What is MistralAI-Training?
Mistral · Training crawlers · User agent only
MistralAI-Training is Mistral's AI training crawler. Crawls web content to build datasets for training Mistral's generative AI models. Not used for search or live answers.
A visit usually means: Public content is being collected that may end up in model training data or large web datasets.
MistralAI-Training user agent
The full user agent string, as Mistral publishes it:
- User-Agent
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; MistralAI-Training/1.0; +https://docs.mistral.ai/robots)
Version numbers change, so match the token MistralAI-Training anywhere in the user agent, ignoring case. Source: Mistral's crawler documentation, checked Nov 18, 2026.
Reading the string
- The published version is MistralAI-Training/1.0. Expect the number to change.
- It starts with Mozilla/5.0, like a browser, so a filter that only checks the start of the string will miss it.
- It carries a link, https://docs.mistral.ai/robots, so anyone reading their logs can look the crawler up.
Paste a string from your logs into the user agent parser to see which crawler it is. New to the term? Read what a user agent is.
Block or allow MistralAI-Training in robots.txt
Mistral says MistralAI-Training follows robots.txt.
User-agent: MistralAI-Training
Disallow: /User-agent: MistralAI-Training
Disallow: /private/User-agent: MistralAI-Training
Allow: /A crawler follows the most specific group that names it, so a rule here overrides your User-agent: * group for this crawler. Well-behaved crawlers re-read robots.txt within about a day.
MistralAI-Training and Mistral's other crawlers
Mistral runs 3 crawlers in this directory, each with its own robots.txt token and its own job. What blocking each one costs you:
- MistralAI-Training · TrainingModel training and datasets; it sends no visitors
- MistralAI-User · AnswersMistral's assistant reading your pages when someone asks about them
- MistralAI-Index · IndexingYour pages in Mistral's AI search answers
Opt out of Mistral's AI training, stay in its search
Name only the training token. MistralAI-Index has its own token, so it keeps following your other rules.
User-agent: MistralAI-Training
Disallow: /MistralAI-Training IP ranges and verification
Mistral publishes no IP ranges for MistralAI-Training, so requests are identified by user agent only and recorded as unverified. Anyone can send this user agent.
What MistralAI-Training means for your traffic and revenue
MistralAI-Training collects pages for model training or datasets. It sends no visitors. Blocking it costs no traffic today; allowing it may help future models know about your product.
Most crawlers never run JavaScript, so browser analytics can't see them, and the ones that do shouldn't count as visitors. MetricRaven's tracker ignores MistralAI-Training, and server-side crawl tracking shows which pages it reads instead. See visits and revenue by AI company.
Questions
- What is MistralAI-Training?
- Crawls web content to build datasets for training Mistral's generative AI models. Not used for search or live answers. MetricRaven groups it under Training crawlers: crawlers collecting public pages that may be used to train AI models.
- What is the MistralAI-Training user agent?
- Mistral documents this user agent: "Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; MistralAI-Training/1.0; +https://docs.mistral.ai/robots)". Match the token "MistralAI-Training" rather than the whole string, because version numbers change.
- How do I block MistralAI-Training in robots.txt?
- Add "User-agent: MistralAI-Training" followed by "Disallow: /" to your robots.txt. Mistral says MistralAI-Training follows robots.txt.
- Does MistralAI-Training publish IP ranges?
- No. Mistral publishes no IP ranges for MistralAI-Training, so only the user agent identifies it, and anyone can send that string.
- Is MistralAI-Training safe?
- The real MistralAI-Training only requests public pages, and Mistral documents it. The risk is impostors using its name. Whether to allow it depends on what you get back: public content is being collected that may end up in model training data or large web datasets.
- What other crawlers does Mistral run?
- MistralAI-User (AI assistant) and MistralAI-Index (AI search). Each has its own robots.txt token, and each does a different job.
- Can I block Mistral's AI training but stay in its search?
- Yes. Add a "Disallow: /" group for MistralAI-Training. MistralAI-Index has its own token, so it keeps following your other rules.
See when AI assistants and crawlers read your site
Add one line to your server and MetricRaven shows which assistants, search engines and training crawlers request your pages, which ones are IP-verified, and the URLs they expected to find, including 404s.