Last updated: 30 October 2026
In short: Aparté collects no personal data. There is no account, no server, and your entries — what you write, your moods, your photos, your voice memos, your handwritten pages, your habits — never leave your device. The only things that can ever leave, if you explicitly allow it, are an anonymous crash report or, if you turn on the beta AI feature, equally anonymous technical diagnostics — never the content of your journal. At launch, the app also checks whether an update is required, without sending anything about you; and if you turn on Whisper transcription, it downloads its model once, again without sending anything.
For any question about this policy: contact@zenith-ai.dev.
Nothing. Aparté asks for no account, no email, no phone number. We operate no server that receives your data. We use no analytics and no advertising, and we embed no third-party trackers.
Everything you write — text, moods, emotions, tags — is stored in an encrypted database (SQLCipher, AES-256) on your device. The encryption key is random and never stored in plain form: it is sealed by your master password (derived with Argon2id), by your device's secure hardware if you enable biometrics, and by your recovery phrase if you generate one.
Your photos, voice memos and handwritten pages are encrypted too (XChaCha20-Poly1305), with a key derived from that same key: on your device, they never exist unencrypted. Your voice, in particular, is encrypted while it is being recorded, not afterwards.
We have no technical means of reading your entries. If you lose both your password and your recovery phrase, nobody — including us — can recover your journal.
Habit tracking (in beta) keeps your habits — their names, the days you log, the time of their reminders — in the same encrypted database as your journal, and they are part of your backups. Their reminders are notifications scheduled by your device, with no server involved, and neutral wording: they never name the habit.
The analysis that suggests a mood from your words runs entirely on your device. No text is ever sent over the internet to be analysed. The suggestion is only a proposal: the mood that counts is the one you pick.
The microphone is only used while you record a memo. Transcription, which is optional, happens on your device: either with the system's speech recognition (Apple or Google), used in its on-device mode only — the app never uses recognition that would go through a server — or with Whisper, an open model you can turn on in Settings. The resulting text is added to your entry; the audio stays attached, encrypted.
The system may need to download its own language model (about 60 MB on Android): it does so itself, after asking you. On iPhone, the system also asks for the "Speech Recognition" permission.
If you turn on Whisper, the app downloads its model once (190 MB) from Hugging Face (huggingface.co), where its authors publish it. This request carries no identifier and nothing about your device or your journal; as with any web page, Hugging Face sees the connection's IP address. The file's fingerprint is checked before use, and you can delete the model at any time.
Handwritten pages are not analysed: they are not read, not searched, and not turned into text.
When you create a backup, Aparté produces an encrypted file that you save wherever you want (iCloud, Google Drive, your computer…). We don't host that file and have no access to it. If you choose the plain-text export, the app warns you explicitly before producing a readable file.
If you share your year in pixels, the image is made on your device, then handed to the system share sheet: you choose where it goes, and we never receive it. It only shows the colours of your moods, day by day, and not a word from your journal. The number of days and the app's name appear if you leave them ticked. If you save it to your photos, iPhone asks you for permission to add it: Aparté uses it for nothing else.
If the app crashes, it can send a technical report through Firebase Crashlytics (Google). This is off by default. It is only enabled if you answer "yes" to the question the app asks after a few launches, or in Settings, and you can turn it off at any time.
A report contains: the error and its stack trace, your device model, OS version and app version, plus an installation identifier generated by Firebase. A report never contains the content of your entries, your moods, your tags, or your password. We attach no personal identifier to these reports.
This data is processed by Google as a processor. Legal basis: your consent (GDPR Article 6(1)(a)), withdrawable at any time. See the Firebase privacy policy.
Aparté offers, as an option, two uses of an AI that runs entirely on your device (Apple's system models or, on Android, Google's Gemini Nano, on compatible devices): a narrative recap of your week, and "Go further", a follow-up question about the entry you are writing, generated only when you ask for it. This feature is currently in beta and off by default: it only turns on after explicit consent — in Settings, or when you ask for a first question —, the app explaining the following before you enable it. Without it, "Go further" offers questions written in advance, with no AI at all. An entry in which the app spots words of distress is never submitted to the model, not even locally.
The text of your journal, the quotes a recap or a question draws on, and the themes extracted from it never leave your device — not during generation, and not in the diagnostics below. If you turn this feature on, the app may send, through Firebase Crashlytics, anonymous technical reports meant to evaluate and improve the beta: whether the model is available on your device (and, on Android, the version of the system's AI service and the download state of its model), whether a given generation succeeded or failed (a recap or a question, and whether a question was discarded by the app's checks, with the reason as a code), a generic error code (not the error message, which could quote your text), technical measurements — durations, the size of the processed text as a number of characters or tokens (never the text itself), available memory and storage, battery level, thermal state —, your device model and interface language, your explicit feedback ("does this recap ring true?") as a simple yes/no, and what you do with a suggested question (answer it, ask for another, close it, or leave it be). That last measure is only sent if you accepted the version of the consent text that mentions it. These reports never contain the content of your journal, your quotes, your themes, or your mood.
This data is processed by Google as a processor, under the same legal basis as crash reports: your consent, withdrawable at any time in Settings.
At launch, the app downloads a small public file from our website (aparte.zenith-ai.dev) stating the minimum version still supported. If your version is older — for instance because it contains a serious defect — the app asks you to update it, while still letting you back up your journal.
This request contains no identifier and no information about your device or your journal: it is the equivalent of opening a page of our website. As with any website visit, the host (Firebase Hosting, Google) sees the connection's IP address; we don't use it or link it to anything.
Aparté is a paid app. The transaction is handled entirely by Apple's App Store or the Google Play Store depending on your platform: we never see or store your payment details. The app contains no in-app purchases and exchanges no data with any payment service.
After about a week of writing, and at most once every four months, Aparté may ask the system to show the Google Play or App Store rating window. Your rating and any review are entered in that window, which belongs to Google or Apple, and are sent by them under their own terms: Aparté has no access to them and sends nothing from your journal.
We do not sell, rent or share any personal data, for the simple reason that we hold none. The only third parties involved are those mentioned above (Google for optional crash reports and AI beta diagnostics; Hugging Face for downloading the Whisper model, also optional, which receives no data about you; Apple or Google for purchases and, if you leave one, your review).
The GDPR grants you rights of access, rectification, erasure, portability and objection. With Aparté, you exercise these directly from the app, since you are the sole holder of your data:
You may also lodge a complaint with your local data protection authority (in France, the CNIL).
Aparté is not directed at children under 13 and does not knowingly collect any data about them.
Aparté is not a medical device. It provides no diagnosis and does not replace care from a health professional.
If this policy changes, the date at the top will be updated. Any change that would widen data collection would be asked of you explicitly inside the app, never imposed silently.