Artificial intelligence
Private AI for sensitive data
Private AI runs open language models on a server in the European Union or on your company's own servers, so sensitive data never passes through public AI services. It is for clinics, law firms, accountants and financial services that want to use AI without losing control of their information. You always know where the data is, who can reach it and which model is processing it.
When data can't leave the building
A patient file, a contract under negotiation or a client's accounts shouldn't be pasted into a public AI service. Yet teams want to summarise, search and draft as quickly as they see others doing, and some are already doing it quietly. Private AI resolves that tension: the model comes to the data, instead of the data going to the model.
Two ways to host it
On a dedicated server from a European provider, in an EU data centre, only your company has access and there's no hardware to buy. On your own servers, the data never leaves your premises, which some sectors prefer or require. Either way, nothing is sent to outside AI providers, and you know exactly which version of the model is answering.
What we build on top
The model is the foundation. On top of it we build what your team will actually use: an internal knowledge assistant that answers from your company's procedures, document intelligence for invoices, reports or applications, or search across old archives. If you'd like to review the security and GDPR position of your current systems first, start with a security and GDPR review. The stages of the work are set out in how we work.
What you get
- Choosing the right open model for the job, such as Llama, Mistral, Qwen or Gemma
- Setup on a dedicated server in the EU or on your company's own infrastructure
- Assistants, document search or data extraction running on that model
- Per-user access control and usage logs
- Testing answer quality against real examples from your work
- Technical documentation for your data protection officer and for audits
Best for
- Clinics, laboratories and practices handling health data
- Law firms and solicitors bound by professional confidentiality
- Accountants, insurers and credit intermediaries
- Companies whose internal policies forbid sending documents to outside services
Typical features and modules
Integrations & automation
AI add-ons
- Document reading
- Smart search
- AI chat assistant
- AI transparency & logging
Compliance & security
- Security hardening
Hosting, e-mail & maintenance
- Admin training
- Monthly maintenance
How we deliver it
Mapping the tasks, the data involved and the rules that apply
Comparing open models on real examples in an isolated environment, including a licence check
Choosing where to host and preparing the server, with access control and backups
Building the application your team will use on top of the model
Launch with training, technical documentation and quality monitoring
Frequently asked questions
Is an open model as good as ChatGPT?
For well-defined tasks such as summarising, classifying or answering from your own documents, today's open models are often enough. For harder reasoning or long, creative writing, a large commercial model may still make a difference. We test with real examples before recommending anything.
What hardware do we need?
It depends on the model and how many people use it. Smaller models run on a server with a single graphics card; larger ones need more memory and several cards. You can also rent GPU servers in EU data centres rather than buying equipment.
Will our data be used to train the model?
No. The model runs on a server your company controls, and neither documents nor questions are sent to third parties. The model is only fine-tuned on your data if you decide it should be, and then the tuned model stays with you too.
Does this make us GDPR-compliant?
It helps a great deal, but it doesn't replace everything else. Health data is a special category under the GDPR and still needs a lawful basis, access control and records. We prepare the technical documentation; the legal assessment stays with your data protection officer or lawyer.
Do users know they're talking to an AI?
Yes. The assistant identifies itself as an AI, as the EU AI Act has required since 2 August 2026, and every answer shows which documents it came from. In sensitive settings, a professional always makes the final call.
How much does it cost?
Every project is quoted by hand from your requirements, so pricing is on request. Hosting depends on the model and the number of users, and we explain it in the proposal. Request a quote
Shall we talk about your project?
Tell us what you have in mind and what you want to achieve.