MistralAI-Index user agent, IP ranges and robots.txt

Mistral AI operates MistralAI-Index. Per its documentation, the bot crawls the web for indexing only, feeding Mistral search, which helps answer user questions in Vibe; crawled content is not used for generative AI training. The robots.txt token is MistralAI-Index; per-bot compliance is undocumented. Site owners verify visits against Mistral's published IP range file; no reverse DNS method is documented.

In server logs the bot identifies itself with the token MistralAI-Index inside a Mozilla/5.0 AppleWebKit-style string, ending with a link to Mistral's robots documentation page at docs.mistral.ai/robots. Mistral describes the visits as automated crawling for indexing purposes only. The pages it fetches go into the index behind Mistral search, which the operator says helps answer user questions in its Vibe product. Mistral states that content crawled by this bot is not used for generative AI training of any kind, which separates it from a training crawler in the operator's own framing.

Because the category is AI search indexing, the trade-off is visibility inside Mistral search results and Vibe answers. Allowing the bot keeps a site eligible to appear there. A Disallow rule under User-agent: MistralAI-Index is the documented way to opt out of that index, though Mistral publishes no sentence confirming that this bot obeys it, so a site that needs a firm block should also filter by IP. The documentation gives no crawl-rate figure, so the request volume a site should expect is unknown. Whether the bot renders JavaScript is also undocumented.

Two documented points help with control. Robots.txt support rests on Mistral's general statement that it uses specific robots.txt tags; no sentence promises that MistralAI-Index itself complies, so treat compliance as undocumented rather than confirmed. Verification is by IP only. Mistral publishes a JSON file at mistral.ai/mistralai-index-ips.json with a creationTime field and a prefixes array of ipv4Prefix objects. A request from an address outside that list is not the documented bot.

Operator note. As with MistralAI-User, robots.txt control rests only on the documentation's general statement that Mistral "uses specific robots.txt tags"; there is no per-bot compliance sentence, and no crawl-rate or JavaScript-execution details are documented. The published IP file is dated 2026-04-19 and lists individual IPv4 addresses (/32 prefixes) only.