MistralAI-Training
Mistral AI operates MistralAI-Training, a crawler that collects web content to build datasets for training Mistral generative AI models. It is not used for search indexing or to answer live user queries in Vibe. Site owners control it with the robots.txt token MistralAI-Training, which Mistral says it honors. Mistral documents no IP list and no reverse-DNS verification method.
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; MistralAI-Training/1.0; +https://docs.mistral.ai/robots)What MistralAI-Training does
In server logs, MistralAI-Training identifies itself with the string MistralAI-Training/1.0 inside a Mozilla-style user agent, followed by a link to https://docs.mistral.ai/robots. Mistral's documentation says the crawler collects web content that helps build datasets for training its generative AI models. The same page states the crawler is not used for search indexing and does not answer live user queries in Vibe, so a visit from this agent has no documented connection to a search result or to a real-time request from a person.
Blocking MistralAI-Training keeps a site's pages out of the training datasets Mistral builds with this crawler. Because the operator states the agent plays no part in search indexing or live queries in Vibe, disallowing it has no documented effect on how the site appears in search results or in the answers Vibe gives users. Allowing it means Mistral may use the content as training material for its models. Mistral's page says webmasters can disallow the user agent in robots.txt, and the entry records that the crawler honors those rules.
Verification is undocumented. Mistral publishes no IP address list for this crawler, and the URL analogous to the address lists for its other two agents, https://mistral.ai/mistralai-training-ips.json, returns HTTP 404. Mistral documents no reverse-DNS suffix either. Crawl rate and JavaScript execution are also absent from the docs, so a site owner has only the user-agent string and the robots.txt rule to work with.
Operator note. The documentation states: "Webmasters can disallow this user agent in their robots.txt file." Mistral publishes no IP address list for this crawler, unlike its other two bots. No crawl-rate, JavaScript-execution, or reverse-DNS details are documented.
Controlling MistralAI-Training with robots.txt
Use the token MistralAI-Training in robots.txt. Mistral AI documents that MistralAI-Training honors robots.txt directives.
User-agent: MistralAI-Training
Disallow: /User-agent: MistralAI-Training
Allow: /Verifying a request is really MistralAI-Training
Anyone can put MistralAI-Training in a User-Agent header. The operator does not document a verification method.
Common questions
Should I block MistralAI-Training?
That depends on whether you want your pages in Mistral's training datasets. Mistral says the crawler builds training data and is not used for search indexing or live user queries in Vibe, so the operator describes no search or Vibe cost to blocking it. Add a robots.txt rule for the MistralAI-Training token; Mistral states it honors that.
How do I verify a request from MistralAI-Training?
Mistral documents no way to verify this crawler. It publishes no IP list, and the URL analogous to the address lists for its other crawlers returns 404 for this one. Mistral gives no reverse-DNS suffix either. You can match the user-agent string, but the operator provides nothing to confirm the source of a request.
Does MistralAI-Training run JavaScript?
The operator's documentation does not say. Mistral's robots page covers the crawler's purpose and the robots.txt token but gives no detail on JavaScript execution, crawl rate, or request frequency, so assume nothing beyond the fact that it fetches web content.
Sources
Every fact on this page was checked against Mistral AI's own documentation, listed below, and re-checked by a second reviewer before publication. Reviewed September 2026.
Related crawlers
This registry documents how operators describe their own bots so site owners can identify and control them. It does not publish third-party IP lists or guess at undocumented behaviour. To see how your own site responds to automated visitors, the bot detection scanner reads a URL's live response and names the protection it finds.