Privacy policy · last updated 11 November 2026
It stays on the iPhone.
Your audio and your transcripts stay on your iPhone, and nothing about you reaches us from the app. There is no account and no sign-up. The app contains no analytics code — ours or anybody else's SDK — no telemetry, no tracking, no advertising and no crash-reporting service, and nothing in it sends your audio, your transcripts, your titles or your edits anywhere. They leave the iPhone only when you send them: through the iOS share sheet, or the clipboard.
Two things that sentence does not cover, said here rather than only further down. If you download a Whisper model, your iPhone's address and the details any web request carries go to Hugging Face, which keeps and analyses them under its own policy — and that, not anything we hold, is why the App Store label declares three kinds of data. And if you write to us, we have your message and your address, because you sent them. Neither is audio and neither is a transcript, and both have their own section below.
This covers Plain Said 1.0 for iPhone, bundle
be.grunt.transcriba, and its share extension,
be.grunt.transcriba.share. It is on the App Store, and this page
changes if anything in it stops being true. The name on the Home Screen is
Plain Said while the project, the bundle identifiers and the App Store
record keep the earlier internal name Transcriba — one app, two names,
and nothing turns on which one you happen to see.
The app can make one kind of network request, and only after you ask for it. If you choose to download a Whisper model in Settings, the model files and their tokenizer come from Hugging Face. That is the whole of it: on the default engine, Plain Said makes no network request at all. The one download below says exactly what those requests carry and what Hugging Face sees. Separately, the first time you use a language with the Apple engine, iOS itself fetches that language's files from Apple — the app never sees that connection and cannot add anything to it. That one download is also the reason the App Store's privacy label names three kinds of data; What the App Store label says, below, sets out which and why.
An app is one thing and the system it runs on is another. Everything above is about Plain Said itself. Where somebody else is involved instead — Hugging Face for a model you asked for, Apple for its speech languages and for the store the app comes from — this page says so in plain words rather than letting a promise about the app be read as a promise about the platform.
Who and what
Plain Said is a transcription app for iPhone, published by Grunt SRL, Rue du Couvent 20, 1332 Rixensart, Belgium — company number 0738.575.618, VAT BE 0738.575.618. It records audio, or takes in audio you already have, and turns it into text on the device — with Apple's built-in speech recognition, or with one of OpenAI's Whisper models run on the phone by WhisperKit.
Contact: stray@grunt.be — the mailbox carries another Grunt app's name and is the right address for Plain Said too.
The one download
Nothing but a download, and never without a tap. When you tap Download on a Whisper model in Settings, the app asks Hugging Face for a listing of that model's files, then fetches each one; right after that, at the model's first load, it fetches the tokenizer files for that size if they are not already on the device. Hugging Face may redirect the file downloads to its own storage or CDN hosts, which it chooses. That is every network request Plain Said makes.
What those requests carry is the names of the public files being fetched — which model, and which tokenizer. No audio, no transcript, no title, no account, no name, and no identifier of ours. There is no Hugging Face token: the app sets none.
What Hugging Face receives anyway is what any web server receives: your device's IP address, the standard metadata of an iOS HTTP request — iOS's own user agent, which names the app's executable and build and the versions of the system networking libraries — and the time. If Hugging Face sets cookies, iOS accepts them into the app's own cookie store. We say this rather than writing "nothing of yours leaves", because an address is yours even when a file name is not.
What Hugging Face does with that is Hugging Face's business, not ours. Its privacy policy (huggingface.co/privacy, read 5 November 2026) says it records service usage, the session date and an approximate location, the IP address, and device and operating-system information, and that it uses them to deliver, improve, analyse and secure its services; it treats an IP address as personal information. It keeps information "as long as necessary to deliver the Services", for legal requirements and for security, and sets no period. We have no agreement with Hugging Face and receive none of this. If you never download a Whisper model, none of it happens.
