AGENT PROFILES

Mistral Vibe

Mistral AI · General-purpose agents · Coding agents

Mistral’s assistant and agent for work and code (formerly Le Chat) on the web, in the app, in the terminal and in the editor.

Typical tasks

Availability in Europe
Added to the catalog on Mistral’s assistant and agent for work and code (formerly Le Chat) on the web, in the app, in the terminal and in the editor.Details
01

What it does

7 features

Mistral Vibe · Assistant and agent for work and code

Product and access

Vibe is the new name of Le Chat; conversations, settings and plans carried over. According to the documentation, the open-source mistral-vibe CLI is the tool of Vibe Code and therefore belongs in this profile.

Core features

7 features from provider sources · checked on 5 November 2026. Availability and limits are directly below.

  • Vibe Work for research and documents

    According to the product page, Vibe researches the web and internal documents, analyses data, drafts documents and decks and uses over 100 connectors.

    Source 1
  • Vibe Code: CLI, editor, web

    The CLI and VS Code extension work on a local checkout, Vibe Code Web on a GitHub repository in a remote sandbox.

    Source 7
  • Scheduled tasks

    A saved prompt runs once or daily, weekly, monthly or yearly. The feature is in public preview and runs in Work mode only.

    Source 11
  • Knowledge Base

    The Knowledge Base (successor to Chat Memories, Work mode only) keeps details from conversations and can be viewed, changed through chat, deleted and exported.

    Source 10
  • Approval modes in the CLI

    The profiles ask, accept-edits (default), smart-approve and auto-approve control what Vibe does without asking. Project configuration loads only from trusted folders.

    Source 8
  • Skills, MCP and subagents

    You extend the CLI with skills, MCP servers, AGENTS.md and custom subagents; a read-only explore subagent is built in.

    Source 5
Show all 7 features
  • Remote coding sessions

    Vibe Code Web runs tasks in a managed cloud sandbox and ends each session with a branch or pull request for review.

    Source 12

Usable AI models

Open choice

In the terminal you use Mistral models by default and pick them by alias. According to the docs, the Vibe Code CLI also works with any model behind an OpenAI-compatible endpoint, including on your own machine. For web and app, the pages checked name only Mistral models and no model choice of your own.

Model families named

  • Mistral
  • Other models can be connected freely
Model choice
You choose
Your own API key
Yes
Router or custom endpoint
Yes
Local models
Yes

The rating covers the generally available main access. Partly documented · Sources accessed on

Mistral: Using offline models · GitHub: mistral-vibe · Mistral: Pricing · Mistral: Vibe Work, get started

02

Does it fit my task?

2 practical tasks · 7 requirements and limits

A good fit if …

  • you want an assistant for research and documents and a coding agent under one account
  • you want to use an open-source terminal CLI with a Mistral account or your own API key
  • you prefer a European provider with data hosted in the EU by default

Less suitable if …

  • you need zero data retention for Vibe conversations
  • you expect unattended tasks without pre-authorised connector actions
  • you expect memory in the CLI as well

Editorial assessment based on provider documentation. Not a product test.

Practical tasks

Editorial examples based on documented features. No product tests performed.

01

Weekly research with report

Your input
A prompt with source guidance, set up as a scheduled task in Work mode.
Possible result
The prompt runs at the chosen time and the result appears as a notification in the sidebar.
What to check
Read the result and sources; pre-authorise only read actions for unattended runs.
02

Code change in the CLI

Your input
A trusted project folder with AGENTS.md and a limited task.
Possible result
Vibe changes files and asks about shell commands.
What to check
Check the diff and tests; use auto-approve only in isolated environments.
EDITORIAL EXAMPLE

Set up a weekly research task as a scheduled task.

Your possible prompt

Compare three providers for [task]. Use their official documentation. Show differences, sources, and open questions in a table.

Adjust the task, data and permissions.

Requirements and limits

  • Requirement

    Account or API key

    For Mistral-hosted models the CLI signs in with a Mistral account through the browser or with an API key, and this also works in Free mode. A Mistral account is optional for local models or other OpenAI-compatible keys.

    Source 6
  • Requirement

    Python 3.12 for manual installation

    The CLI needs a GIL-enabled build of Python 3.12 or later; uv can install it. On Windows it works per the README, but UNIX is the officially supported target.

