not sent anywhere. It is unlimited, free, and works with no network connection.

This applies to the on-device path specifically. If you use the cloud pass, data does leave your device — see below.

Cloud refinement

On iPhone, cloud refinement is off by default. The first time it would apply — the first time you're online and about to dictate — Verbatype asks whether to turn it on, and no recording is sent anywhere until you say yes. You can change your answer at any time in Settings → "Data & Privacy", and on-device recognition keeps working exactly the same either way, online or off.

On Mac, cloud refinement runs automatically whenever you're online — there is currently no separate on/off control for it on that platform. On-device recognition still works without a network connection either way; going offline is the only way to keep a Mac dictation entirely on-device today.

When cloud refinement runs, Verbatype sends your dictation to two providers to produce a more accurate result:

• Groq receives your audio and runs speech-to-text (whisper-large-v3-turbo). If you have custom dictionary terms, those terms are sent to Groq with the audio as a recognition hint. Groq can also receive your transcribed text for punctuation and phrasing normalization: it is one of two providers we use for that step, and which one handles a given request depends on provider availability and, if you supply your own API keys (see "Bring your own keys" below), on your own key setup.
• Google Gemini can also receive the transcribed text for the same normalization step — punctuation, casing, spoken-to-written phrasing. Gemini does not receive your audio.

By default these requests go through our own server, a Cloudflare Worker, which relays them, meters your usage, and pays the providers. We send each provider only what is needed to fulfil that request, under our own commercial agreements with each of them, and we do not authorize either one to use your data to build advertising profiles. We select processors we believe handle data with protections consistent with this policy, but we do not control what Groq or Google do on their end beyond fulfilling the request — their own terms and privacy policies apply to that leg, and we encourage you to review them if you want the details of how each vendor itself handles a request once it reaches them.

Weekly speech reports

If you ask for a weekly speech report, a week of your dictation text is submitted to Google's Gemini Batch API to produce the analysis. This is text only — no audio. The resulting report is stored on our server under your account.

Bring your own keys (BYOK)

If you buy the BYOK unlock, you can enter your own Groq and Google API keys. On that path your device calls Groq and Google directly. Your audio and text do not pass through our server, and we never see them.

Your API keys are stored in your device's Keychain, in a keychain group shared between the Verbatype apps, and Apple's iCloud Keychain synchronizes them to your other devices. They are stored on your devices, not on our servers. We never receive them and we cannot read them.

We still receive metadata on the BYOK path: your account identifier, entitlement checks, history sync (if it is on), and the per-call usage telemetry described above, labelled as BYOK usage. Your provider bills you directly for the usage.

Model downloads

The speech-recognition model is not bundled in the app. Your device downloads it once, about 470 MB, from Hugging Face (the argmaxinc/whisperkit-coreml repository). The optional Qwen text-normalizer model is a separate download of about 470 MB, served from our own storage on Cloudflare R2.

These are downloads to your device. They do not send your dictation anywhere.

We do not train on your dictations

Your audio and your text are not used to train or fine-tune AI models, and are not used for any other machine-learning purpose. They are processed to produce your transcription, to sync your own history to you, and to generate a report when you ask for one. That is all.

We do not sell your data, and we do not use it to build advertising profiles.

Sub-processors

These are the third parties that may receive your data, and what each one is for. This is the complete list.

Payments

All purchases go through Apple's In-App Purchase system. There is no other payment processor and no billing intermediary. We never receive your card details or your Apple ID.

What Apple sends us about a purchase: the product identifier, transaction identifiers, the purchase and expiry dates, the environment (production or sandbox), the app's bundle identifier, and later events such as renewals, revocations, and refunds. Apple signs these payloads. We verify the signature on our server and retain the signed payload as the record of the transaction.

The keyboard and "Full Access"

Verbatype includes a custom keyboard extension for iOS. iOS requires "Full Access" for any keyboard extension that needs to talk to its host app. Verbatype's keyboard uses that permission only to talk to the main Verbatype app over a private on-device channel, so it can show and edit the text you dictated. The keyboard makes no network requests of its own, and does not log, read, or transmit what you type.

