AI that makesnothing up
The two questions every business owner asks are the same: can I trust it, and where does my data go? Both are technical choices, not a matter of promises. A constrained system with source references makes nothing up, and a model on your own server sends nothing out.
Does this sound familiar?
- 01An ordinary chatbot gives a fluent answer that is factually based on nothing.
- 02Nobody knows exactly which company data is being pasted into which tool.
- 03There is no record of what the system decided and why.
- 04The GDPR question only comes up once something has already been built.
What you get.
- Answers constrained to your own sources, with a reference attached
- Open models that run entirely on your own server, so no data goes to third parties
- A hybrid setup if that suits better: sensitive work locally, the rest in the cloud
- A record of what the system did, usable towards a regulator
- A data processing agreement and a clear description of which data goes where
Operating it yourself says more than a description.
The same question, two answers side by side: an unconstrained model that may make things up, and a constrained answer with sources or an honest 'I don't know'. This is a demo on sample data, not client work.
Demo · prepared in advance · illustrative · no live model
Choose a question
Illustrative demonstration with examples prepared in advance. No live AI call and no API key needed; the answers are fixed and contain no invented figures or names.
Four steps, in this order.
We name what is sensitive
Which data must not leave the building, and which does not matter? That line determines the rest.
We choose the setup to match
Local, in the EU or hybrid. The choice follows from the line, not from my preference.
We put it in writing
Data processing agreement, roles and a description you can show a customer or a regulator.
We keep testing
A fixed set of questions with which you can see for yourself whether the quality holds up.
Lead time: depends on scope, usually this starts with a single conversation.
Open modellen · On-premise · EU-hosting · Evaluatiesets · Monitoring · AVG
What people ask me about this.
Q1Can AI really run entirely on my own server?
Yes. In recent years open models have become good enough for most business work: reading documents, summarising, sorting, answering questions from your own knowledge. For the heaviest reasoning work the large cloud models are still stronger. I would rather name that difference up front than afterwards.
Q2How do you stop AI from making things up?
By not letting the model answer from memory. It is given the relevant passages from your own sources and may not claim anything beyond them, with the source visible next to the answer. If it is not in there, the answer should be 'that is not in here'. That is a design choice, not a setting you switch on.
Q3Who is responsible under the GDPR?
You remain the controller, I am the processor and work on your written instructions. We set that out in a data processing agreement: what security is in place, how I help you with requests and data breaches, and that everything is erased once we are done. If I bring in a party that takes part, it is bound by the same obligations.
Q4We work with patient or client data. Is this still possible?
Often yes, but then the setup is stricter: running locally, only the data needed for that one task, and access per role. That is not an extra service, that is the condition. If it cannot be done responsibly in your situation, you will hear that from me before anything is built.