    Source 5
  • Limitation

    Scheduled tasks: preview and pre-authorisation

    Sensitive connector actions such as sending email need a pre-granted Always allow for unattended runs. The page recommends read-only prompts.

    Source 11
  • Limitation

    Smart approve is not a sandbox

    According to the safety page, smart approve is a classifier, not a sandbox. For untrusted code it recommends the ask profile or an isolated environment.

    Source 8
  • Limitation

    Zero data retention does not cover Vibe conversations

    The page excludes Vibe Work and Chat because they store conversation data. Organisations on paid plans can request ZDR from support.

    Source 13
  • Not yet confirmed

    Model training on your data

    The pricing page lists Opt-out three times in its model training row without it being clear which plan each entry belongs to.

    Source 4
  • Not yet confirmed

    Remote agents, hooks and teleport by plan

    The pricing page does not say which plan includes these features.

    Source 4
03

Can I use it?

With requirements

Europe & regional availability

Help articles on storage location and retention, the product page and documentation reviewed. The storage location article does not name Vibe explicitly. The trust centre and privacy policy could not be evaluated.

Reviewed scope: Cloud assistant with a local CLI; Mistral AI is a European provider
Reviewed on:

  • With requirements

    Data hosted in the EU by default

    The help article on storage location says data is hosted in the EU by default; if you choose the US API endpoint, it is hosted in the United States. Depending on the feature, data may temporarily go to subprocessors outside the EU. Enterprise customers can have some of these features turned off at organisation level. The article does not name Vibe or Le Chat explicitly.

    Region, reason and source

    Region covered: EU/EEA

    Documented reason: Statement on storage location with exceptions; the article does not name Vibe.

    Official source · help.mistral.ai · Official source · mistral.ai

  • Availability still unclear

    Availability by country: no statement

    The reviewed pages give no country list for Vibe. The product page mentions full data residency for enterprise customers deploying on-premises, in a private cloud or on Mistral Cloud, without naming a location.

    Region, reason and source

    Region covered: EU/EEA, United Kingdom and Switzerland

    Documented reason: The reviewed sources name no country release for this.

    Official source · mistral.ai · Official source · help.mistral.ai

Europe and regulation explained

Accounts & permissions

Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.

Configuration covered: Mistral account for web and CLI, alternatively an API key

Sources reviewed:

Show all 3 points
  • Documented

    Account used

    Vibe opens at chat.mistral.ai with the existing Le Chat login. The CLI signs in through the browser with a Mistral account (recommended) or takes an API key. Enterprise lists SAML SSO.

    Evidence · Evidence · Evidence
  • Documented with conditions

    Admin approval

    Administrators can disable auto-approve and turn off the Knowledge Base organisation-wide. For enterprise organisations, admins can set the date of the merge of Chat and Work themselves.

    Evidence · Evidence · Evidence
  • Documented

    Data access & team sharing

    Work asks before sensitive actions such as sending email. Vibe Code asks before shell commands and by default applies file edits without asking. Access outside the working directory and to .env files requires confirmation.

    Evidence · Evidence · Evidence
04

Memory and conversation continuity

4 of 6 criteria documented

Vibe Work keeps details in a visible Knowledge Base that you can view, delete and export; the CLI resumes sessions but describes no memory.

Knowledge Base and conversations in Vibe Work; session logs of the CLI

Sources accessed:

Documentation by criterion

StorageDocumented

Storage and scope

Vibe stores conversation data so you can use your history; the Knowledge Base holds entries with a title and description. The CLI writes session logs under ~/.vibe/logs/ when session logging is enabled.

SelectionDocumented

Selection and timing

Entries are created explicitly (“remember”, “save”) or implicitly when Vibe detects recurring useful details. For the CLI, the README describes no memory feature.

Edit & deleteDocumented

Inspect, edit and delete

You see each entry, cannot edit the text directly but ask Vibe in chat to change it, and delete a topic through the menu. Deleting everything is irreversible; settings allow an export.

Access boundariesPartly documented

Agent and channel boundaries

The Knowledge Base exists in Work mode only; admins can disable it organisation-wide. The page describes no split by person, workspace or project. The CLI’s project configuration loads only from explicitly trusted folders.

