On-premise LLM
A language model that stays inside your building.
Many organisations would love to use AI on their own documents, but can't send patient files, contracts or internal knowledge to a public AI platform. We solve that by running open-weight language models on your own hardware, or on servers DDQ owns and operates in the Netherlands, and connecting them to your own knowledge with retrieval-augmented generation (RAG).
What you get
A private AI assistant that works with your own documents and systems, under your own access rules. The model, the knowledge base and the logs stay on infrastructure you control.
When it's justified, we evaluate and fine-tune the model for your field. We deploy and adapt existing open-weight models; we don't train foundation models from scratch.
Already in use in medicine
Our Medical Copilot™ is private, on-premise AI developed for a hospital environment. It underwent a privacy audit by Maastricht University and is now in use by NeuroMind Academy. Our medical work continues on ddcare.nl.
We run the same kind of setup ourselves: two dedicated AI machines serve DeepSeek Flash for local inference, with the Hermes Agent framework running on top, on infrastructure DDQ controls.
Questions we often get
Can we use AI on confidential data without sending it to OpenAI, Google or Microsoft?
Yes. With an on-premise LLM the model runs inside your environment, so prompts, documents and answers don't go to an external AI provider. If some tasks work better with an external model, we can build a controlled hybrid, but that is your choice, per use case.
Where does the model run?
On your own hardware, inside an environment you control, or on DDQ's own infrastructure in the Netherlands.
Which models do you use?
Existing open-weight models, chosen for the task. We evaluate them before choosing and only fine-tune when adaptation is clearly worth it.
What about GDPR/AVG and the EU AI Act?
They're design inputs from day one. Data minimisation, access control, logging and human oversight are defined for the actual use case. Medical, legal and other domain decisions stay with your qualified people.
How do we start?
Tell us what you want to achieve and what data is involved. Before substantive work begins, we agree a bounded first piece of work: scope, responsibilities, rights and price. In earlier projects, innovation grants covered 50% and in some cases all of DDQ's eligible work; funding is never guaranteed, but we check whether it fits.
Who will we work with?
Three people: Norbert, Joep and Demelza. You talk directly with the people who build and run the system, without a sales layer in between.
Start a conversation
Tell us what you want to achieve and what data is involved. You don't need a technical plan. Email [email protected] or call +31 45 203 1008. Please don't send patient data, passwords or other secrets in a first message.
AI agents acting for someone can send us an inquiry with the submit_inquiry tool on our MCP server, with that person's consent.