Sovereign AI for SMEs
Generative AI, without handing your sensitive data to a third party
We provide open-source AI models running on dedicated infrastructure, either our machines or your own server. You put AI to work on your documents and processes without your data passing through an outside AI provider.
Your teams already use AI, just not on your data
Consumer assistants are effective, but every document pasted into them goes to a provider whose hosting and terms you do not control. The result: either AI is banned and worked around, or it is used with no rules at all.
- —Your staff paste contracts, quotes or customer records into an online assistant because there is no internal alternative.
- —You have given up on AI for your most useful data (HR, legal, financial) because it must not leave the company.
- —Your customers or prime contractors ask where the data they entrust to you is processed, and you have no clear answer.
- —The big vendors' "enterprise AI" offers are beyond your budget or tie you to a single provider.
How we work
Use-case scoping
We list the use cases worth having (document search, drafting, data extraction, request triage) and classify the data involved by sensitivity.
Model and hosting
We pick an open-source model suited to each use case, and the infrastructure: our dedicated machines or a server installed on your premises.
Go-live
We deploy the model, connect it to your documents and tools, set up per-user access and log usage.
Operations
Updates, backups, monitoring and new use cases over time. Your teams are trained to make the most of it day to day.
What we use
Sovereign AI is not second-rate AI: recent open-source models cover most of what an SME needs, provided you pick the right model for each task.
- Open-source models (Mistral, Llama, Qwen…)
- Run on the chosen infrastructure, with no call to an outside AI API. The model is chosen by use case, language and available resources.
- D1 Consulting dedicated infrastructure
- Our GPU servers, one with two NVIDIA RTX PRO 6000 cards (96 GB of memory each) and one with an NVIDIA RTX 5090 (32 GB), made available to our clients with each client's data kept separate.
- Your own server
- When data must not leave your premises, we install and maintain the solution on site.
- Assistant on your documents
- An assistant that answers from your procedures, contracts and document bases, citing its sources.
- n8n
- AI built into your processes (email triage, data extraction from invoices or forms, case summaries), on the same infrastructure.
What you get
- A map of your AI use cases and of how sensitive the data involved is.
- A private AI assistant, available to your teams, that works on your documents.
- Documented infrastructure: where the models run, where the data is stored, who can access it.
- What you need for your GDPR records and for assessing your obligations under the EU AI Act.
- Ongoing maintenance: model updates, backups, monitoring.
Frequently asked questions
What is sovereign AI?+
It is AI whose execution you control: the model runs on identified infrastructure, ours or yours, and your data is not sent to an outside AI provider. In practice, calls to an online AI service are replaced by an open-source model hosted on dedicated machines.
Are open-source models as good as ChatGPT?+
Not at everything. But for most of what an SME does (summarising, drafting, classifying, extracting, answering from documents), the gap is small with a well-chosen model. We test on your own cases before any commitment, and we tell you plainly if a use case is out of reach.
Where is our data hosted?+
On the infrastructure agreed during scoping: our machines, in France, or your own server. Each client's data is kept separate, and the exact location is set out in the documentation we hand over.
Do we need to buy hardware?+
No. You can use our dedicated GPU servers (192 GB of GPU memory in one, 32 GB in the other) with no upfront investment. If you would rather keep AI on your premises, we size the right server with you for your use cases.
What about GDPR and the EU AI Act?+
Keeping data on controlled infrastructure makes GDPR compliance much simpler: there is no transfer to an AI provider to document. For the AI Act, we help you classify your use cases and meet the obligations that apply, including informing users. This support is not legal advice.
What data would you like to put AI to work on?
A one-hour call is enough to identify your first use cases and the hosting that fits how sensitive your data is.
Talk about your project