Mistral Vibe

• Getting a single briefing on unread email, chat, and calendar items before a meeting, with attendee context and talking points
• Running deep research across the web and internal documents and turning it into an editable brief, report, or RFP response
• Cleaning, consolidating, and analyzing Excel or CSV files and downloading the result as an .xlsx workbook
• Scheduling a weekly prompt that compiles a status update from connected tools
• Delegating test generation, a module refactor, or a dependency upgrade to a remote coding agent that opens a pull request
• Starting a coding task in the terminal and teleporting it to the cloud to keep running after you close the laptop

Vibe is the clearest way to get Mistral's models doing agentic work without building anything yourself. The appeal is breadth at a low price: one Pro subscription covers a work agent with connectors and Canvas outputs and a coding agent that moves between your terminal, VS Code, and cloud sandboxes. For organizations, the main differentiator is deployment: Enterprise can run it on premises, in a private cloud, or on Mistral Cloud with data residency. The caveats are maturity and transparency: the product was renamed in May and restructured again in September, key pieces are still beta or preview, paid limits are relative rather than fixed, and the web and mobile apps cannot use another vendor's model. Teams that care most about model choice or published usage caps should test it against ChatGPT or Claude first.

About Mistral Vibe

Vibe is what Le Chat became. On May 28, 2026 Mistral renamed its assistant and repositioned it as an agent for long-running, multi-step work, keeping existing conversations, settings, and plans at the same chat.mistral.ai address. Since September 22, 2026 the separate Chat and Work tabs have been merging into one experience, with a fast/think toggle instead of a choice of surface; self-serve Free, Pro, and Team accounts migrate first, and Enterprise admins get about six months to choose their migration date. In Work mode you describe an outcome and Vibe gathers context from your prompt, files, libraries, the web, and connectors (Google Workspace, Outlook, SharePoint, Slack, Notion, GitHub, and others, more than 100 in total, plus custom MCP servers), breaks the task into steps with a live todo list, and asks for approval before sensitive actions such as sending email, posting messages, or changing data. Output lands in Canvas as documents, presentations, Excel workbooks you can download as .xlsx, or React-based mini apps and dashboards. Skills replace the old Chat Agents (Deep Research is now a Skill), memories became an editable Knowledge Base (in public preview), and prompts can be scheduled to run once or on a recurring cadence. Code mode is the same agent pointed at software: the open-source Vibe CLI (Apache-2.0) and VS Code extension work on a local checkout, Vibe Code Web runs sessions against a GitHub repository in isolated cloud sandboxes, and /teleport moves a live CLI session to the cloud and back. Remote sessions can run in parallel while your machine is off and finish as a branch or draft pull request. The apps run on Mistral's own models, with Mistral Medium 3.5 (an open-weight 128B model with a 256K context window) as the default. Plans start with Free, Pro at $14.99 a month, Team at $24.99 per user per month, and a custom Enterprise plan that adds on-premises, private-cloud, or Mistral Cloud deployment with data residency, SAML SSO, audit logs, and white labelling. The tradeoffs: agents, long-horizon tasks, and all-day coding need Pro or higher, paid limits are stated as multiples of Free and fair-use terms rather than fixed numbers, several pieces (connectors, custom MCP connectors, the Knowledge Base) are still beta or preview, and the web and mobile apps run only on Mistral's model family.