Private AI · GDPR
Private LLMs — use AI without handing over your data
Contracts, HR files, engineering data: much of it must not go to an external AI service. A private language model on your own infrastructure or in an EU data centre gives you an internal assistant that knows your documents — while everything stays under your control.
Why companies want private AI
Confidential data
Customer, HR or R&D data must not leave the company, contractually or legally.
Works council & compliance
External AI services fail at approvals, works agreements or internal policies.
Scattered knowledge
Answers are buried in thousands of PDFs, wikis and inboxes — people search instead of work.
Set up in four steps
1. Requirements & data protection
Which data, which users, which permissions? This defines architecture, deletion concept and documentation for the DPO and works council.
2. Model & hardware
Choosing a suitable open-source model (e.g. Llama, Mistral, Qwen) and sizing: GPU server, EU cloud or existing infrastructure.
3. Knowledge connection (RAG)
Documents are indexed and searchable with permissions; answers cite their sources.
4. Operation
Updates, monitoring, quality measurement and training — or handover to your IT.
Your benefits
- No data to third parties — on-premise or in an EU data centre of your choice
- Answers with sources and permission checks
- Predictable costs instead of per-token pricing
- Documentation for your data protection officer and works council
- Open standards — no proprietary black box
Frequently asked questions
Are local models good enough?
For well-defined tasks such as questions about internal documents, summaries or classification, current open-source models perform very well. We measure this upfront with your real questions.
What hardware is needed?
For a team of 20–100 users a server with one or two GPUs is often enough. Alternatively we rent GPU capacity in a German or EU data centre.
Is it GDPR-compliant?
If data never leaves your network, there is no third-country transfer. We also document purposes, permissions and retention periods for your records.
What does a private AI assistant cost?
Setup and integration typically land in the low five figures (EUR), plus hardware or rent. You get a fixed-price proposal after the first call.
Check private AI for your company
In a free first call we clarify data, requirements and a realistic architecture — no sales pressure.
Have an idea or a process that eats hours every week?
Let's find out in 30 minutes whether AI or a custom product can fix it, and what it would cost.