Mistral kills Le Chat, bets it all on one agent for inbox and pull requests

Mistral AI launched Le Chat in February 2024 as a straightforward chatbot aimed at the consumer market. Barely a year later, the company pulled the ripcord on that identity. Le Chat is now Vibe, and the name change is the least interesting part of what Mistral just announced.

The architecture is the story. Vibe is one agent meant to handle two very different jobs: enterprise workflow orchestration and software development. It is a bet most competitors have not made. OpenAI keeps ChatGPT and its Codex-powered API as separate categories. Anthropic's Claude straddles the line but maintains distinct product surfaces. Google's Gemini is the closest parallel, and even there, coding and enterprise productivity remain largely siloed. Not everyone is betting on a single do-everything agent; Sakana AI's Fugu system is built around orchestrated swarms of models instead.

A unified agent, not a collection of tools

Vibe's Work Mode takes on multi-step tasks across Google Workspace, Outlook, SharePoint, Slack, GitHub, and custom connectors. It can search enterprise knowledge bases, analyze structured data from databases or spreadsheets, draft documents and reports, and run scheduled workflows on daily, weekly, or monthly cadences. The agent makes its reasoning visible as it executes. Every tool call and thought chain is expandable, and it asks for user sign-off before beginning a plan.

Code Mode runs remote coding agents from a dedicated web surface, with isolated sandboxes and session persistence even when the local machine shuts down. It connects to GitHub, manages projects, and ships work through pull requests. The new Mistral Vibe extension for VS Code brings the same agent into the IDE as a side panel that reads, edits, and executes commands across the entire project. Mistral had already previewed this cloud-based, asynchronous session model with its move to cloud coding agents.

Mistral frames the move as deliberate. It expects knowledge work and software development to blend into one capability: reasoning over context, calling tools, producing artifacts, and iterating until the job is done. The company's flagship models have been optimized for reasoning, agentic tasks, tool calling, and code generation. That same technical stack now powers both modes.

The pricing and positioning

Vibe launches with four tiers: a free plan for quick answers and simple tasks; Pro at $14.99 per month for complex tasks and all-day coding; Team at $24.99 per user per month with shared workspace and admin controls; and Enterprise for custom deployments, model training, and dedicated support.

The $14.99 Pro price undercuts ChatGPT Plus and Claude Pro, both at $20, and shaves a dollar off Copilot Pro's $15. That aggressive pricing suggests Mistral is willing to trade margin for adoption, especially among developers who might otherwise reach for GitHub Copilot or Cursor.

A single license covers both Work and Code modes. A user paying for Pro can manage their inbox, run research, and switch to coding sessions without a second subscription. For individual developers and small teams, that bundling could be the decisive advantage. It also responds to a market where coding agents have run ahead of enterprise adoption.

Open-source roots, enterprise reality

Mistral has always straddled open source and commercial products more carefully than most peers. It releases powerful base models under permissive licenses while keeping its flagship products proprietary. Vibe continues that strategy: the agents run on Mistral's optimized flagship models, but enterprises can also build through the API in Mistral Studio, with a free Experiment plan for testing and prototyping.

The reusable skills system, which extends Vibe through open standards, is another nod to the developer audience that made Mistral's early reputation. Users can define custom skills to automate repeatable workflows and trigger them as slash commands in Code Mode or scheduled tasks in Work Mode. That is a practical answer to tool sprawl: package a recurring workflow once and call it from either mode.

Whether the market actually wants one agent for both jobs is an open question. The coding half has a known hurdle: models that clean up on public benchmarks can stumble badly on private enterprise code, as SWE-bench comparisons keep showing. Mistral is willing to unify its entire product line to find out.

Vibe is live starting today at chat.mistral.ai, with mobile apps on iOS and Android. Existing Le Chat users will have their history, plans, and settings carried over. The company says it will soon sunset the original Chat mode, though conversation history will be preserved in Work Mode. It is also adding new connector controls for enterprise governance.