Language-model independent
On the way to sovereign AI: you choose the language model
vragen.ai is not tied to one language model. You choose which model writes the answers, and every model shows where it runs. Choose a model that runs in the Netherlands and you have a Dutch AI site search and chatbot on your website, on Dutch soil from question to answer.
-
In the Netherlands
Open models we run our own copy of, in Dutch data centers. The question does not leave the country.
-
In the EU
The GPT models from OpenAI, through Microsoft in a European data center. They do not train on your data; we put that in the data processing agreement.
-
In the United States
Claude from Anthropic, for those who want it. It runs at Anthropic in the United States, and Anthropic does not train on your data either.
Which language models can we choose?
Answer by vragen.ai
vragen.ai works with 13 language models, 5 of which run in the Netherlands
Which model fits best depends on your content, your visitors and where your data is allowed to be. We are happy to help you choose and advise which model suits your situation.
New models are added all the time. Missing one? Let us know, and we will see whether we can add it. This list is from .
| Model | Made by | Runs in | Intelligence | Speed |
|---|---|---|---|---|
| Kimi-K3 | Moonshot AI | NL Netherlands | 4 out of 4 | 3 out of 4 |
| Mistral Large 3 | Mistral AI | NL Netherlands | 3 out of 4 | 3 out of 4 |
| VLAM Mistral medium | Mistral AI | NL Netherlands | 3 out of 4 | 3 out of 4 |
| GLM-5.3 | Zhipu AI | NL Netherlands | 3 out of 4 | 4 out of 4 |
| Qwen3.6 35B | Alibaba | NL Netherlands | 3 out of 4 | 4 out of 4 |
| GPT-6.1 Sol | OpenAI | EU Europe | 4 out of 4 | 2 out of 4 |
| GPT-6.1 Terra | OpenAI | EU Europe | 3 out of 4 | 3 out of 4 |
| GPT-6 Luna | OpenAI | EU Europe | 2 out of 4 | 4 out of 4 |
| GPT-5.6 Sol | OpenAI | EU Europe | 4 out of 4 | 2 out of 4 |
| GPT-5.6 Terra | OpenAI | EU Europe | 4 out of 4 | 3 out of 4 |
| GPT-5.6 Luna | OpenAI | EU Europe | 2 out of 4 | 4 out of 4 |
| Claude Sonnet-5.5 | Anthropic | US United States | 3 out of 4 | 2 out of 4 |
| Claude Haiku-5.5 | Anthropic | US United States | 2 out of 4 | 4 out of 4 |
Source: Language models in the Netherlands Hosting and data storage Sovereign AI platform
How does choosing a different language model work?
Answer by vragen.ai
How AI site search works with a language model of your choice
The language model only writes the answer. Your knowledge base stays in the Netherlands and your settings stay the same, whichever model you choose. To switch, you pick another model, test the answers in the playground and put it live.
-
A visitor asks a question
The question arrives at vragen.ai, on our own servers in the Netherlands. Your knowledge base is there too.
-
vragen.ai searches your own content
The passages from your knowledge base that match the question go to the language model you chose for this agent, together with the question.
-
The language model writes the answer
Based on those passages, with the source included. The model does not train on your content or on your visitors' questions.
Source: How it works The playground
Why not one fixed model?
Answer by vragen.ai
Language-model independent, so not dependent on one AI provider
If you work with one fixed model, the provider decides which versions exist and when a version stops. With vragen.ai you choose: the smartest model, the fastest, or a model that runs in the Netherlands.
Language models improve fast. A new version comes out almost every month. We test those new models and add the best ones to the list. That way you can always move to a better model.
For the models in the Netherlands we have our own copy. It stays available, even if the maker stops supporting it. Does your organization already work with an approved AI provider? Then we can in principle connect it.
Source: Language models in the Netherlands Where is our data?
Can everything stay in the Netherlands?
Answer by vragen.ai
Sovereign AI: your AI site search runs entirely in the Netherlands
Sovereign AI, AI on Dutch soil where you decide on the answer yourself, is what we at vragen.ai have been working toward for some time. With the models in the Netherlands, that now holds for the whole chain.
If you choose a model labeled NL, the whole chain stays in the Netherlands. The application and your knowledge base are at TrueFullstaq in Almere. Large open models run at Nebul, a Dutch AI cloud company from Leiden with data centers in the Netherlands. Nebul does not store the content of questions, does not log it and does not use it for training.
Kimi, Qwen and GLM were developed in China, Mistral in France. The processing happens here: an answer needs no connection to the maker, no question goes to China and the maker sees no conversations.
Source: Language models in the Netherlands Sovereign AI platform Security
What if we would rather use GPT or Claude?
Answer by vragen.ai
You can choose GPT and Claude too, hosted in Europe or the United States
We use the GPT models from OpenAI through Microsoft in European data centers. Processing stays within Europe and there is no training on your data. For those who want it, Claude from Anthropic is available too. That model runs at Anthropic in the United States.
Every provider you choose is added as a subprocessor in your processing register. The docs list which parties are in the chain for your choice.
Source: Security DPIA and processing register
Can we try it with our own content?
Answer by vragen.ai
Create your own demo
You create an environment with your own content in it, for free. That way you see for yourself what your visitors will see: one answer, with the source included. If you like it, you choose a subscription.
Enter your details
- Free, no obligation
- With your own content
Timon Zuurmond
Account manager
Not sure whether your content is suitable, or which plan fits? Timon will take a quarter of an hour to look at it with you.
+31 6 12 15 05 90 tzuurmond@swis.nl
Mon to Fri, 9:00 to 17:30
Form not loading? Refresh the page, or get in touch with us.