And it does not stop with Hugging Face. Its policy says it shares information with its affiliates and service providers, naming Amazon Web Services, Google Cloud Platform, OVHCloud and Hugging Face SAS among them. Its servers are in the United States and it may store and process data "in the United States or any other country". Its policy names no transfer mechanism for that — no standard contractual clauses, no Data Privacy Framework — and we would rather write down that it is absent than let you assume there is one.
Asking Hugging Face about it. Hugging Face, Inc. is in the United States; its establishment in the EU is Hugging Face SAS, 9 rue des Colonnes, 75002 Paris, and the authority for it is the French CNIL. Its policy takes access and deletion requests in writing, at privacy@huggingface.co. We cannot make one for you: we are not a party to the request and hold nothing to match it against.
Transcribing itself is offline. Once a model is on the phone, or an Apple language is installed, turning audio into text makes no network request — Whisper runs through CoreML on the CPU and the Neural Engine, and Apple's engine is on-device by design.
Apple's speech languages are downloaded by iOS, not by us. The first time you use a language with the Apple engine, iOS fetches that language's files from Apple through its own asset machinery. That is a connection between your iPhone and Apple, under Apple's privacy policy; the app asks for the language and never sees the request.
What is stored, and where
Everything the app keeps lives in its own app-group container on the iPhone, shared between Plain Said and its share extension, plus some short-lived working files in the temporary folders iOS gives each of them. Nothing is on a server of ours, because there is not one.
-
Recordings you made in the app — captured as 16 kHz mono
audio and re-encoded to a smaller
.m4aonce they have been transcribed, or left as they were if that re-encode fails. Kept until you delete the recording. - Audio you imported or shared in — held separately, and deleted as soon as its transcript is done, or if the job fails in a way that will not be retried. Otherwise it is subject to the seven-day rule below.
- Transcripts, titles, your edits and each recording's own settings — in the app's database. Deleted when you delete the recording, which takes its audio with it.
- Whisper models and their tokenizers — kept until you delete the model in Settings.
- Two small bookkeeping files, recording which transcription jobs are claimed and how long an import has been waiting. They hold no content.
- Settings — the engine, the default language, the chosen model — in the app group's preferences, with one per-model flag the app keeps in its own preferences so a model that crashed while loading is not loaded the same way twice.
- Working copies, briefly. A file arriving through the share sheet lands in the extension's temporary folder before being copied in. A normalised copy of the audio being transcribed, and the audio being re-encoded, live in a temporary folder while the job runs and are removed by it. A transcript you export is written to the app's temporary folder for the share sheet to pick up, and the app does not delete that one — iOS clears temporary folders in its own time, so a copy of an exported transcript can sit there after you have sent it.
Unfinished imports: the seven-day rule
Audio you import is deleted once it has been transcribed. If an import is never transcribed, Plain Said removes it automatically once it has waited at least seven days — the next time you open the app, or a later time if it cannot then. That deletion takes the whole row — the audio, the title, and any partial transcript that had been produced.
The exceptions are the part worth reading, because each of them keeps your data rather than removing it:
- It is not a timer that fires on its own. Nothing is deleted while the app is closed; the check runs when Plain Said comes to the foreground. If you never open it again, nothing is deleted automatically at all.
- The clock starts when you first open the app after the import, so something shared in while the app was shut still gets its full seven days.
- It is measured on the device's own running clock, not the calendar, so changing the date does nothing to it. A restart can only lengthen the wait, never shorten it.
- It is not promised for any particular opening. An import that is due is skipped that time — and tried again at a later opening — if it is still being worked on, if the bookkeeping or the database cannot be read, or if the deletion itself fails. Every one of those skips keeps your data rather than tidying it away early.
None of this applies to recordings you made in the app. Those are kept until you delete them.
