Transcribatron — Privacy Policy

Effective date: August 3, 2026

This Privacy Policy explains how the Transcribatron app for iOS and macOS ("the App"), published by Marc Maguire ("we", "us"), handles your information.

"The App" includes everything we ship with it: the iPhone, iPad, and Mac apps, the Transcribatron keyboard, the Share extensions, and the assistant connector on your Mac. Where one of them behaves differently, this policy says so.

The short version

Transcribatron is local-first. Your recordings are transcribed on your device. We do not have accounts, we do not run a server that collects your content, we do not use analytics or advertising trackers, and we do not sell your data. Two things happen without you switching them on: App Store purchase validation, which runs whenever you open the App and never involves your content, and syncing to your own private iCloud, which follows your device's iCloud setting rather than a switch inside the App. Everything else leaves your device only through features you turn on. This policy explains exactly what, and to whom.

Information that stays on your device

• Audio recordings never leave your device through the App. Recording and playback are local.
• Transcription runs entirely on-device (using on-device speech models). No audio is uploaded for transcription.
• Your AI provider API keys are stored in the device Keychain. They are not transmitted to us. This includes any key you add for a cloud voice (text-to-speech) provider.
• Your calendar (optional). If you grant calendar access, the App reads your upcoming events on your device to show you meetings you may want to record, remind you before they start, and — when you choose — add a link back to your recording on the event. Your calendar is read on your device and is never uploaded to us. One caveat: when you link a recording to an event, the event's title, agenda, and attendee list are saved with that meeting — so if iCloud sync is on, they sync to your own private iCloud along with the rest of the meeting (see "Sync to your own iCloud" below). Nothing goes to us either way.
• Document scans (optional). If you use the camera to scan a document so it can be read aloud, the scan is converted to text on your device. The images are not uploaded.

The Transcribatron keyboard and "Full Access" (iOS)

The Transcribatron keyboard is optional and off until you add it yourself in iOS Settings. When you do, iOS asks you to grant it Full Access and shows a strong warning — because a keyboard with Full Access could, in principle, transmit what you type. That warning is Apple's, it applies to every third-party keyboard, and it is worth taking seriously. So here is what ours actually uses it for.

Full Access is what lets the keyboard reach the App's own data and settings on your device — the two share a private App Group container — so that dictation and AI cleanup work in the keyboard the same way they work in the App. If you have chosen a cloud AI provider, cleaning up your dictated text sends that text under your own API key, exactly as described in "Cloud AI processing" below. If you haven't, nothing is sent.

The keyboard never transmits what you type on its own, and it does not log or store it. It contains no analytics, and we run no server for it to report to.

Two buttons can send text off the device, both only when you press them and both only if you have chosen a cloud AI provider: cleaning up what you just dictated, and the wand — which takes the text already sitting in the field you are editing, including text you typed yourself before opening the keyboard, and runs it through that same cleanup.

If you would rather not grant Full Access, don't — dictation and AI cleanup in the keyboard will be unavailable, and nothing else about the App changes.

Information that leaves your device

Everything in this section happens only when you turn the feature on, with two exceptions. Purchase validation (item 5) runs automatically so the App knows what you've bought. iCloud sync (item 2) is active whenever your device is signed in to iCloud, unless you disable it for the App in your device's iCloud settings — there is no separate switch for it inside the App. The App also downloads AI model files the first time it needs them (item 4); that is a download, not an upload of anything of yours.

1. Cloud AI processing — under your own key

If you choose a cloud provider for features such as cleanup, summaries, chat, or reading your content aloud (text-to-speech), the App sends the relevant transcript or document text (never the audio you recorded) directly from your device to the provider you selected, using your own API key and account. Supported providers include Anthropic, OpenAI, Google, and xAI; for cloud voices, OpenAI, Google, and ElevenLabs. On-device voices are also available, and using one sends nothing.

The audio is never sent. Several other things are, and you should know what they are:

• The style instruction you chose, including any custom style you wrote.
• Your custom vocabulary — the terms, the "sounds like" phrases, and any hints you added — travels with cleanup, analysis, and chat requests so the model spells your terms correctly. If you use a cloud provider, your vocabulary list goes with them.
• Your questions, when you chat with a transcript.
• Calendar details of a linked meeting — the event title, the agenda, and attendee names — when you run AI analysis on a meeting you linked to a calendar event. Those attendee names are other people's.
• A recipient's first name, when you dictate into a supported mail app on your Mac. The App reads it from the "To" field of the message you are composing, so the greeting can address that person by name.

That data is handled under the provider's own privacy policy and terms, not ours — we do not receive, store, or proxy it. Most providers' API tiers do not use data submitted through the API to train their models by default, but this is governed by your provider's terms — review them. If you prefer to keep everything local, you can use an on-device AI model instead, and nothing is sent.

