The short answer
Le Chat, Mistral AI's assistant, is now called Vibe. Since 2026 it works in three modes: Vibe Work for multi-step tasks across connected tools, Vibe Code for coding agents, and the older Vibe Chat. Mistral, based in Paris, hosts data in the EU by default and asks for approval before sensitive actions. There is a free plan, and Work mode launched in May 2026 alongside Mistral Medium 3.5.
| Made by | Mistral AI, Paris, France |
|---|---|
| Free plan | Yes, with capped messages and searches, basic images and up to 5 scheduled tasks |
| Models | Mistral's own models, including Mistral Medium 3.5, Devstral for coding and Voxtral for voice |
| Best at | Agent work across connected tools with approvals, EU data hosting, coding in the same plan |
| Shows sources | Yes for Library answers, which cite their documents |
| Team plans | Team and Enterprise |
| Your data | EU-hosted by default. Individual plans train by default until you opt out; Enterprise is opted out by default. |
Le Chat is now Vibe: what changed in 2026
If you are looking for Le Chat, you are in the right place, but the name on the screen has changed. Mistral AI, the Paris-based AI company, announced on 28 May 2026 that Le Chat would become Vibe, and its help centre confirmed the switch on 12 August. Same address (chat.mistral.ai), same login, and your account, plan, conversations and saved content all carried over. In Apple's App Store the app is now listed as Vibe by Mistral (ex-Le Chat), so searching for either name finds it.
The rename came with a new shape. Vibe has three modes:
- Vibe Work, the main mode, takes a goal, breaks it into steps and uses your connected tools to get it done.
- Vibe Code runs coding agents, in the browser in isolated cloud sandboxes, in a command-line tool, or inside VS Code, JetBrains and Zed.
- Vibe Chat is the old turn-by-turn chat. Mistral said in May that it would soon retire Chat mode, but the August help article still lists it as available, so expect it to fade out.
On models, Work mode launched in preview on 22 May 2026 alongside Mistral Medium 3.5, a 128-billion-parameter open-weight model with a 256,000-token context, built for long coding and productivity tasks. The command-line coding tool runs on Devstral, and voice input uses Voxtral. Mistral's newest flagship, Mistral Large 4, arrived on 6 October 2026 as a preview for developers; Mistral has not said whether Vibe uses it, and it does not name the model behind each mode. Business context: Mistral raised a 3 billion euro funding round in September 2026, so it is not going anywhere soon.
What Vibe (Le Chat) is genuinely good at
Multi-step work across your tools, with a brake pedal
Ask Vibe Work to "find every unanswered client email from this week, draft replies and put a follow-up on my calendar for Friday" and it plans the job, shows its tool calls and progress as it works, and stops to ask before it sends a message, writes a document, creates an event or changes data. That approval step is the right default for people who are new to agents, because nothing irreversible happens without you.
Working from your own documents
Libraries let you build an indexed collection of your files and web pages, attach it to a task and get answers that cite where each point came from. Mistral says Library documents are kept in its European cloud and removed permanently when you delete the Library or your account. Projects, custom instructions and a Knowledge Base (which replaced the old Memories) keep context from one task to the next.
Repeating jobs on a schedule
Scheduled tasks run a prompt once or on a daily, weekly, monthly or yearly rhythm. A Monday briefing on your sector, a monthly summary of a shared folder or a Friday status note are natural fits. Even the free plan allows up to five.
Coding in the same subscription
Vibe puts a coding agent in the same plan as your everyday work: start a remote coding session in the browser that can end in a pull request, or use the CLI and editor extensions. For a team that writes some code and a lot of documents, one subscription covers both.
Keeping data in Europe
Mistral's help centre says your data is hosted in the European Union by default. Some features can send data temporarily outside the EU through subprocessors that Mistral lists in its Trust Center, and Enterprise customers can switch some of those features off. For European teams under GDPR pressure, this is often the deciding reason to pick Vibe.
Where it falls short
- Training is on by default for individuals. Free and Pro users have to opt out themselves. Details below.
- The product is still moving. In six months it changed name, added modes and announced the end of another. Menu names in older tutorials are often wrong, and you may meet both "Le Chat" and "Vibe" in the same week.
- Little transparency on models. Mistral says Vibe runs on its flagship models but does not say which one powers a given mode, which makes it hard to know what you are getting on each plan.
- Not a research engine. Vibe can search the web and read a URL, and Deep Research exists as a skill, but for quick cited answers Perplexity is still the stronger habit.
- Zero data retention is not offered for Vibe Work or Chat on any plan, so if you need nothing stored at all, look at a self-hosted model instead.
How to get better results from Vibe
- Describe the outcome, not the steps. Work mode is built to plan. "I need a one-page summary of last quarter's support tickets for the leadership meeting" works better than a list of instructions.
