AI agent directory / MistralAI-Training
MistralAI-Training
Crawler from Mistral
Crawls web content to build datasets for training Mistral's generative AI models.
MistralAI-Training is the crawler Mistral uses to collect web content for datasets that train its generative AI models. Mistral says webmasters can disallow it in robots.txt.
Mistral states that this crawler is not used for search indexing or to answer live user questions in Vibe. Those jobs belong to MistralAI-Index and MistralAI-User, which have their own tokens.
| Run by | Mistral |
|---|---|
| Type | Crawler |
| Purpose | Training |
| Acts for | Its operator |
| robots.txt token | MistralAI-Training |
| Documentation | Mistral's documentation |
MistralAI-Training user-agent string
Full user-agent string (source):
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; MistralAI-Training/1.0; +https://docs.mistral.ai/robots)
Match on MistralAI-Training; the rest of the string changes between versions.
How Airglow identifies it
- DeclaredIt names itself in the User-Agent:
MistralAI-Training.
Every visit in Airglow says which of these applied, so you can decide how much to trust it.
Seeing it on your site
It usually reads pages without running JavaScript, so Airglow counts every visit through a server connection (Cloudflare). When it does run JavaScript, the script sees and replays it too.
Allowing or blocking MistralAI-Training
MistralAI-Training reads the MistralAI-Training group in your robots.txt. To keep it out of the whole site:
User-agent: MistralAI-Training Disallow: /
Mistral runs a separate bot for each purpose, so blocking MistralAI-Training doesn't block its other agents listed below.
Airglow shows, for every agent that visits, whether your robots.txt lets it in, and helps you write rules by purpose (training, search, acting for a person).
Mistral's three robots.txt tokens
Mistral documents three separate user agents, each with its own robots.txt token:
MistralAI-Trainingcrawls to build datasets for training Mistral's generative models. Not used for search or live answers.MistralAI-Indexcrawls automatically to build the Mistral search index. Not used for training.MistralAI-Userfetches a page when a Vibe user asks a question. Not used for automatic crawling or training.
Because the tokens are separate, you can block training and still let your pages appear in Mistral search and Vibe answers.
No published IP list
Mistral publishes IP address lists for MistralAI-User and MistralAI-Index, but its documentation lists none for MistralAI-Training. That means there is no operator-documented way to confirm a request really comes from Mistral. Treat the user agent as a claim.
Questions
What does MistralAI-Training do?
It crawls web content to help build datasets for training Mistral's generative AI models, according to Mistral's crawler documentation.
What happens if I block MistralAI-Training?
Mistral's training crawler should stop collecting your pages. Blocking it does not block MistralAI-Index or MistralAI-User, so your pages can still show up in Mistral search and Vibe answers.
Does blocking MistralAI-Training remove me from Vibe answers?
No. Mistral says MistralAI-Training is not used for search indexing or live user queries. Vibe relies on MistralAI-Index and MistralAI-User, which you control with their own robots.txt tokens.
Does Mistral publish IP addresses for MistralAI-Training?
Not as of November 2026. Mistral's documentation links IP lists for MistralAI-User and MistralAI-Index only.
Other agents from Mistral
Sources
Last reviewed November 10, 2026.
See when MistralAI-Training visits your site
Airglow names every AI agent on your website, replays what it did and links the visitors it sends you. Free for one website.