Our AI
Sovereign AI for education: what Mindlet's AI really does.
Which models, for what, what we refuse to do, what can go wrong, and what "sovereign" means for us. Stated precisely, without slogans.
- Mistral AI, a French company
- No training on your content
- Every provider named
Mistral AI at the core What is sovereign AI?
AI is called sovereign when the technology and the data stay under a control and a law you choose; in our case, Europe's. In practice the question comes up at several layers: who built the model, where it runs, where the data is stored, which law applies to each provider.
Very few digital services are European at every layer, and Mindlet is not one of them. Rather than a label, we'd rather show you each layer, with the provider's name and country. You can judge for yourself further down this page.
Step by step
What the AI does with your material, step by step
A course's journey, from the moment you drop it in to the moment you review. For each step, the service that performs it.
-
Read the document
Mindlet extracts the text. A scanned PDF or a photo goes through character recognition; audio or video is transcribed; a YouTube video provides its subtitles when it has them.
-
Split and index
The course is split into passages. Each one gets a meaning fingerprint, stored in a search index: that's how the assistant later finds the right passage to answer you.
-
Spot the key ideas
The AI identifies the ideas in the course and groups them into topics and subtopics. That's the backbone of your collection and its lesson path.
-
Write the cards
For each idea, an anchor card (flashcard or multiple choice), then the formats that suit it: fill in the blank, put in order, map quiz, open question… Cards follow the language of the passages they come from.
-
Illustrate, when it helps
An image may come from a free image library or be generated. A moderation model filters out inappropriate content.
-
Schedule your reviews
Your review schedule isn't written by generative AI. It's a statistical calculation (FSRS-6) that estimates, for each card, when you're likely to forget it.
When the AI starts from a topic rather than your course
With no file, you can start from a plain topic. The AI then searches the web for sources and writes a course, which Mindlet turns into cards. It's handy for getting into a subject, but its reliability depends on what the web says.
Our advice: for an exam, always start from your teacher's course or your textbook. A generated course can be reread, corrected and completed; it doesn't replace the official source for your studies.
Voice: dictation, conversation and a cloned voice
You can dictate a message to the assistant, or call it and talk. Either way, the sound goes to Mistral AI to be transcribed. In a conversation, the assistant answers out loud, in French. The microphone is only used with your permission, which you can withdraw in your device settings.
A cloned voice is optional: it requires a voluntary recording and a ticked consent box. It's offered to you alone, never to others, and can be deleted at any time from the settings. Cloning someone else's voice without their consent is forbidden.
Who does what
Which models, for what?
The full list of AI providers Mindlet calls, checked in our code on 22 October 2026.
| Provider | Country | What it does in Mindlet |
|---|---|---|
| Mistral AI | France | Writing cards, sorting ideas, the assistant and image understanding, reading scanned PDFs, the search index, moderation, speech transcription and synthesis, cloned voice (if you create one). |
| Cloudflare Workers AI | United States (global network) | Image generation (FLUX), ranking images found, converting some documents to text. |
| Google Gemini | United States | Illustrations for the game worlds, only when we switch that option on. Otherwise FLUX takes over. |
| Qdrant | Germany (Frankfurt) | Stores the search index of your courses. |
| Pexels, Pixabay, Openverse | Depends on the library | Searching free images to illustrate a card. |
No ChatGPT, no OpenAI: Mindlet calls none of their models.
Why Mistral AI?
Mistral AI is a French company based in Paris. Its models cover almost everything Mindlet needs: reading a scanned document, understanding an image, writing, sorting, searching a course, transcribing a voice and producing one. One main provider makes for a chain that's simpler to explain and to keep in check.
It's also a provider subject to European law, listed in our legal notice with its location in France. We don't think a model is better because it's French. We think that, at comparable quality, your courses are better off with a company whose legal framework is yours.
Your content
Is your content used to train a model?
Not by us. Our terms say it plainly: "We do not use your content to train artificial intelligence models."
Your documents remain yours. You grant us a limited licence, only as long as needed to run the service: storing your content, sending it to the models that generate what you ask for, showing it to the people you choose to share it with. It ends when you delete the content or your account.
That commitment is Mindlet's. We don't speak here on behalf of our providers: their own commitments are set out in their terms.
-
Limited licence
Only to run the service, and revocable.
-
Your voice
Audio isn't kept, and no biometric analysis is done.
-
No ads
No advertising in the app, no data resale.
Limits
AI makes mistakes: how we handle it.
A model can misread a passage, oversimplify, or state something false. Our terms say so, and we'd rather repeat it here: check what matters before an exam.