- Connect only what the task needs. Work can ask you to connect a missing tool mid-task. Connect the ones this job needs and leave the rest unconnected.
- Read the plan before you approve. Vibe shows what it intends to do. A five-second check at that point saves you from a wrong email going out.
- Give it a Library for anything recurring. If you answer the same questions from the same documents, put those documents in a Library once and attach it every time.
- Turn good prompts into skills or scheduled tasks. When something works, save it so you do not rewrite it next week.
- Check facts and figures. Like every assistant, Vibe can be confidently wrong. Ask it to cite its source and open the important ones.
Our free lessons at /learn?tool=lechat walk through each habit in a few minutes.
Privacy settings in Vibe and where to find them
Mistral's help centre says that for individual Vibe users, input and output data are used to train its models by default until you opt out. Enterprise customers are opted out by default, and on Team plans an administrator can switch training off for the whole organisation. To opt out:
- On iPhone or Android: open Settings, then under Account choose Data & Account Controls and untick Enable data sharing.
- On the web: the help centre describes the switch in the admin panel. Under Manage, choose Vibe, then under Privacy switch off Allow your interactions to be used to train our models.
Three details people miss. The opt-out also covers your memories. The API has its own separate toggle, so developers need to switch both. And giving a thumbs up or down with a comment authorises Mistral to use that rating, prompt and answer on any plan, so skip the comment on sensitive chats.
Which Vibe plan suits whom
Free is a real starting point: everyday tasks, web search, connectors and a few scheduled jobs, with capped messages and only basic image generation.
Pro suits anyone who uses it daily. You get more messages and searches, longer thinking, deep research reports, many more images and all-day coding. Verified students can get the Education plan, a version of Pro for up to 12 months that adds spreadsheet and slide generation.
Team is the plan for a company that wants a shared workspace, admin controls, more storage and the ability to switch off training for everyone at once. Enterprise adds private or on-premises deployment, custom models, audit logs and SAML single sign-on, and is opted out of training by default. Prices are on Mistral's pricing page.
Who should skip it: people who mostly need quick, cited research answers will be happier with Perplexity, and anyone who needs a guarantee that nothing is stored should know that zero data retention is not available for Vibe Work or Chat. Teams that cannot accept any training on individual accounts should buy Team or Enterprise seats rather than letting staff sign up for personal plans, because each personal account has to opt out on its own and it is easy to forget. Finally, if you depend on one specific named model, Vibe is harder to plan around, since Mistral does not publish which model sits behind each mode.
Pick Vibe if EU data hosting matters to you, or if you want one subscription for both documents and code. Pick ChatGPT if you want the widest set of features, and Perplexity if most of your work is research. See our comparison tool, the ChatGPT alternatives list and tool stacks for how people combine them. If you are weighing it against a free model from China, read our DeepSeek guide, which explains why data location matters.
Prompts to copy
Clear your inbox with approvals
Using my email connector, find every email from this week that asks me a question and has no reply yet. Draft a short reply to each, list them for me to review, and do not send anything until I approve each one.
You get: Draft replies waiting for your sign-off
Prepare a meeting brief from your files
Using the [Library or folder name], prepare a one-page brief for my meeting with [person or client] on [date]. Include the open issues, the last decisions made with dates, and three questions to raise. Cite the document for every point.
You get: A sourced brief built from your own documents
Summarise a long document for leaders
I have uploaded [document]. Summarise it for [audience] in under 300 words: the decision it asks for, the three strongest arguments, the main risk and what happens if we do nothing. Flag anything that looks inconsistent.
You get: A decision-ready summary with red flags
Schedule a weekly sector briefing
Every Monday at 8 a.m., search the web for news from the past seven days about [sector or competitor]. Give me five bullet points with a source link each, newest first, and put anything about regulation at the top.
You get: A recurring Monday briefing as a scheduled task
Turn notes into a slide outline
Here are my notes for a [length]-minute presentation to [audience]: [paste notes]. Turn them into a 10-slide outline with a headline and three bullets per slide, and suggest one chart I should add.
You get: A slide-by-slide outline ready for your deck tool
Analyse a spreadsheet
I have uploaded [spreadsheet]. Tell me the three biggest changes between [period A] and [period B], what might explain each one, and draw a chart of the trend. Show the numbers you used.
You get: Changes, explanations and a chart with the working shown (charts use Code Interpreter, on paid plans)
Build a reusable checklist skill
Create a reusable skill called '[name]' that reviews any [document type] against this checklist: [paste checklist]. It should return a table with each item, pass or fail, and a one-line fix for each fail.
You get: A skill you can run on every new document
Fix a small piece of code
In Vibe Code, open [repository] and fix this issue: [describe bug]. Write a test that fails before the fix and passes after it, explain the change in three sentences and open a pull request for my review.
You get: A tested fix waiting as a pull request