Mistral adds managed agents and apps to Studio, with customer-controlled deployment

Mistral is giving enterprise developers more of the machinery needed to turn an AI model into a working business system. On November 5, 2026, it added managed agents and apps to Mistral Studio, alongside workflow, enterprise-context, monitoring and evaluation features. Customers can run the platform in Mistral’s cloud, their own infrastructure or a combination.

Incident alerts and feedback queues

Mistral’s internal examples show the work it wants developers to package. Its engineering team built an incident-investigation agent that starts when an alert fires. The agent follows the team’s runbook, seeks approval before sensitive actions and posts findings in Slack. Engineers then take over with context already gathered. Mistral says this saves the team hundreds of hours each month.

A product-operations workflow handles a more routine queue: internal ideas and customer feedback. A model identifies bugs and feature requests, while routing rules send submissions to the appropriate team. Mistral puts the saving at about 35 hours a week. Both figures are company-reported savings from its own use, not independent measurements.

The company also describes a staffing app built by its Applied AI team. Instead of manually comparing project requirements with profiles and past experience scattered across tools and documents, staffing leads receive an agent’s shortlist. The app supplies supporting evidence explaining how each person’s experience fits the project, leaving the recommendations ready for human review.

Three building blocks, one operating platform

Studio brings three distinct jobs onto shared infrastructure: agents investigate and act, workflows coordinate execution and approvals, and apps give employees an interface. Developers connect models to company data and systems through connectors and MCP servers, which provide connections to external tools and information. Studio handles execution and lets teams inspect what happened.

• Managed agents can follow reusable procedures, run code in isolated sandboxes and delegate tasks to other agents. Developers can add custom logic and restrict the tools, data and actions available.
• Workflows combine fixed rules, model reasoning and human decisions. Interrupted runs can resume; new enterprise-system event triggers start processes, while a visual graph exposes their logic.
• Apps receive hosting, authentication and access controls. Teams can use pre-approved components, connect agents and workflows, and retrieve enterprise documents through search indexes.

The development path also reaches outside Studio’s interface. Through the new Mistral MCP server, developers can work in Mistral Vibe or their existing coding agent, test locally and deploy with one command. For workflows, Vibe can generate, test and deploy a process from a description, while leaving the resulting code available to inspect and edit.

Keeping execution and evidence under customer control

Deployment flexibility covers both the models and the platform. Studio supports customer cloud environments, virtual private clouds and on-premises infrastructure, as well as Mistral-hosted or mixed deployments. Customers choose models, govern access to data and tools, and decide which actions need human approval. Mistral says sensitive data and the processes built around it can stay within the customer’s perimeter.

For inspection, Studio provides OpenTelemetry traces: records of individual steps, including model calls, tool use and approvals. Teams can use these to locate slowdowns or failures. Evaluations test changes to prompts, models or tools against customers’ own datasets before release. Mistral says those traces and evaluations remain the customer’s, preserving the operational knowledge collected through use.

Mistral names ABANCA, BNP Paribas, Accenture, Capgemini and CMA CGM among customers and partners already using Studio in production. The next step described by CMA CGM’s VP of AI, Julien Smadja, is more building by its own teams and partners with the new version. He says that approach should let the company automate more processes while keeping expertise in-house.