RetrievalDocumented

Retrieval

Vibe reviews the entries at each turn to decide which are relevant; retrieval is not deterministic. A read or write pill shows accesses. In the CLI, --resume (picker) and --continue reopen earlier sessions.

New conversationsPartly documented

Session reset and compaction

In the CLI, /clear starts a new conversation; /compact compacts it while, per the README, the same session stays. For Vibe Work, the reviewed page describes no such command.

05

Cost and effort

4 plans · Prices as of 5 November 2026

Prices and included usage

Pricing checked

Free costs nothing, Pro 14.99 US dollars a month excluding taxes, Team 24.99 per user per month; Enterprise is on request.

PlanPriceIncluded
Free0 US dollars

Limited messages, web searches and coding sessions; basic image generation and editing; 100+ connectors; web and mobile.

Pro14.99 US dollars per month (excl. taxes)

Higher limits, faster responses, all-day coding in CLI, IDE or web, 15 US dollars of API credits a month.

Team24.99 US dollars per user per month (excl. taxes)

Higher limits, up to 30 GB storage per user, domain verification, data export.

EnterpriseOn request

Custom models, agents and workflows, audit logs, SAML SSO, white label, custom deployments (self-hosted, private cloud or on-premises).

Price source

  • Besides 24.99 US dollars for Team, the pricing page also lists 50 US dollars a month without making clear what it applies to. Pro for verified students costs 5.99 US dollars per the page.
  • Beyond the limits you can continue with pay-as-you-go credits at API rates.

Setup

Sign in on the web with a Mistral account. For code you install the CLI and sign in through the browser or with an API key.

Operation & oversight

Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured.

Deployment covered: Web app, CLI and Vibe Code Web

Sources reviewed:

Editorial assessment

Approvals are finely adjustable, but autonomy depends on your profiles and pre-authorisations. For scheduled tasks, prefer read-only prompts and check the results.

Show all 6 points
  • Documented

    Files and working state

    Conversations and the Knowledge Base persist in Vibe Work, and CLI sessions can be resumed if session logging is enabled. On Vibe Code Web, past sessions can be inspected but not resumed once the sandbox is deprovisioned.

    Evidence · Evidence · Evidence
  • Documented

    Sign-ins

    The CLI signs in through the browser or stores an API key, per the README in ~/.vibe/.env. The reviewed pages do not say how long a sign-in lasts.

    Evidence · Evidence
  • Documented

    Starting further runs

    Scheduled tasks run once or daily to yearly without you, in Work mode and as a public preview. The CLI runs programmatically with vibe --prompt, Vibe Code Web asynchronously in a cloud sandbox.

    Evidence · Evidence · Evidence
  • Documented

    Human intervention

    Sensitive actions need confirmation; for unattended scheduled tasks you must pre-authorise connector actions with Always allow. In programmatic mode Vibe denies calls that would need approval unless you pass --auto-approve.

    Evidence · Evidence
  • Open in reviewed sources

    Detecting errors

    The scheduled tasks page describes notifications when a result is ready; the retrieved version contains no details on behaviour on failure.

    Evidence
  • Documented

    Resuming after failures

    Scheduled tasks can be edited, paused, resumed and deleted; results stay in the history after deletion. In the CLI, --resume and --continue resume sessions.

    Evidence · Evidence
06

Evidence and open questions

16 sources · 3 open points

Open points

The reviewed sources contain no reliable information on these points. A gap says nothing about the product.

Europe: Availability still unclear

Requirements and limits: not yet confirmed

Sources and evidence

Product page, documentation, pricing page, help articles and repository reviewed. Feature descriptions come from the provider; example tasks are editorial. The trust centre and privacy policy could not be evaluated. Hands-on tests are still pending.

Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.

Read profile as Markdown

How we assess the sources

Provider

Mistral AI develops the models and offers Vibe as a product for work and code. The mistral-vibe repository contains the CLI under Apache-2.0.

Operator
Mistral AI
Organization
Provider
GitHub stars
5.303 · Reviewed
License
Apache-2.0
Provider sourcesSource 1 Source 2 Reviewed
Changes to this profile 1
  • Mistral’s assistant and agent for work and code (formerly Le Chat) on the web, in the app, in the terminal and in the editor.

These notes describe editorial changes. Source and pricing review dates appear in their respective sections.