Mistral · Training
MistralAI-Training
Dataset construction is the declared function: Mistral says it "crawls web content to help build datasets for training Mistral generative AI models," and rules out the other two roles its siblings hold: "This crawler is not used for search indexing or to answer live user queries in Vibe." What makes it worth a page of its own is the verification asymmetry inside one vendor's documentation. `MistralAI-User` and `MistralAI-Index` each carry a published IP endpoint; this section carries none. So the token that decides whether your text enters a training set is the one token of the three that cannot be network-verified: a user-agent string is all there is to match on, and a user-agent string is a claim anybody can type.
- Operated by
- Mistral
- Purpose
- Training
- robots.txt token
- MistralAI-Training
- Verification
- User agent only
Collects pages into a corpus used to train models.
How to verify MistralAI-Training
Mistral publishes no list of IP addresses for MistralAI-Training, so nobody (us included) can prove that a request carrying this user agent really came from Mistral. The name is trivial to copy. Treat it as a claim the visitor is making about itself, not as an identity anyone has checked.
There is no range list to confirm. This entry was last reviewed against Mistral's own documentation on .
User agent
Mistral publishes this user agent for MistralAI-Training. Match on the MistralAI-Training product token rather than the whole string: vendors revise the surrounding version and URL fragments without notice.
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; MistralAI-Training/1.0; +https://docs.mistral.ai/robots)robots.txt for MistralAI-Training
Yes, and stated as a direct instruction rather than a policy: "Webmasters can disallow this user agent in their `robots.txt` file." Given that no IP file exists for this token, that directive is not merely the recommended control: it is the only one Mistral documents.
Block
User-agent: MistralAI-Training
Disallow: /Allow
User-agent: MistralAI-Training
Allow: /robots.txt is a request, not an enforcement mechanism. It is honoured by convention, and a crawler that ignores it is stopped at your edge, not in a text file.
What blocking MistralAI-Training costs you
Nothing visitor-facing moves. No citation slot closes, no Vibe answer loses a source link and no search placement changes, because those outcomes belong to two other tokens with two other directives: this one only determines whether your material can be folded into a training corpus, which returns no referrals and no attribution by construction. Set against that, allowing it is a decision you cannot audit: with no published ranges, you have no way to confirm that traffic bearing this name is Mistral rather than someone borrowing the string, and no way to distinguish a compliant crawl from an imitation of one.