Backup and sync — and what their absence costs you
Your recordings, imports, transcripts, models and the bookkeeping files are excluded from iCloud Backup. Each of those five folders is flagged, and the flag is read back to confirm it took; if the app cannot confirm it, it refuses to write your data there at all rather than write it somewhere that would be backed up. There is no iCloud sync and no CloudKit: the app's entitlements carry nothing but the shared container.
The exception, stated rather than left to be discovered: your settings may be backed up. The container itself is flagged as a best effort only, so the preferences — engine, language, chosen model — can ride an iCloud backup the way any app's preferences do. They hold no audio and no text.
And the cost of all this is real. Your recordings and transcripts having no second copy means no copy comes back: delete the app, or move to a new iPhone, and they are gone. Exporting what you want to keep is the only copy there will be. The settings may survive, which is no consolation — they are the engine, the language and the chosen model, not a word of what you recorded. We would rather say that plainly than let "it never leaves your phone" be read as "it is safe".
Permissions, and the one you are asked for
The microphone is the only permission Plain Said asks you about, the first time you start a recording. iOS shows the app's own words for why: that it records audio so it can transcribe it on this iPhone, and that it never uploads your recordings.
Transcribing asks for nothing. The app and the extension both carry a speech-recognition usage description, and neither ever requests speech-recognition authorization — the Apple engine works on-device without it, so no prompt appears. The description is there in case iOS ever shows one; today it does not.
Not requested, at all: location, contacts, photos, camera, notifications, Bluetooth, local network, health, and app-tracking permission — the last because there is no tracking to ask about. The store label's Location line does not contradict this, and What the App Store label says, below, is why: it is what Hugging Face derives from an address, not anything iOS hands the app.
Two things run in the background and neither is a permission prompt. A recording keeps going when the screen locks, which is what the app's audio background mode is for; and a long transcription carries on through a continued-processing task, which iOS shows you as a progress indicator.
The share extension
Sending one audio file to Plain Said from another app's share sheet — Voice Memos, Files, Mail — runs the extension, which shares the same container as the app. With the Apple engine it transcribes the file there and then, inside the share sheet, and saves the transcript. With Whisper it leaves the file in the inbox and the app transcribes it the next time it runs. Either way the file is copied in from the extension's temporary folder, and nothing about it goes anywhere else.
What reaches us
Nothing, from the app. No data of yours arrives here, because there is nothing in the app that sends any and no service of ours to send it to.
Diagnostic logs stay on your device. The app writes to iOS's own
log — under the subsystem be.grunt.transcriba — which stays on the
iPhone unless you deliberately share a system diagnostic with somebody.
In the TestFlight beta, crash reports reach us automatically — regardless of Share With App Developers in your iPhone's privacy settings. That is Apple's channel, not something the app switches on or can see. That setting governs App Store customers. TestFlight stays automatic.
Apple tells us about the store and about the beta. TestFlight shows us tester status, device, sessions, crashes, and any feedback you send through it. A public link enrols testers anonymously, so no name or email address comes with that — unless you put your email address into a piece of feedback, which Apple then shows us on that feedback alone. What App Store Connect adds is Apple's own measurement of its own store rather than anything the app sends, and Apple's terms govern it rather than ours.
When you write to us
The support page asks you for things, so this page has to account for them. Writing to us is not the app doing something — it is you choosing to send an email — but it is still us holding personal data of yours, and everything above would be misleading without this section.
- What arrives: your email address, the name your mail program puts on it, and whatever you choose to put in the message. The support page asks for the iPhone and the iOS version, the engine and model, the language, where the audio came from, and what you did before it went wrong.
- What we ask you not to send: the audio and the transcript. We do not need them to start looking. If a particular file turns out to be the only way to reproduce something, we will say so and you can decide then.
- Why we have it: to answer you, and nothing else. The basis is legitimate interest, Article 6(1)(f), in answering a question you chose to ask us.
- How long: your message and our replies are deleted twelve months after the conversation ends.
