Coaley Peak GEO Research Centre Engine guide
Mistral Vibe How Europe's own AI assistant finds, cites and learns from your pages.
Vibe, called Le Chat until May 2026, is the assistant from Mistral AI, Europe's best-known AI lab. It answers from Mistral's own models, many of them open-weight, and from a search index Mistral now builds itself. Here is its history, the four layers behind its answers, its three crawlers and how to opt out of each.
Independent guide: Coaley Peak is not associated with Mistral AI. Trademarks
- Owner
- Mistral AI
- Launched
- February 2024 as Le Chat; renamed Vibe in May 2026
- Home market
- France and Europe; offered worldwide
- Languages
- Multilingual; no official list published
- Open weights
- Many models, including Vibe's default, Mistral Medium 3.5
- Crawlers
- MistralAI-User, MistralAI-Index and MistralAI-Training
Overview
What Mistral Vibe is, and who uses it
Mistral AI was founded in Paris in April 2023 by Arthur Mensch, Guillaume Lample and Timothée Lacroix. It made its name with open-weight models, Mistral 7B in September 2023 and Mixtral 8x7B that December, and launched its assistant, Le Chat, in beta in February 2024. Mobile apps and a Pro tier followed in February 2025, and Le Chat passed a million downloads in about two weeks.
On 28 May 2026 Mistral renamed Le Chat to Vibe and added Work Mode, which carries out multi-step tasks across tools such as Google Workspace, Outlook, SharePoint, Slack and GitHub, and Code Mode. The address, chat.mistral.ai, and users' accounts and conversations carried over. Much third-party material still says Le Chat; Mistral's own documentation now says Vibe.
Vibe matters for GEO for two reasons. Many European organisations choose it because they want a European provider, and it is sold on Free, Pro, Team and Enterprise plans. And Mistral's models reach far beyond Vibe: its Apache 2.0 open-weight models, Mistral Large 3, Mistral Small 4 and Ministral 3 among them, are self-hosted inside other people's products, carrying what they learned in training with them.
Mistral is also one of the clearest providers to work with technically. It runs three separate crawlers, one for training, one for its search index and one for live user requests, each with its own robots.txt token and, for two of them, a published IP list. A site can refuse training and still be indexed and cited.
Timeline
Mistral Vibe, from launch to today
Arthur Mensch, Guillaume Lample and Timothée Lacroix found Mistral AI in Paris.
Mistral releases Mistral 7B as open weights under Apache 2.0.
Mixtral 8x7B, a sparse mixture-of-experts model, is released under Apache 2.0.
Le Chat launches in beta, alongside the Mistral Large model.
The UK's Competition and Markets Authority concludes that Microsoft's partnership with Mistral AI does not qualify for a merger investigation.
Mistral Large 2 is released under the Mistral Research Licence, with a commercial licence for business use.
Le Chat adds web search with citations, a Canvas editor and image generation.
Mistral signs an agreement with AFP that gives Le Chat access to AFP newswire stories, about 2,300 a day in six languages. It is Mistral's first content partnership.
Le Chat launches on iOS and Android with a Pro tier. Brave says the apps use the Brave Search API for live web results.
The EU's General-Purpose AI Code of Practice is published. Mistral AI is a signatory, and its copyright chapter commits signatories' crawlers to follow robots.txt.
Mistral 3 launches: Mistral Large 3, a 675-billion-parameter mixture of experts, and Ministral 3, both as open weights under Apache 2.0.
Mistral Medium 3.5 (128 billion parameters) is released as open weights under a modified MIT licence and becomes the assistant's default model.
Le Chat is renamed Vibe, adding Work Mode and Code Mode.
Mistral announces Mistral Large 4 in public preview. Its open weights are reported to be planned for 27 October.
Things to know
Things to know about Mistral Vibe
Use the current name: Vibe, formerly Le Chat. The address is still chat.mistral.ai, and Mistral's crawler documentation now refers to Vibe.
Mistral runs three robots.txt tokens: MistralAI-Training for training data, MistralAI-Index for its search index and MistralAI-User for live fetches. You can refuse training and still be indexed and cited.
Mistral publishes JSON IP lists for MistralAI-User and MistralAI-Index, but none for MistralAI-Training.
Its crawlers' IPs sit in Microsoft Azure address space (AS8075, according to RIPEstat), so allow-lists keyed to a Mistral-owned network will not match them.
Vibe's default model since May 2026 is Mistral Medium 3.5, an open-weight model under a modified MIT licence. Companies with more than US$20 million monthly revenue need a commercial licence to self-host it.
When a Vibe answer relies on a fetched page, it links to the source. Mistral says third-party content shown through web search or verified news is not used to train its models.
Brave said in February 2025 that Le Chat used the Brave Search API for live results. Mistral has since built its own index crawler and does not publish its current mix of sources.
Mistral's AFP agreement gives Vibe newswire stories in French, English, Spanish, Portuguese, German and Arabic.
