Which AI models does Maditon use, and does my data reach a US provider?

Two rules govern every AI call.

The provider must be European or self-hosted. Maditon uses Mistral (France), with Ollama and vLLM on EU infrastructure as the self-hosted options. US-headquartered AI providers are excluded — not by preference but by the same jurisdictional reasoning that governs the rest of the stack. An EU endpoint of a US model provider would not satisfy it.

Personal data is stripped before the prompt leaves. The model needs to know what your AI system does; it does not need names, email addresses or other identifying detail to classify it. Those are removed at the boundary, at each point where a prompt is built.

Under the EU AI Act’s own vocabulary this makes Abiton Ventures a deployer of AI, not a provider of AI models — and Maditon applies to itself the rules it helps you apply. AI involvement is disclosed rather than hidden, AI output is a draft until a human accepts it, and AI actions are logged as metadata without the prompt contents.