-
Who else sees it: Proton Mail, which carries
grunt.be's email — Proton AG, Route de la Galaise 32, 1228 Plan-les-Ouates, Geneva, Switzerland, under its published Proton Mail privacy policy. There is nothing from the app to join it to, because the app sends us nothing. - What Proton can and cannot see, precisely. Proton's Mail policy says it does not have access to encrypted message content and cannot decrypt it on its servers or in its backups. The same policy says that, because of how SMTP works, it does have access to the metadata: sender and recipient addresses, the IP address an incoming message came from, attachment names, the subject line, and the times a message was sent and received. And a message you send us from an ordinary mail account is not end-to-end encrypted — Proton's policy says inbound mail from external providers is scanned for spam in memory and then encrypted and written to disk. So your subject line and the fact that you wrote are not secret from Proton. We would rather tell you which is which than let "encrypted mail" cover both.
- Switzerland, and why there is no extra paperwork for it. Proton AG is outside the EU, so this is a transfer and has to be accounted for like the one to our host further down this page. The European Commission lists Switzerland among the countries it has recognised as adequate, which it says means personal data can flow there "without any further safeguard being necessary". That recognition is the safeguard, and for once there is genuinely nothing further to show you.
What the App Store label says
The store's privacy label names three kinds of data, and this is the page that explains them rather than the place you find them. It declares Location (coarse), Usage Data (product interaction) and Diagnostics (other diagnostic data), each marked as linked to you, used for app functionality and analytics, with tracking set to No.
All three are the model download, and nothing else is behind them. They are what Hugging Face records when the app fetches a Whisper model you asked for, and they are declared because the app is what makes those requests. None of it arrives here. Download no model and none of it happens at all — on the default engine the app makes no network request.
The Location line is the one worth reading twice. It is not a location the app asks for or is given: Plain Said requests no location permission and iOS never hands it one. It is the approximate place an address implies, which Hugging Face works out at its end from the request arriving. Diagnostics is the same shape — the device and operating-system details any web request carries, which Hugging Face says it may use to diagnose and fix its services — and Usage Data is the fact of the download itself.
Inside the app, the privacy manifests say the machine-readable half. The build submitted to the store declares those same three types in the app's own manifest, linked to you and not used for tracking; the share extension's declares none, because the extension makes no network request at all. Neither names a tracking domain, because there is none.
Retention and deletion
Deleting a recording deletes its audio and its transcript together. Deleting a Whisper model in Settings removes the model and its tokenizer. Deleting the app removes the container and everything in it — the recordings, the imports, the transcripts, the models, the bookkeeping and the settings.
Two things are not ours to delete, and both of them are somebody else's: a crash report, which is held in Apple's system, and whatever Hugging Face logged when you downloaded a model, which is held under its own policy and its own retention. What is ours to delete: a message you sent us, and, in App Store Connect, you as a tester and the feedback you sent through TestFlight. Asking is enough. The app itself sends us nothing.
Third parties, ads, children
No advertising, no third-party analytics kits, no data brokers, no sale or sharing of personal information, no accounts, no in-app purchases and no subscriptions. Hugging Face is the only third party the app itself ever talks to, and only if you download a Whisper model. Three others stand around the app rather than inside it, and each has its paragraph on this page: Apple, for the store and for the speech languages iOS fetches; Netlify, which hosts this website; and Proton Mail, which carries our email, if you write to us.
Children. Plain Said is rated 4+, which is Apple's lowest age rating and means it holds nothing unsuitable — not that it is aimed at children. There is no account to make and nothing of a child's reaches us. The one thing a model download does not spare a child is the request itself: it shows Hugging Face an address and the same details as anybody else's.
What is inside it
One third-party library: WhisperKit 1.1.1 by Argmax, Inc., under the MIT licence, linked into the app. Its downloader is adapted from Hugging Face's swift-transformers, under Apache 2.0. Nothing else of anyone else's is in the binary.
