A chatbot that invents nothing. Because it knows only your knowledge.

Yes, and they learn it at first contact. Since 2 August 2026, Article 50(1) of the EU AI Actthat an assistant identifies itself as an AI system – visibly, understandably and accessibly, not hidden in the imprint. We build the notice into the chat window from the start and avoid human-sounding fantasy names without clarification. That is better operation anyway: an assistant posing as a human gets found out and damages trust.

From the sources you provide: website, documents, FAQ, internal instructions. It answers only from those, names the source – and if it finds nothing, it says so and passes on instead of guessing.

That is exactly what the RAG architecture prevents: the answer must be demonstrable from found passages of your knowledge. Edge cases go to a human. There is no hundred per cent in technology – but „with a source or not at all“ is a different level of safety from free invention.

A clean handover to the right person – including the conversation so far, so your customer need not repeat anything. For us the handover rate is a metric, not an embarrassment.

Yes: EU processing, no model training on your data, access logged. For sensitive documents we build the knowledge layer on your own infrastructure.

The assistant synchronises from your sources – change the shipping terms and the answer changes. We set up the maintenance workflow so updating is not a project but a simple action.

Absolutely – that is exactly what we recommend: a pilot with the 20 most common questions, tested internally, put live, measured. What proves itself gets extended. Setup from about 3,000 euros.