Mistral does not disclose the datasets it trains on, for open or commercial models. Its privacy policy says it trains on publicly available internet data and third-party datasets.
On the Free and Pro plans, users' own chats are used for training unless they opt out; Enterprise is opted out by default. Under the EU Digital Services Act, Mistral reports fewer than 45 million monthly active recipients in the EU.
The four layers
How Mistral Vibe builds an answer
Every AI answer is made in four layers: what the model learned in pre-training, how it was tuned, what it retrieves live, and how it puts the answer together. Here is each one for Mistral Vibe, and what it means for a brand that wants to be cited.
Layer 1
Pre-training: the base (foundation) model
Mistral trains its own models and publishes many as open weights: Mistral 7B and Mixtral, then Mistral Large 3, Ministral 3 and Mistral Small 4 under Apache 2.0, and Mistral Medium 3.5, Vibe's default, under a modified MIT licence. Mistral does not disclose its training data. Its model cards give no knowledge cut-off, though its recommended system prompts give 'knowledge base last updated' dates of 1 October 2023 for Large 3 and Ministral 3, and 1 November 2024 for Small 4 and Medium 3.5.
What it means for you: Anything published about your brand after those dates is not in the weights, so Vibe can only know it through retrieval. And because the same open weights run in thousands of self-hosted products, what Mistral's base models learned about you travels well beyond Vibe. Base model targeting matters more here than for closed assistants.
Layer 2
Post-training: tuning and alignment
Mistral publishes little about tuning. Medium 3.5 combines instruction-following, reasoning and coding in one model, with the reasoning effort set per request, and the Magistral models are reasoning variants. Mistral also offers separate moderation models. On the Free and Pro plans, users' conversations are used to train future models unless they opt out.
What it means for you: Tuning shapes how confidently Vibe names brands and how much it hedges. Because consumer conversations feed future training, how people already talk about your brand in Vibe can, over time, shape what the next model believes about it.
Layer 3
Retrieval: live search
Le Chat added web search with citations in November 2024, and Brave said in February 2025 that it used the Brave Search API for live results. Mistral has since documented MistralAI-Index, a crawler that builds an index for Mistral's own search, and MistralAI-User, which fetches a page when a user's question needs it. The AFP agreement adds newswire stories, about 2,300 a day in six languages.
What it means for you: To be cited, your pages must be reachable by MistralAI-User and indexable by MistralAI-Index, and being well placed in Brave Search probably still helps. Mistral warns that blocking MistralAI-Index will likely reduce your chances of being cited, so allow both even if you refuse training.
Layer 4
The answer: what people read
When Vibe uses a web page, it includes a link to the source in its answer, and web and news results can show links, thumbnails or snippets. Vibe also carries out multi-step tasks in Work Mode, writes code in Code Mode, and has memory and voice, carried over from Le Chat.
What it means for you: Links in answers make Vibe a source of visits as well as mentions. Write pages whose key facts can be quoted on their own, with your brand named, in each language you sell in, so that a single sentence of yours can stand as the answer.
Crawlers
Mistral Vibe's crawlers and user agents
The user-agent tokens to look for in your server logs and to name in robots.txt.
| User-agent token | What it does | robots.txt |
|---|---|---|
| MistralAI-User Documentation | Fetches a page when a Vibe user's question needs it, and may cite it. Mistral says it does not crawl automatically and is not used to gather training data. User agent: Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; MistralAI-User/1.0; +https://docs.mistral.ai/robots). | Follows robots.txt: the MistralAI-User token governs which sites these fetches may reach. |
| MistralAI-Index Documentation | Crawls automatically to build the index for Mistral's own search, which helps answer Vibe questions. Mistral says this content is not used for generative AI training of any kind. | Documented as a robots.txt token. Blocking it will likely reduce citations, Mistral says. |
| MistralAI-Training Documentation | Crawls to build datasets for training Mistral's generative models, including its open-weight releases. Not used for search or live answers. | Follows robots.txt: Mistral says it can be disallowed. |
The user agents for MistralAI-Index and MistralAI-Training follow the same pattern as MistralAI-User, with their own token and version 1.0 in its place.
Crawl sources
Where Mistral Vibe crawls from, and how to verify it
Mistral publishes IP lists for two of its three crawlers. MistralAI-User requests come from the addresses in mistral.ai/mistralai-user-ips.json: 20.240.160.161, 20.240.160.1, 20.240.194.83 and 51.12.243.114. MistralAI-Index crawls from the addresses in mistral.ai/mistralai-index-ips.json, dated 19 April 2026: 135.225.57.92 and 20.240.192.194.
No list is published for MistralAI-Training, and Mistral documents no reverse DNS hostnames for any of its crawlers. RIPEstat places every listed address in Microsoft's AS8075, which is Azure. To verify a request, match the user-agent token and check the IP against the current JSON list; never rely on the user agent alone, which anyone can copy.
If your firewall or CDN challenges cloud-hosted traffic by default, it may be blocking Mistral too. Check that the published addresses are allowed before concluding that Vibe cannot see your site.
