Privacy and your data

What leaves your device

The audio of each dictation or note. When you finish speaking, VoiceSnap Pro uploads the recording, with your language setting and your personal vocabulary, over an encrypted connection to our server in Ireland, which passes it to OpenAI for transcription and sends the text back. Cloudflare sits in front of our server and handles that traffic on the way.

For a dictation, the audio is held in memory for the seconds it takes to transcribe and is then discarded — not saved, not archived, not reviewed by anyone. For a voice note, the text and its recordings are stored, encrypted, in your account so they reach your other computers, along with up to 50 earlier versions of the text. They stay until the note is deleted for good, and the recordings only for as long as your Keep recordings setting says. Notes you type by hand in VoiceSnap Pro are saved to the account the same way. Re-transcribing a note sends its recordings through our server to OpenAI again, the same way.

With AI titles on — the default — a note's text also goes from our server to OpenAI to write the note's title: when a voice note is made, after a re-transcription, or when you choose Regenerate title — the only way a typed note gets one. Its recordings are never sent for this, and dictations never get titles. Turn AI titles off in Settings → AI and no note's text is sent for a title.

With AI tags on — also the default — a note's text and the names of your tags go to OpenAI the same way, once, so it can pick up to three of your tags for the note. It can only choose among tags you made; it never creates one.

With Reminders from your voice on, the sentences around a request such as “remind me tomorrow at 9” in a voice note go to OpenAI the same way, to read the time. The rest of the note does not.

AI actions send text only when you choose one. The text you run it on, from a note or a dictation, and your own action's prompt go through our server to OpenAI, and the result comes back to you. Our server keeps none of it, only a count of the actions you run. Turn AI tags, reminders from your voice or AI actions off in Settings → AI and nothing is sent for them.

Your account is identified by the email address you sign in with, from Google or from the one-time code we email you. To check your account and plan, the app also sends its sign-in session and an identifier for each computer, taken from its hardware ID, which is how the devices on your account are counted. The privacy policy lists everything our server keeps.

What never reaches our server

The list below is not a matter of settings. Apart from the one exception it names, none of it is ever sent to us.

• Your dictation history, including the name of the app each dictation went into. It stays on your computer. The exception: when you run an AI action on a dictation, its text — never the app's name — passes through our server to OpenAI, which our server does not keep. An AI assistant you allow can also read the history on your computer (see below).
• Your shortcut and microphone settings.
• The contents of the field you dictate into, or of any other window on your screen.
• Your keystrokes in other apps. The shortcut listener sees key events so it can notice your shortcut, but it records none of them.

Nothing is used to train AI models

Your audio is not used to train or fine-tune any model, and neither is the resulting text, nor the text, tag names and prompts sent for titles, tags and AI actions. There is no setting to find and no box to untick, because there is no opposite state to opt out of.

This is the one that matters most for people dictating client work, medical notes, legal drafts or unreleased product plans — the material that is confidential is precisely the material a dictation app hears first.

Not training on it is a different thing from not storing it, though. Notes, unlike dictations, are stored in your account on our servers, and our server holds the keys that decrypt them. If a note would contain health details or anything else you would rather not have on a server, dictate it into a document on your own computer instead: the audio is still transcribed through our server and OpenAI, but then discarded, and the text stays with you.

An AI assistant you connect to your notes is the one exception you control: what it reads is handled under its provider's terms, not ours.

Why Accessibility permission, and what it is not used for

macOS treats “type into another application” as a privileged operation and puts it behind the Accessibility switch. It is the same permission a text expander or a window manager needs, and it is the only way to put text into the app you are already using rather than into a window of VoiceSnap Pro's own.

VoiceSnap Pro uses it to insert text and to read the name of the frontmost app — that is what appears on the mic pill and gets stored with each dictation in the history on your computer. It is never sent to us. It does not read the contents of your fields, does not take screenshots, and does not record keystrokes. Setup also asks for Input Monitoring, though the current version does not rely on it. The shortcut listener is not a keylogger: it sees key events but records none of them, and it acts only on your shortcuts, on Escape during a session, and on any other key you press while holding a shortcut, which cancels that recording.

Your notes are yours to keep or delete

Deleting a note moves it to the Trash, where it stays for 30 days, on every computer signed in to your account, so it can be restored. After that, or at once with Delete forever or Empty Trash, it is deleted for good from your computer and from your account, with its earlier versions and recordings — the recordings leave our storage within about 10 minutes, and encrypted database backups expire within about 16 days. Dictations only ever exist on your computer, and so does their Trash: a deleted dictation waits there for 30 days before it is deleted from your computer, and clearing the history deletes every dictation at once, the Trash included.

Keep recordings, in Settings → Voice notes, sets how long recordings are kept at all: forever, 30 days, or not at all. Settings → AI has a switch each for AI titles, AI tags, reminders from your voice and AI actions, which decide whether text is sent to OpenAI for each. Tags and your own AI actions are deleted in Settings, on every computer. You can also turn off Keep dictation history and use VoiceSnap Pro as a pure typing tool with no dictation history at all. To have everything our server holds for your account deleted, use the delete-my-data page linked in the footer.

AI assistants, off by default

VoiceSnap Pro can let an AI assistant on your computer, such as Claude Desktop, search and read your notes. It is off until you turn on Let AI assistants read my notes in Settings → AI assistants, and the switch is per computer. Your dictation history is included only if you turn on the switch below it as well.

When it is on, the assistant starts VoiceSnap Pro in a read-only mode that reads the notes stored on that computer — text only, never recordings or anything in the Trash — and never goes online. VoiceSnap's server is not involved. What the assistant reads is sent to the assistant's provider, Anthropic in the case of Claude Desktop, and handled under that provider's terms.

Diagnostics

The app has no crash reporter and no analytics. Nothing about what you dictate — not the words, not the app you dictated into — is sent as analytics.

What our server does keep is a running total, per account and computer, of seconds transcribed and number of transcriptions, used to count your free allowance and your plan's limits, plus 30 days of request logs that never contain audio, transcripts, note text, note titles, tag names, prompts or AI results. For AI actions it also keeps a count of the actions you run.

For your IT or security team

The short version to pass on: a desktop app that uploads microphone audio over TLS to a transcription endpoint after an explicit user-initiated action, receives text, and stores dictation text locally. Users sign in with Google or an emailed one-time code. Notes, their titles, their earlier versions and their recordings are also stored encrypted in the user's account on the vendor's servers in AWS Ireland, with transcription by OpenAI. While the account's AI titles and AI tags settings are on (the default), note text, never audio, is also sent to OpenAI to generate a short title and to pick up to three of the user's existing tags, whose names are sent with it; tags and the user's custom AI actions are stored encrypted in the account. AI actions, which run only when the user chooses one and can be turned off per account, send the selected text of a note or dictation, up to 20,000 characters, and any custom prompt to OpenAI through the vendor's server, which keeps only a count of actions per account. No background listening and no continuous recording. An optional local MCP server, off by default, lets a user's own AI assistant read the notes on that device; it makes no network connections itself, but the assistant sends what it reads to its own provider.

Two operational notes matter in a managed environment. VoiceSnap Pro needs the microphone open only during a dictation or note the user starts, so it is compatible with policies that forbid always-on recording. And because the library and its note recordings sit unencrypted in the user's profile, full-disk encryption — FileVault or BitLocker — is the control that protects them, alongside your normal endpoint backup policy.