AI Transparency
Ce document n'est pas encore dans ta langue
La traduction est encore en attente. Un texte qui engage doit être rédigé et relu dans chaque langue plutôt que traduit à la machine : d'ici là, le texte ci-dessous est la version qui s'applique à toi.
Disponible en: English
Brouillon — pas encore relu par un avocat
Ce document a été rédigé en interne et n'a pas été relu par un juriste qualifié. Il est publié pour pouvoir être lu et corrigé avant le lancement. Ne le prends pas pour un avis juridique définitif, et ne le soumets pas à un store comme version finale sans cette relecture.
En bref
Daily+ has four optional AI features. They run on our servers and send your request to a third-party language-model provider. This notice says exactly what each one sends, who receives it, what is kept, and what the features cannot do.
1. In short
AI features are optional, premium, and off until you turn them on. Nothing is sent to a model provider until you have given consent, and withdrawing consent in Settings stops it immediately.
Your content is not used to train models — not by us, and not by the providers under our agreements with them.
2. What each feature sends
Every AI request goes from the app to the Daily+ API, and from there to the model provider. The app never talks to a model provider directly, which is what lets us keep the provider swappable and keep the request minimal.
| Feature | What is sent | What comes back |
|---|---|---|
| Assistant | Your message, the recent conversation, and the specific figures needed to answer it (for example today’s intake if you asked about today’s intake) | A generated text reply, labelled as AI-generated |
| Photo-to-food (Snap Meal) | The photo you took, and nothing else from your logs | A list of candidate foods and estimated portions for you to confirm or correct before anything is logged |
| Recipe import | The recipe URL or photo you supplied | A structured ingredient list and method for you to review before saving |
| Weekly review | Aggregated figures for the week — totals, averages and trends, not individual entries | A written summary and suggestions |
3. Who processes the request
Our default provider is Google (Gemini API). The system is provider-swappable and may also use Anthropic (Claude) or OpenAI. All three act as our processors under a data-processing agreement, are located in the United States, and receive data under EU Standard Contractual Clauses. The current provider for your request is disclosed in the app’s AI settings.
We do not use a provider that reserves the right to train on our customers’ content.
4. What is kept
We do not store the content of your AI requests after the response has been returned. We do keep an audit record of the fact a request happened — timestamp, feature, model and token counts — for 12 months, for abuse prevention and cost control. That record contains no request content.
Providers may retain request data for a short period for abuse monitoring under their own terms; we select configurations that minimise this and never one that permits training.
Any AI output you choose to save — a logged meal, an imported recipe, a saved summary — is your content from that point on and is stored like anything else you logged.
5. Consent and control
The app asks for consent before your first AI request, states which provider will receive the data, and records the version of the notice you agreed to. You can withdraw consent at any time in Settings; withdrawal disables the features and no further data is transmitted.
Where an AI request would carry health data, that consent is explicit consent under Art. 9(2)(a) GDPR and is requested separately from the general terms.
6. What AI in Daily+ cannot do
Language models produce plausible text, not verified fact. The features here will sometimes be wrong — a misidentified food, an invented nutrient value, a confident but incorrect answer.
- AI output is always presented for your review. Photo-to-food and recipe import never write to your log without you confirming.
- AI output is never medical advice, and the assistant will decline clinical questions rather than answer them.
- No decision with a legal or similarly significant effect is made about you by AI, or by any automated means.
- Deterministic calculations — energy targets, one-rep-max, progression, urgency scores, correlations — are ordinary tested code, not AI. AI is used only where the input is genuinely fuzzy.
7. Reporting bad output
If an AI feature produces something wrong, unsafe or offensive, report it from the message in the app or write to support@dailyplus.app. Output that could put someone at risk is treated as a priority report.