European legal order
The platform runs in European data centres. The legal basis of the operation is processing under a GDPR data processing agreement; location, operator and terms are confirmed in the offer before anything is provisioned.
EU models
European On Demand runs the same enterprise AI in European data centres: billed per user, with no hardware of your own, and with a 3-day trial. The same rules apply as on the appliance.

How European On Demand works
The path is short because no hardware of your own has to be procured.
3 days free, with no infrastructure of your own and no rebuild. You ask your own questions and see sourced answers.
Location, operator, scope of service and terms are confirmed in writing. That includes the data processing agreement with the technical and organisational measures.
Your separate tenant is set up in European data centres. Tenant, workspace and personal content stay strictly apart.
Your sources are connected after your approval; the permissions of your file stores are taken over. The documents stay where they are.
Billed per user and month, monthly, with no hardware contract. The number of users follows your team, not a procurement cycle.
Why European
Where your enterprise AI computes decides which law applies to your data. Everything else stays as you know it from the appliance.
The platform runs in European data centres. The legal basis of the operation is processing under a GDPR data processing agreement; location, operator and terms are confirmed in the offer before anything is provisioned.
The models run inside the platform. Prompts and documents are not passed on to model services of US vendors or to other external interfaces.
Sources, permission checks, approvals and integrations work here just as they do on the local appliance. The operating path changes nothing about the functional scope.
Both operating paths use the same platform. A move onto a local appliance runs as a project and is recorded in the offer; the other way, a local operation can be complemented by European hosting.
One tenant, one legal order
Your tenant runs separately in European data centres. The models work inside the platform, so there is no passing of prompts or documents to model services outside the platform. The data processing agreement, with the technical and organisational measures, belongs to the offer and is concluded before anything is provisioned.

Where the models come from
We rely on open-weight models that run inside the platform, locally and in European hosting. Which models work in your tenant is recorded in your offer.
The platform runs open-weight models itself; it does not call external model services. On the reference appliance this operation is in use and tested.
European vendors such as Mistral AI publish open-weight models that the platform runs. Which of them work in your tenant you settle with us in the introduction project.
The model list for your tenant sits in your offer: there we confirm the choice, the scope and the terms in writing. What binds is that offer, not this page.
Naming Mistral AI refers to the origin of freely available model weights. It is not a partnership, not a certification and not a commitment as to the scope of delivery.
Local or EU-hosted?
The choice turns on whether your data may leave the building and whether you want to run hardware of your own. Local is the Private Spark plan; hosted in Europe it is European On Demand.
| Criterion | Local on the appliance | European On Demand |
|---|---|---|
| Data location | Your appliance in your building; no connection to external model services. | European data centres; location and operator are confirmed in the offer before anything is provisioned. |
| Hardware | A DGX appliance included in the plan, sized for one team. | No hardware of your own required. |
| Billing | A fixed monthly amount per appliance, 6-month minimum term. | Per user and month, billed monthly. |
| Getting started | Setup and an accompanied start as set out in the offer. | Trial: 3 days free, with no infrastructure of your own. |
| How the models run | Open-weight models run inside the platform on your appliance. | The same open-weight models run inside the platform in your tenant. |
| Platform rules | A source on every answer, a permission check on every question, changes only after approval. | Identical: the same platform, the same rules. |
The most sealed-off path is local operation; the fastest start, with no hardware of your own, is European hosting. Compare operating models and prices
What holds on both operating paths
Five rules that apply to every answer and every access.
Common questions
Inside your tenant in European data centres. There is no passing of prompts or documents to model services outside the platform. Location and operator are confirmed in the offer before anything is provisioned.
No. It is the same platform with the same rules: a source on every answer, a permission check on every question, approvals for changes. What differs is location, hardware and billing.
The selection that applies to your tenant sits in your offer. The basis is open-weight models that the platform runs itself, including models from European vendors such as Mistral AI.
Yes, the operation is processing under a data processing agreement. The agreement, with the technical and organisational measures, belongs to the offer and is concluded before anything is provisioned.
Yes. Both operating paths use the same platform; the move runs as a project and is recorded in the offer. The other way, a local operation can be complemented by European hosting.
Because prompts and documents would then leave the European legal order, and a change of vendor can land on you. Open-weight models run under our control. You decide the location.
EU models
The trial runs for 3 days in European data centres. You see sourced answers to your own questions before an offer is on the table.