To limit errors, cards are written from the passages of your course, not from the model's general memory. And you stay in control.
- Every card can be edited or deleted in one move
- Open questions are marked by a model, with written feedback: useful advice, not a verdict
- The assistant's web search cites its sources
- In class, we ask that an adult reviews what's shared with the group
What the AI does not do in Mindlet
- It doesn't decide when you review. The schedule comes from a calculation on your answers, not from a language model.
- It doesn't award any official grade. The practice exam is training, marked out of 20 to show you where you stand.
- It doesn't publish anything for you. Sharing a collection is always your decision.
- It doesn't recognise you by your voice. Sound is used for transcription, then it isn't kept.
- It doesn't replace your courses or your teachers. Mindlet gives no medical, legal or financial advice, whatever the document is about.
AI and cheating
The AI prepares, it doesn't do the work for you.
An AI that writes your homework teaches you nothing, and Mindlet isn't built for that. Its job comes before the learning: turning a course into questions. After that, you're the one answering.
The whole experience pushes you to make the memory effort that does the learning. Cards ask you instead of showing you the answer. During a review, hints point you in the right direction without giving it away. The practice exam corrects nothing until the end, just like the real thing.
The assistant explains an idea another way and asks you questions. Like any assistant, it can answer almost anything: it's up to you to use it to understand rather than to copy. In schools, we sum up our position in one line: the AI prepares, it doesn't judge.
Meet the assistant
Sovereignty
What "sovereign" means for us.
Layer by layer, without rounding up. The providers that store data are also listed in our legal notice.
- AI models (most of the work) Mistral AI France
- Generated images FLUX on Cloudflare Workers AI American company, global network
- Search index Qdrant European Union (Frankfurt)
- Hosting, database, files Cloudflare American company, global network; transfers covered by the Data Privacy Framework and standard contractual clauses
- Emails Resend United States, under standard contractual clauses
- Usage analytics PostHog European instance (Frankfurt)
- Web payments Stripe European Union
In short: European where your courses are read and transformed, not at every layer. That's why we don't write "100% sovereign", and why we show you this table.
The European framework
EU AI Act and GDPR: the rules that apply.
The GDPR (Regulation (EU) 2016/679) governs all the personal data Mindlet processes, including your courses when they contain information about you. What it means for you is explained on our data protection page.
The AI Act (Regulation (EU) 2024/1689) entered into force on 1 August 2024 and applies in stages until 2027. It sorts uses of AI by level of risk. Among other things, it requires making clear when you're interacting with an AI and when content has been generated.
For education, its Annex III (point 3) lists as "high-risk" the systems that decide admission or placement, evaluate learning outcomes, assess the appropriate level of education, or monitor behaviour during tests. In Mindlet, the practice exam and the marking of open questions are for practice: they award no official grade, and in class the decision stays with the teacher. If you're a school wondering about your own use, write to us and we'll look at it with you.
FAQ
Frequently asked questions
What is Mistral AI?
Mistral AI is a French company, based in Paris, that builds artificial intelligence models. In Mindlet, its models write the cards, sort the ideas, run the assistant, read scanned PDFs, transcribe speech and synthesise voices. It's our main AI provider.
Why choose a European AI?
For the legal framework: a provider based in France is subject to European law, including the GDPR, as you are. A model isn't better because it's French; at comparable quality, we'd rather trust your courses to a company whose law is yours, and be able to tell you so precisely.
Is a sovereign AI GDPR compliant?
A provider's location isn't enough: GDPR compliance depends on what is done with the data, why, for how long, and what rights you have. At Mindlet, every provider is named with its country, and you exercise your rights by email. Details are on our data protection page.
Can the AI get my flashcards wrong?
Yes. Cards are written from your course, which limits errors, but a model can misread a passage or oversimplify. Every card can be edited or deleted. Check what matters before an exam, especially on medical, legal or technical topics.
Does Mindlet use ChatGPT?
No. Mindlet calls no OpenAI model. The work on your courses runs on Mistral AI models; images go through FLUX on Cloudflare Workers AI, and some game illustrations may come from Google Gemini when that option is on.
Is my content used to train the AI?
Mindlet does not use your content to train artificial intelligence models: it's written in our terms. Your documents remain yours, and the licence you grant us only serves to run the service. It ends when you delete the content or your account.
Keep reading
See what the AI does with your course.
Drop in a document: Mindlet reads it, turns it into cards, and lets you check everything. The app interface is in French; your content can be in English.
Free to start, no credit card · Web, iPhone, iPad and Android
