Sovereignty
French generative AI, and checkable
"Sovereign AI" has become a claim you read everywhere and can almost never verify. This page states what is true here, what is not, and where to go and check each statement. It is the only format that survives a security review or a public-sector buyer.
What is true, and checkable
The publisher is a French company
BeeMM is a French SAS. The contract you sign is French, the invoice is in euros, support answers in French, and the competent court in a dispute is French. That is not a detail: it is what actually changes when something goes wrong.
The processing agreement is published
Our GDPR Article 28 agreement is online and readable before any signature, rather than "available on request". Your impact assessment has a document to cite today, without going through a salesperson.
Every model provider is named
A public annex identifies, model by model, the publisher that runs it. So you can check each line against your own policy instead of taking our word for it. That is the piece missing from almost every "sovereignty" page.
Inputs and outputs are not used for training
What you generate is not reused to train models. The detail, including what each provider does on their side, is in our terms and in the providers annex.
No lock-in to one publisher
The catalogue gathers 98+ models from around fifteen different publishers. Depending on a single AI vendor is precisely what sovereignty is meant to avoid — including when that vendor is European.
And what is NOT true
The models themselves run at their publishers, several of which sit outside the European Union. So we do not claim your data never leaves France. We say where it goes, for each model, and let you decide which ones to use.
How to check it yourself
Four public documents, and no sales call required.
- 1 · The publisher’s identityThe legal notice gives the company form, the registered office and the publication director. The company number is verifiable on the French public business registry.
- 2 · The processing agreementThe DPA page sets out the Article 28 agreement: purposes, retention, sub-processors, security measures. To be read before signing, not after.
- 3 · The providers annexThe "Models and providers" page names, for every model in the catalogue, the publisher that runs it. That is where what leaves the EU, and what does not, can be checked.
- 4 · The privacy policyWhat is collected, why, for how long, and the rights you can exercise. Alongside the cookie policy, which lists what is set and under what condition.
Frequently asked questions
- Does my data stay in France?
- No, and nobody honest can promise that while aggregating frontier models. The publisher is French and the platform operates under GDPR, but the models run at their publishers, several of them outside the EU. Our public annex names them one by one so you can decide model by model.
- So what is actually French?
- The publisher, the contract, billing in euros, support, the competent jurisdiction, and the platform orchestrating the generations. Those are the parts you have recourse against — which is the point of the question.
- Do you have a GDPR processing agreement?
- Yes, and it is published rather than supplied on request: the Article 28 agreement is readable online before any signature, together with the sub-processors.
- Do you train models on our content?
- No. Inputs and outputs are not reused to train models. The detail, including what each provider does on their side, is in our terms and in the providers annex.
- Can we restrict ourselves to European models?
- You pick the model at every generation, and the public annex states the publisher of each. Restricting use to a list you have approved is therefore possible, and that is exactly what naming the providers is for.
Check before you try
All four documents are public and readable without an account. The free plan then gives access to the whole catalogue.
Keep reading
- A canvas, not a chat boxAn infinite AI canvas to generate, compare and chain: every image and video model on one surface, collaboratively, shippable as an app.
- The AI image generator that does not lock you to one modelAn AI image generator with the whole catalogue in one picker: compare models on the same brief, edit, upscale, on a single credit balance.
- Nano Banana on Beemm VisionUse Google's Nano Banana 2 and Nano Banana Pro image models on one canvas: legible in-image text, conversational editing, output up to 4K.
- Kling on Beemm VisionGenerate video with Kling V3 and Kling O3 4K: text, image or reference to video, up to 15 seconds, multi-shot sequences and generated audio.
- The full catalogueEvery image and video model available in the canvas.