Mistral
Mistral is a French AI company, and the name of the family of large language models it releases, several with weights that anyone can download and run. The models are met behind other products more often than under their own name, so a business rarely encounters Mistral as an assistant a customer is talking to.
A model family, not an assistant
An assistant is a product: a chat interface, a retrieval step that searches and fetches the live web, and a model that composes the answer. A model is the last of those three. Mistral publishes models and also runs its own chat product, but its models are equally likely to be running inside someone else’s application, on someone else’s servers, with someone else’s retrieval. Whether such a deployment fetches the web at all, and under what user agent, is the deployer’s decision. There is no single robots.txt line that addresses every place a Mistral model is running, so a site owner should read the documentation of the product in front of the customer, not the model vendor’s.
Open weights and what they change
When a model’s weights are downloadable, the training data is fixed at release and nothing a business publishes afterwards reaches that copy of the model. What the model knows about a business at answer time therefore comes from retrieval, if the deployment has any, or from nothing. This is the general case for every model, open or closed, but open weights make it visible: the model on a laptop cannot have read yesterday’s page.
Measured here
This site has not probed any Mistral model or product. Its probe ran against Claude, and its site-side readiness figures describe what pages serve to a crawler regardless of which model later reads them; those are in the 9,483-site audit.
A common question
Which robots.txt line covers the places a Mistral model is running?
There is no single line that covers them all. A model is one part of a product; the chat interface and the retrieval step around it belong to whoever deployed it, and that deployer decides whether the web is fetched at all and under which user agent. So the documentation worth reading is that of the product the customer is actually talking to, not the model vendor's. This site has not probed any Mistral model or product.