MistralAI-User: what the Le Chat and Vibe fetcher does
MistralAI-User is the fetcher Mistral's assistant uses when someone asks it about a page. What Mistral documents, how robots.txt applies and how to verify it.
By Crawlito · 6 min read
In short
- MistralAI-User fetches pages when a user of Mistral's assistant, Le Chat, now renamed Vibe, asks a question that needs a live web page.
- Mistral says the MistralAI-User robots.txt token governs which sites these user requests can be made to, and that it is not used for automatic crawling or training.
- Mistral runs two other agents with separate tokens: MistralAI-Index for its search index and MistralAI-Training for model training data.
- Mistral publishes the IP addresses MistralAI-User uses at mistral.ai/mistralai-user-ips.json, so you can check a request claiming to be it.
- Blocking MistralAI-Training does not block MistralAI-User or MistralAI-Index, because each token needs its own robots.txt group.
On this page
- What MistralAI-User is and when it visits
- The full user-agent string and robots.txt token
- Mistral's three agents compared
- How to allow or block MistralAI-User in robots.txt
- How to verify a request really comes from Mistral
- Why MistralAI-User might fail to read your page
- Frequently asked questions
- Check what Mistral's fetcher receives
MistralAI-User is the user agent Mistral's assistant uses when a person asks it a question and it needs to open a web page to answer, then link to that page as a source. It is a user-triggered fetcher, not a crawler: Mistral says it is not used to crawl the web automatically or to collect training data. You control it with its own robots.txt token, MistralAI-User, separately from Mistral's search and training crawlers.
The assistant was called Le Chat until 28 May 2026, when Mistral announced that Le Chat is now Vibe. Mistral's crawler documentation now describes MistralAI-User as the agent for user actions in Vibe. The token and the user-agent string did not change with the name, so rules you wrote for Le Chat still apply.
What MistralAI-User is and when it visits
Mistral's crawler page lists three agents. MistralAI-User is the one that acts for a single person in a single conversation. When someone asks Vibe something that needs a current page, such as "what does this pricing page say" or a question about recent news, the assistant may fetch the page to help answer and include a link to the source.
That has two practical consequences for site owners:
- Visits follow user questions, not a schedule. You will see MistralAI-User on the pages people ask about, in bursts, and rarely on pages nobody mentions. A quiet log does not mean you are blocked.
- It is the agent closest to a click. A MistralAI-User fetch means a person was asking about your page right then, and the answer can carry a link back to you.
Mistral states that MistralAI-User is not used for crawling the web in any automatic fashion, and not for generative AI training. Those statements are Mistral's own and cannot be checked from outside, as with every operator's description of its own bots.
The full user-agent string and robots.txt token
The string Mistral documents for this agent is:
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; MistralAI-User/1.0; +https://docs.mistral.ai/robots)The part you write in robots.txt is the product token, MistralAI-User. Matching is case-insensitive under RFC 9309, so mistralai-user works too, but copying the documented spelling keeps the file readable.
Mistral's three agents compared
Mistral splits its bots by purpose, like OpenAI, Anthropic and Perplexity do. The training vs search vs user crawlers guide explains why that split matters for your policy.
| Token | Purpose per Mistral | What blocking it costs |
|---|---|---|
MistralAI-User | Fetches pages for user questions in Vibe and links to the source | Vibe cannot open your pages when users ask about them |
MistralAI-Index | Automated crawling for Mistral's search index, which helps answer questions in Vibe; Mistral says this content is not used for generative AI training | Your pages are less likely to be in the index Vibe searches |
MistralAI-Training | Crawls content to build datasets for training Mistral models; not used for search or live answers | Opts your content out of Mistral's training crawl, with no documented effect on search or user fetches |
Mistral's page includes an explicit sentence for MistralAI-Training only: webmasters can disallow it in robots.txt. For MistralAI-User the page says the token governs which sites user requests can be made to, which is the same kind of control expressed for a fetcher. For MistralAI-Index the page lists the token and the string but makes no separate robots.txt statement. Mistral's introduction says it uses these robots.txt tags so webmasters can manage how their sites interact with AI.
The full list of agents across operators is in our AI crawler user agents list.
How to allow or block MistralAI-User in robots.txt
Each token is its own group, and a named group does not inherit anything from User-agent: *. Our guide to robots.txt user-agent groups explains that rule in detail, because it is where most mistakes happen.
Allow user fetches and search, opt out of training
This is the common choice for sites that want to stay reachable from AI assistants but keep content out of training sets:
User-agent: MistralAI-Training
Disallow: /
User-agent: *
Disallow: /admin/
Disallow: /cart/MistralAI-User and MistralAI-Index are not named, so they follow the * group and are kept out of /admin/ and /cart/ only. There is no need to add an empty group that "allows" them.
