Mistral Vibe folds work agents and coding PRs into one subscription

The AI assistant market is moving from chat responses to agents that can carry work across applications, documents, calendars, code editors, and pull requests. On May 28, 2026, Mistral AI launched Vibe, replacing Le Chat with a single agent product that combines long-running work tasks and coding workflows under one license.

The work side is aimed at multi-step enterprise tasks rather than one-off answers. Work Mode can search across Google Workspace, Outlook, SharePoint, Slack, GitHub, and custom connectors; analyze structured data; draft reports and other deliverables; schedule recurring tasks; and use reusable skills for repeatable processes. Mistral says the agent plans first, asks for sign-off, then shows progress and tool calls as it works through the task.

The coding side is more direct competition with cloud coding agents. Code Mode gives users a dedicated web surface to connect GitHub, create sessions, inspect changes, and move work toward a pull request. Sessions run in isolated sandboxes, can persist while a user’s local machine is off, and can run in parallel. Mistral says third-party triggers such as Slack are coming in June, which would turn the coding agent into something teams can invoke from their normal coordination tools.

Mistral is also shipping a VS Code extension and CLI updates. The extension places Vibe in a side panel where it can read across the project, edit files, execute commands, and use open files or selected line ranges as context. The CLI gains session controls, skills as slash-command-style repeatable workflows, custom modes, subagents, editable plans before execution, mid-run multiple-choice questions, session-scoped permissions, and /teleport for moving a live session between terminal and cloud.

The pricing puts Vibe in the mainstream assistant tier: free for lighter use, Pro at $14.99 per month, Team at $24.99 per user per month, and Enterprise for custom deployments and model training. The important question is not whether Mistral can add another agent surface. It is whether its flagship models, tool calling, permissions, and review flow can make the same product credible for both office work and code changes that reach production review.