Opting out
How to opt out of Mistral Vibe, or limit what it uses
Refuse training only
User-agent: MistralAI-Training with Disallow: / stops future collection for training. It does not affect indexing or citations. Mistral says erasing data from models already trained is technically limited.
Official pageLeave the search index, with care
User-agent: MistralAI-Index with Disallow: / keeps your pages out of Mistral's search. Mistral warns that this will likely reduce your chances of being cited in Vibe answers.
Official pageStop live fetches
User-agent: MistralAI-User with Disallow: / stops Vibe fetching your pages when a user asks about them.
Official pageMake a GDPR request
For access, rectification, erasure or objection, use the privacy request form linked from Mistral's help centre, or write to Mistral AI, Attn: Privacy Team, 15 rue des Halles, 75001 Paris. Mistral warns that requests about model training have technical limits. Complaints can go to France's CNIL.
Official pageReport illegal or infringing content
Mistral's Digital Services Act notice form covers intellectual property infringement, defamation and other categories. Authorities have a separate address, dsa-official@mistral.ai.
Official pageKeep your own team's chats out of training
On the web, open the Admin panel, then Vibe, then Privacy, and turn off 'Allow your interactions to be used to train our models'. In the mobile app, untick 'Enable data sharing' under Data & Account Controls. The API has a separate setting.
Official page
Opting out works going forward only. Mistral's open-weight models that have already been downloaded keep whatever they learned, and Mistral says erasure from trained models is technically limited. Blocking MistralAI-Training costs you no visibility; blocking MistralAI-Index or MistralAI-User does.
How we help
How we get brands cited in Mistral Vibe
The Coaley Peak GEO Research Centre researches and implements GEO for SEO agencies and in-house teams, with foreign markets as our speciality.
Crawler policy that keeps you citable
We set robots.txt and firewall rules so MistralAI-User and MistralAI-Index can reach you, with training refused or allowed as you choose, and check the result against Mistral's published IP lists.
Visibility audit across your European markets
We ask Vibe the questions your customers ask, in French, German, Spanish and your other languages, and record whether you are named, cited and described correctly in each.
Pages Vibe can quote
We rework key pages so each states its facts plainly enough to be lifted into an answer with a link: product names, prices, specifications and claims, in every language you sell in.
Base model targeting for Mistral's open weights
Mistral's open models run inside thousands of other products. We probe them with retrieval switched off to see what they learned about you, and work on the sources future training draws on.
Measurement
We track citations and referrals from Vibe alongside ChatGPT, Gemini and the rest, market by market, so you can see what Europe's own assistant says about you and where it sends people.
More on our research, services and team at the Coaley Peak GEO Research Centre.
FAQ
Mistral Vibe GEO: questions
Is Le Chat the same as Mistral Vibe?
Yes. Mistral renamed Le Chat to Vibe on 28 May 2026, adding Work Mode and Code Mode. The address (chat.mistral.ai), accounts and conversations carried over, though many older articles and guides still say Le Chat.
Which robots.txt token stops Mistral training on my site?
MistralAI-Training. A group with User-agent: MistralAI-Training and Disallow: / stops future collection for training, without affecting MistralAI-Index or MistralAI-User, so your pages can still be indexed and cited in Vibe.
Does Mistral Vibe cite its sources?
Yes. Mistral says that when Vibe uses a web page it includes a link to the source in its answer. Web search and verified news results can also show links, thumbnails or snippets.
Where does Vibe get live information?
From MistralAI-User fetches, Mistral's own search index built by MistralAI-Index, and AFP newswire stories. In February 2025 Brave said Le Chat used the Brave Search API for live results; Mistral does not publish its current mix.
Are Mistral's models open source?
Many are open-weight. Mistral Large 3, Ministral 3 and Mistral Small 4 are Apache 2.0, and Mistral Medium 3.5, Vibe's default, uses a modified MIT licence that requires a commercial licence for companies with more than US$20 million monthly revenue. Some models remain closed.
Where do Mistral's crawlers come from?
Microsoft Azure address space. Mistral publishes JSON lists for MistralAI-User and MistralAI-Index, and RIPEstat places the addresses in AS8075. There is no published list for MistralAI-Training, and no reverse DNS scheme.
Want a free chat about Mistral Vibe?
Tell us which market and brand you are working on. If we can help, we will set up a free, no-obligation chat with Stephen or Ethan.
More engine guides
Related engines
Independence and trademarks. Mistral AI, Mistral, Le Chat and Vibe are trademarks of Mistral AI or its affiliates. Coaley Peak is independent: we are not associated with, endorsed by or a partner of Mistral AI. This guide is our own research from public documentation and reputable reporting, checked in October 2026.
Reviewed by Stephen Grindley, research centre lead. Last checked .
Document reference: ISO_webpage_geo_engine_mistral_v1
Last modified: 7 October 2026
GEO Research Centre·Engine guides·Mistral Vibe