2. Sync to your own iCloud (optional)

If iCloud is enabled, your transcripts and related data sync through Apple's CloudKit to your own private iCloud database. This is your Apple account and Apple's infrastructure — the data does not go to a Transcribatron server. Apple's handling of iCloud data is governed by Apple's privacy policy. You can manage or disable this in your device's iCloud settings.

3. Diagnostic logging (off by default, opt-in)

This is off by default and exists for one purpose: so that if you hit a bug, you can help us reproduce it. Nothing is sent unless you switch it on yourself.

When you do, the App sends its diagnostic log — a record of in-app actions (e.g. "started recording", "opened settings"), app and error events, screen identifiers, and limited technical metadata. It never includes your transcripts, audio, or other content. Only an approved list of technical fields is allowed to leave the device; anything not on that list is dropped before sending.

The switch turns itself back off automatically after about one hour, so you cannot leave it running by accident, and there is a cap on how much can be sent in any 24-hour period. You can also turn it off yourself at any moment.

The logs are received by Grafana Cloud, a hosted logging service operated by Grafana Labs, which acts as our processor for this and handles the data under its own terms. Along with purchase validation (item 5), this is one of only two flows in this policy that reach a service running under our account rather than yours — which is precisely why it is opt-in, content-free, and self-expiring.

4. AI model downloads

The App downloads on-device AI model files from Hugging Face, a public model host. (Some are fetched by the FluidAudio library, but they come from Hugging Face too — it is a piece of software we use, not a separate host.) These are downloads of model software, not uploads of your content — no personal information is sent in the process beyond what a normal file download requires (e.g. your IP address, as with any internet request).

5. Purchase validation (RevenueCat) — always on

The App is free to try and then offers a one-time lifetime purchase through Apple's App Store. To check what you've bought and unlock it across your devices, the App uses RevenueCat, a purchase-infrastructure service. This is the one flow you do not switch on: it runs when you open the App.

RevenueCat receives a randomly generated anonymous ID, your App Store purchase and entitlement information, and basic technical details (such as device model and OS version, and — as with any internet request — your IP address). It is used solely to determine what you have purchased.

On iPhone and iPad the request also carries the identifier Apple gives developers for your device (the "IDFV"). It is the same for every app we publish on that device and resets when you delete them all. We do not ask for it, use it, link it to you, or keep it against any record — RevenueCat's software attaches it to each request it makes. On a Mac, nothing of the kind is sent.

RevenueCat does not receive your recordings, transcripts, notes, name, email, or Apple ID. Unlike the AI providers above, this runs under our RevenueCat account rather than one of yours, but it is not linked to your identity by us, and there is no sign-in. RevenueCat's handling is governed by their own privacy policy.

6. AI-assistant connections on your Mac (optional, off by default)

On macOS you can let AI assistant apps running on your own computer read your meetings and dictations, using the Model Context Protocol ("MCP") — for example, to ask an assistant about a past meeting. This is off by default, and full transcript text sits behind its own separate switch even after you turn the connection on.

A separate switch, also off by default, lets an assistant do things rather than only read: transcribe a file you point it at, clean up text, or run an analysis. Those actions use whatever AI provider you have configured, so turning them on means an assistant can cause text to be sent to your cloud provider under your own key without you initiating it in the App each time.

That same switch also covers voice: an assistant can play your recordings aloud, speak text, and ask you a question out loud — recording your spoken answer, transcribing it on your device, and returning the text to the assistant. The recording is made only when the assistant asks and you answer, and the audio does not leave your Mac, but an assistant being able to start a recording at all is the reason this switch is off until you turn it on.

Anything you allow an assistant app to read is then handled under that app's privacy policy, including any cloud AI service it sends it to. You can turn this off at any time in the App's settings.

What we do NOT do

• No user accounts with us.
• No advertising, no ad identifiers (IDFA), no cross-app/website tracking.
• No third-party analytics SDKs (no Firebase, Sentry, Mixpanel, etc.). Our purchase-validation provider (RevenueCat, above) receives purchase data, not analytics or behavioural tracking.
• No selling or renting of your data.
• No training on your content — we could not if we wanted to. Your recordings and transcripts never reach us, so there is nothing for us to train a model on. There is no setting to opt out of, because there is nothing to opt out of.

If any of these commitments ever had to change, we would say so in the App before the change took effect.

Sending your notes somewhere else

Some features write your content to a place you chose:

• Obsidian. If you connect a vault, the App writes each analysis into that vault folder as a markdown file, automatically, as it is produced. Where the vault then goes is up to you — many are synced by iCloud Drive, Obsidian Sync, or Dropbox, and that is outside the App.
• Exports and sharing. Anything you copy, share, export as subtitles, or send to Apple Notes goes wherever you send it, under that service's terms.