Block MistralAI-User from one section
If you want Vibe to open your public pages but not, say, a members area, name the token and repeat every general rule inside its group:
User-agent: MistralAI-User
Disallow: /admin/
Disallow: /cart/
Disallow: /members/Leaving out /admin/ and /cart/ here would open those paths to MistralAI-User, because it stops reading * once a group names it.
Block all three Mistral agents
User-agent: MistralAI-User
User-agent: MistralAI-Index
User-agent: MistralAI-Training
Disallow: /Think about the trade-off first. Blocking the user fetcher means a person who pastes your URL into Vibe gets an answer without your page in it. Our post on whether you should block AI crawlers works through that decision.
How to verify a request really comes from Mistral
Anyone can send a user-agent string. Mistral publishes the addresses its fetcher uses, which is how you tell real traffic from imitations:
| Agent | IP list |
|---|---|
| MistralAI-User | mistral.ai/mistralai-user-ips.json |
| MistralAI-Index | mistral.ai/mistralai-index-ips.json |
| MistralAI-Training | No list on Mistral's page at the time of writing |
The files use the same layout as other operators' lists: a prefixes array of ipv4Prefix entries. At the time of writing the MistralAI-User file holds a handful of single addresses (/32 prefixes), so expect it to change and fetch it again rather than hard-coding it.
A quick check from the command line:
curl -s https://mistral.ai/mistralai-user-ips.json \
| jq -r '.prefixes[].ipv4Prefix'Then compare the client IP from your log line against that list. A request with MistralAI-User in the string from an address outside the list is not Mistral's fetcher, whatever it claims. The general method, including reverse DNS for operators that support it, is in how to verify real AI crawlers.
Find MistralAI-User in your logs
grep -i "MistralAI-User" access.log | awk '{print $1, $7, $9}' | sort | uniq -c | sort -rn | headThat prints the client IP, the path and the status code, grouped. Look for 403 and 429 responses on pages you meant to allow. The guide to AI crawlers in server logs covers the log formats and what each status means.
Why MistralAI-User might fail to read your page
If Vibe says it cannot access a page that robots.txt allows, the cause is usually somewhere else in the request path:
- A WAF or bot rule blocks it. Bot protection products often treat unfamiliar agents as unverified. If you use an IP allowlist for good bots, add Mistral's published addresses and refresh them on a schedule.
- A challenge page is served. A fetcher that does not solve JavaScript challenges receives the challenge HTML, not your content, even with a
200status. - The content is rendered in the browser. Mistral does not document whether its fetcher runs JavaScript. If your page ships an empty shell and builds the text with JavaScript, assume a fetcher may see only the shell. Test with
curland read the raw HTML. - Rate limits. A user question can trigger several fetches in a short time. A strict limit can return
429to the second or third request.
You can reproduce the first and third problems yourself by requesting the page with the documented string:
curl -s -o /dev/null -w "%{http_code}\n" \
-A "Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; MistralAI-User/1.0; +https://docs.mistral.ai/robots)" \
https://www.example.com/pricingThis tests rules keyed on the user-agent string. It does not reproduce rules keyed on IP address, because the request comes from your machine, not Mistral's.
Frequently asked questions
Is MistralAI-User the same as Le Chat's crawler?
Yes. Le Chat was renamed Vibe in May 2026, and Mistral's documentation now says MistralAI-User is for user actions in Vibe. The token and user-agent string stayed the same.
Does blocking MistralAI-Training also block MistralAI-User?
No. Each token is matched separately in robots.txt. A group for MistralAI-Training applies only to that crawler, and MistralAI-User follows its own group or, if none names it, the * group.
Does MistralAI-User use my content for training?
Mistral says it does not: the agent is not used to crawl content for generative AI training. Training collection is done by MistralAI-Training, which you can disallow on its own.
Does MistralAI-User run JavaScript?
Mistral does not document it. Serve the main text of each page in the initial HTML so any fetcher can read it, whether or not it renders scripts.
Where can I find Mistral's IP addresses?
Mistral publishes JSON files for MistralAI-User and MistralAI-Index, linked from its crawler page. There is no list for MistralAI-Training on that page at the time of writing.
Check what Mistral's fetcher receives
Robots.txt is only the first gate. A firewall rule, a challenge page or a JavaScript-only page can stop MistralAI-User even when your file allows it. Run the free check on one of your key pages to see what an AI fetcher receives, with no account needed.
Filed under AI crawlers. Topics: ai-crawlers, robots-txt, server-logs, bot-protection