This website

This site sets no cookies and runs no analytics or advertising scripts of our own, and we do not track you across sites. It is hosted on Cloudflare Pages, which — like any Cloudflare site — automatically injects a small script that de-obfuscates the email addresses printed on this site, and sends the browser's own network-error reports to Cloudflare (a standard Network-Error-Logging header); neither is something we configured, and neither sends anything about you to us. Our own code's only network calls are submitting the launch-updates form and the Mac download link, which streams the app installer and stores nothing about you. If you submit the launch-updates form, we store the email address you entered, an unsalted SHA-256 hash of your IP address, and which form on the page you used; that hash is written but is not currently used to detect or block duplicate submissions (submissions already dedupe by email address). To be removed from that list, email [email protected].

How long we keep things

Kept until you delete it. We do not run a scheduled deletion of your account data, and we do not set an expiry on your dictation history, dictionary, reports, or usage records. They stay until you use one of the controls below, or ask us to remove them.

The one exception is the purchase and credit records described under "Deleting your account", which we keep after deletion for the reasons stated there.

Your controls

These controls are in the app, under Settings → "Data & Privacy", on iPhone and on Mac.

Cloud refinement (iPhone)

Off until you turn it on; a toggle in Settings lets you turn it on or off at any time. This control does not exist yet on Mac, where cloud refinement runs automatically whenever you're online — see "Cloud refinement" above.

Sync dictation history to server

On by default. Turning it off stops uploading new dictation history. Cloud recognition still works the same way and transcription quality is unchanged. Per-call usage telemetry is still sent, because that is how usage is metered.

Delete history

This deletes your dictation records only. "Delete from server only" removes them from our server and leaves the copy on this device. "Delete everything" removes them from our server and from this device, including local audio recordings. Neither option touches your dictionary, your reports, your usage records, or your account. If sync is still on, new dictations will upload after the deletion, so turn sync off first if you want the deletion to stick.

Delete account

This deletes the account itself. On our server it deletes your dictation history, your custom dictionary, your weekly reports and suggestions, your weekly usage meter, your diagnostics, your per-call usage telemetry, and your account record. On the device it then erases your local history, your local audio recordings, and the account identifier in the Keychain. This cannot be undone.

What is kept after account deletion, and why.

• Purchase and entitlement records, and the word-credit ledger. These are re-pointed to an anonymous placeholder identifier that cannot be used to sign in or to reach any data. We keep them for accounting, for handling refunds Apple sends us after the fact, and to prevent the same purchase being granted credits twice.
• Aggregated daily usage/cost totals. Like the records above, these are de-identified — re-pointed to the same anonymous placeholder identifier — rather than deleted, because they are our only cost ledger. This is separate from your per-call usage telemetry, which is deleted outright; only the daily rollup total behind it survives, detached from you.
• Apple's signed purchase notifications. Apple continues to send these for past transactions, and they are the signed record of what was bought.

These retained records contain Apple's transaction identifiers, the product identifier, the relevant dates, credit amounts, and aggregate usage/cost totals. They contain no dictation content — no audio, no text — and no identifier that we assigned to you.

A push notification token, if you have one, is not linked to your account identifier in our database, so it is unaffected by account deletion either way; it is removed independently when it stops working (for example, when the app is uninstalled).

What deleting your account does to your purchases.

• Any unused word credits are lost. Word credits are consumable purchases; Apple cannot restore them, and neither can we.
• An active subscription and the BYOK unlock are tied to your Apple ID, not to the account we deleted. They come back with "Restore Purchases" on a new account.
• Deleting your account does not cancel your subscription. Cancel it in your Apple ID subscription settings.

Getting a copy of your data

There is no self-serve export control in the app yet. For an export, or for anything the in-app controls do not cover, email [email protected] from the address you would like us to reply to.

No ads, no tracking

The apps carry no advertising and no third-party analytics or tracking SDKs. We do not sell your data to anyone.

Children's privacy

Verbatype is not directed at children under 13, and we do not knowingly collect data from children under 13.

Changes to this policy

We update this policy as Verbatype changes. We update the "Last updated" date above when we do.

Contact

Questions about this policy or your data: [email protected]

← Back to Verbatype.com