Trust Center

How we handle data

No marketing language. If something is missing, email us — it will be here at the next update.

There is an AI assistant in the bottom right of this site. It says so, throughout the conversation.

Where it runs

On our own hardware in the Netherlands: a Mac Studio and a Synology NAS under our own control, reachable via a Cloudflare tunnel with Caddy as reverse proxy. No cloud provider sits between the site and the server.

Which models we use

Not “leading AI”, but names. Local open-weight models on our own hardware, or routed through our own model router to external providers such as DeepSeek, Mistral, Groq, Google Gemini or Cohere. Per project it is fixed which model runs where.

Is input stored or used for training

No. With external APIs we use settings under which input is not used for model training. If a system runs entirely locally, the data never leaves the machine.

What we log and for how long

What is needed to demonstrate what happened: which request, which model, how many tokens, what cost. Not the content of documents. Retention and rights are stated in the privacy statement.

Citations and hallucination

Every answer from our RAG systems shows where it came from: document and passage, clickable. That is not an extra feature but the starting point of the architecture. We measure the hallucination rate with a fixed golden set of questions with known answers; the method and the outcome are published here once the series is running — including when the outcome disappoints.

The AI assistant on this site

Bottom right you will find the site assistant: an AI assistant that answers only from what is on this site, with the source under every answer. If it finds no source it says so and refers you to a human it does not guess. The disclosure that you are talking to AI sits at the top of the window and stays there throughout the conversation, per Article 50(1) and (5). Here we are both provider and deployer, so all four obligations rest with us.

What happens to your question: no cookies, no localStorage, no tracking. The conversation lives in your tab and is gone when you close it. On our side we log the question, the answer and which sources were used — so we can measure how often it gets things wrong. Do not enter personal data here.

What we do not do

  • No legal, medical or financial advice without human involvement
  • No automated decisions about individuals
  • Data is never shared with third parties for commercial gain — ever

Disclaimers

Our systems support people, they do not replace them. For legal or high-risk questions we refer to a human. Neuralex does not provide legal advice and is not affiliated with the Dutch Judiciary or government.

Article 50 EU AI Act

How we handle it ourselves

Since 2 August 2026 the transparency obligations of Article 50 of the EU AI Act apply. This is how it is arranged on this site: the disclosure stays visible for the entire conversation.