We are not in the middle of any of these. We cannot see what you export, and we cannot retrieve or delete it afterwards.

Other people in your recordings

Your recordings will usually contain other people's voices and personal information. On a Mac, if you turn on system-audio capture, a recording also takes in the audio your computer is playing — which on a call means everyone else on it, not only what your microphone hears. Because that data lives only on your device and under your control, you — not we — are the one handling it. You are responsible for recording lawfully (see the Terms of Use), for how you share transcripts, and for honouring a participant's request to delete a recording of them. We cannot do any of those things for you, because we hold no copy.

If you use the App for work, your employer's or client's privacy obligations may also apply to what you record.

App Store privacy declarations

Consistent with the above, the App's privacy manifest declares that any collected data is not used for tracking. It covers Other User Content, Other Diagnostic Data, and Purchase History — each used solely for app functionality and none of it linked to your identity by us.

Data retention and deletion

• Content you create lives on your device (and your own iCloud, if enabled). Deleting items in the App, or deleting the App, removes local copies; iCloud copies are controlled through your Apple account.
• Local diagnostic logs are kept briefly on-device (about seven days) and then removed.
• Diagnostic logs you chose to send us are retained by our logging provider for 14 days and then deleted automatically. They contain no transcripts, audio, or other content.
• Purchase and entitlement data held by our purchase-validation provider is retained under their policy for as long as we use the service.
• Data you sent to a third-party AI provider is subject to that provider's retention practices; consult their policy and account controls.

Because we hold no copy of your content, we cannot recover anything for you. Deleting the App deletes its content from that device. Keep your own backups.

Two small records live in the system Keychain and can survive deleting and reinstalling the App: any API keys you saved, and a note that a free trial was already started on this device. That second one holds a date and nothing else — it is how the trial resists being reset by reinstalling.

Your recording responsibilities

Recording laws vary by location and may require the consent of everyone being recorded. You are responsible for complying with applicable laws when you record people, calls, or meetings. See the Terms of Use.

Children

The App is not directed to children. You must be at least 13 years old, or the higher minimum age required where you live, to use it. We do not knowingly collect personal information from children.

Cloud AI features are 18+. On-device recording, transcription, on-device AI models, and on-device voices are available at the age above. Connecting a cloud AI provider is not: some providers (currently including Google and ElevenLabs) require you to be 18, and Google's terms bar use of its API in an app likely to be accessed by people under 18. That requirement runs between you and the provider whose key you supply — see the Terms of Use section 3.

Your rights

Because we do not collect your content on our own servers, most data-subject requests (access, deletion, portability) are satisfied directly on your device and through your own iCloud and AI-provider accounts. For questions, contact us below. Where you have rights under laws such as the GDPR, CCPA/CPRA, PIPEDA, or Quebec's Law 25, we will honor applicable requests we are able to fulfill.

To be concrete about what we can and cannot do: for your content there is nothing for us to disclose, correct, or delete, because we never receive it. Requests about purchase data should come to us and we will pass them to our purchase-validation provider. Requests about data in your iCloud are handled by Apple through your Apple account.

Withdrawing consent. Every optional flow in this policy can be switched off in the App's settings at any time: cloud AI, diagnostics, the assistant connector, calendar access, and Obsidian. Turning one off stops it going forward. iCloud sync is controlled in your device's iCloud settings, not ours.

Automated decisions. We make no automated decisions about you that produce legal or similarly significant effects. The AI features process text you give them; they do not evaluate, score, or profile you.

Complaints. If you think we have handled your information badly, tell us first — but you can also complain to a regulator. In Canada that is the Office of the Privacy Commissioner, or the Commission d'accès à l'information for Quebec. In the UK, the Information Commissioner's Office. In the EU/EEA, your national data protection authority.

Where your information goes

We hold nothing on our own servers, but two services running under our accounts are in the United States: our purchase-validation provider and, if you switch diagnostics on, our logging provider. What reaches them is described in items 3 and 5 above — no recordings, no transcripts, no notes.

Anything you send to a cloud AI provider goes wherever that provider operates, under your account with them.

Person responsible for privacy

Marc Maguire is the person responsible for the protection of personal information in connection with the App, including for the purposes of Quebec's Law 25. You can reach him at marc@transcribatron.ca.

Third parties referenced in this policy

Changes to this policy

We may update this policy. Material changes will be reflected by an updated effective date and, where appropriate, an in-app notice.

Contact

Questions about privacy, and requests under the rights described above: support@transcribatron.ca.