The Whisper models are not shipped with the app and are fetched only if you ask
for one: the CoreML conversions come from Argmax's
argmaxinc/whisperkit-coreml, carrying an MIT licence tag, and the
tokenizer files from OpenAI's openai/whisper-<size>
repositories, carrying Apache 2.0. OpenAI's Whisper code and weights are MIT.
Plain Said uses no encryption beyond the HTTPS that iOS itself provides for that one download.
This website
The same standard applies to grunt.be itself, including this page and
the other Plain Said pages. They are plain static files.
- No cookies. Nothing is written to your browser's storage, so there is no consent banner — there is nothing to consent to.
- No analytics and no advertising. No Google Analytics, no ads, no pixels of any kind, no third-party tags.
-
No third-party requests at all. The fonts are served from
grunt.berather than loaded from a font service, and so is every image, stylesheet and script. These pages carry a content security policy that forbids the browser from connecting anywhere else. - Nothing to submit. No forms, no newsletter, no sign-up, no account.
One honest caveat: the host sees your address. This site is served by Netlify, Inc., 512 2nd Street, Suite 200, San Francisco, CA 94107, United States — the address Netlify gives for itself as the data importer in its own Data Processing Agreement. Like any web host it receives the IP address of every request, because that is how the page reaches you. There is no analytics product on this site, no tag and no pixel, and nothing from the app to join a visit to.
The basis for that is the same legitimate interest — in having a website that can be loaded at all, which is not possible without an address to send the page to. Netlify is a United States company, so this is a transfer out of the EU and the safeguard has to be named. For data passing through a customer's site Netlify acts as a processor, with the customer as controller, so for this site we are the controller and what Netlify processes on our instruction is the serving of a page.
The safeguard, as Netlify's own agreement sets it out. Netlify's Data Processing Agreement (its own last-updated line reads 9 June 2026) says that where a transfer of customer data protected by the GDPR to Netlify is a restricted transfer, it "shall take place on the basis of the EU-US Data Privacy Framework", and that where the services are covered by more than one transfer mechanism it is the Data Privacy Framework that applies. The European Commission's standard contractual clauses under Implementing Decision (EU) 2021/914 are the fallback rather than a second concurrent safeguard: the same agreement applies them only to a restricted transfer that "is not subject to the EU-US DPF", and in that case the module that applies is Module Two, because the customer is a controller. So the Framework is the basis today and the clauses are what the transfer falls back on if it ever stops being. Netlify's privacy notice is where the certification to the U.S. Department of Commerce that the Framework rests on is stated.
How long, from the same agreement. Netlify processes customer data for the duration of the Agreement. When our access to and use of the Services ends, Netlify either permits us to export the data at our request or deletes it in accordance with the Service's capabilities; afterward, it deletes customer data remaining on our behalf under its deletion policies. The DPA does not say that taking this site down by itself triggers deletion.
Netlify's privacy notice governs processing for which Netlify acts as controller and limits its processor-data exclusion to enterprise customers. For Netlify's processing on our behalf, the relevant document is the DPA: it forms part of Netlify's self-serve agreement and expressly includes visitors to customer websites hosted by Netlify and IP addresses.
GDPR
Grunt SRL is established in Belgium, and the General Data Protection Regulation (EU) 2016/679 applies.
Everything the app does with your audio and your text happens on your own device and none of it is transmitted to us, so there is no record here to access, correct, port or erase. You keep control through the app itself: what it holds is yours to read, edit, export or delete at any moment, and deleting the app ends it entirely.
What we can hold at all is short: a message you sent us, and what Apple passes on — public-link tester data, TestFlight crash reports and feedback, plus App Store information. We would rather name it than call the whole picture anonymous: Apple's crash reports carry a device identifier, and an online identifier is personal data under Article 4(1) whether or not a name sits beside it. We ask nobody for a name or an address, and there is no account anywhere in this. TestFlight crash reports arrive regardless of Share With App Developers, and Apple can take a day to pass tester metrics on, so leaving the beta does not stop what is already on its way. We can delete you as a tester and your TestFlight feedback; the rest has to be asked of Apple. If you think we hold something of yours in spite of all this, ask, and we will tell you honestly what we find.
