AI transparency and model card
How the AI assistant in our documentation works, which model powers it, and where your questions go.
Last updated: 2026-07-14
What this AI is for
The "Ask our AI" assistant in our documentation answers questions using only the published Nordvec documentation. It is a convenience for finding and understanding information faster. It is not legal, financial, or professional advice, and it is not a substitute for reading the authoritative documentation or contacting our team for account-specific matters.
Model and provider
Answers are generated by Mistral (mistral-small), operated by Mistral AI in France. Mistral is our sole generation provider; there is no US-jurisdiction model or fallback. We select the model in configuration and can change it; this page names the current provider and jurisdiction.
Where your questions go
Your question and the retrieved documentation are processed within the EU. Nothing is sent to a non-EU model or a third-party analytics service as part of answering. This is deliberate: generation runs in France and embeddings in Berlin.
Training and retention
Your questions are not used to train any model. Zero-retention is enforced at the model-provider account level, our Mistral account has training and data retention turned off, so your inputs are not kept to improve public models.
How it produces an answer
The assistant uses retrieval-augmented generation: it selects the most relevant documentation passages, gives them to the model as context, and asks it to answer only from that context. Every answer carries numbered citations linking back to the exact source pages, so you can verify each claim rather than trust it.
Accuracy and limitations
AI-generated answers can be incomplete or wrong, and the model can misread a source. Always check the answer against the linked sources; where an answer and a source disagree, the source is authoritative. We run an automated citation-accuracy evaluation that verifies each cited source actually supports the claims attributed to it, and we treat regressions as release blockers.
Human oversight and reporting
You are always in control: you can ignore the assistant and read the documentation directly, and every answer links to the underlying pages. If an answer is wrong, misleading, or cites a source that does not support it, please report it so we can correct the documentation or the retrieval.
AI marking (EU AI Act Art. 50)
Under Article 50 of the EU AI Act, we tell you clearly when you are interacting with an AI. The assistant is labelled as an AI in the interface, and every answer carries a machine-readable and human-readable disclosure that it was AI-generated and may be imperfect.
Contact
Questions about how this AI works, or reports of an incorrect answer, can be sent to privacy@nordvec.com.