The one processing this page cannot describe as purely local is the model download, and it is worth being exact about it. It happens only when you ask for a model, it is how the file you asked for reaches you, and the recipient is Hugging Face, Inc. in the United States together with whatever hosting it routes the download through. We are not a party to it and we receive nothing from it; for anything Hugging Face holds about that download, Hugging Face's own policy and its own process are the route — privacy@huggingface.co, or the CNIL, the authority for its establishment in Paris.
The lawful basis for that one piece of processing, named: our legitimate interest in providing the feature you asked for, Article 6(1)(f). The purpose is to deliver the model you chose, at the moment you tap Download, because those requests are the only way the file can reach you. Nothing else in the app needs a basis set out here, since nothing else involves anybody but you and your iPhone.
Your rights, and the one that actually bites here. Under the GDPR you have the right of access to personal data concerning you, and the rights to rectification, to erasure, to restriction of processing, to data portability, and to object to processing based on legitimate interest — which is the basis named above, so that last right is the live one.
How to object, and what an objection can actually reach. Write to stray@grunt.be and say what you object to. Under Article 21 we have to stop the processing you object to unless we can demonstrate compelling legitimate grounds that override your interests, rights and freedoms — and if we ever thought we had such grounds we would write them out to you rather than carry on in silence. Being exact about what that reaches here matters more than sounding reassuring, so:
- Anything we or a processor acting on our behalf holds — this can include messages you send us, Apple-provided TestFlight tester, crash or App Store information we receive, and visitor data such as IP addresses processed by Netlify for this site. You may ask us to access, correct, erase or restrict applicable data, provide a portable copy where applicable, or object to its processing. We will check what we control and ask our processor for reasonable assistance where necessary. We may need enough information to identify your data; if we cannot identify it or a right does not apply, we will explain why.
- The model download — there is nothing ongoing for us to stop, and we are not going to describe a switch we do not have. The requests are made by the app on your iPhone when you tap Download. There is no account and no server of ours in it, we receive nothing from it, and we have no way to connect an email address to an installation, so we could not block it for you on request even if you asked us to. To avoid these requests, do not download a model. If a model is already installed, deleting it in Settings, or deleting the app, removes the local copy; it does not undo the completed download or erase data already received by Hugging Face. Do not start another model download if you do not want further requests. On the Apple engine the app makes no network request at all.
- What Hugging Face does with what it received — that is Hugging Face's own processing under its own policy, and the rights you have over it are exercised against Hugging Face, not through us: privacy@huggingface.co, with Hugging Face SAS, 9 rue des Colonnes, 75002 Paris as its establishment in the EU and the CNIL as its authority. We are not a party to that processing and hold nothing from it, so we can point you there and no further.
Where there is nothing to act on we will tell you that plainly rather than let silence stand in for a refusal. We make no automated decisions about you and build no profile of you.
Questions can go to stray@grunt.be. You also have the right to lodge a complaint with a supervisory authority: ours is the Belgian one, the Autorité de protection des données / Gegevensbeschermingsautoriteit (dataprotectionauthority.be), and you can equally go to the authority of the country you live in.
Changes
If any of this stops being true, this page changes. First published 5 November 2026 and revised three times: on 7 November 2026 the App Store label, this section's siblings on support email and on this website, and the rights above; on 10 November 2026 what TestFlight gives us; on 11 November 2026 the app reached the App Store. For Plain Said 1.0.
Contact
Grunt SRL, Rue du Couvent 20, 1332 Rixensart, Belgium. Questions about this policy, or about anything the app does with what you give it: stray@grunt.be.