Mistral Studio Rolls Out Enterprise AI Agents

Mistral is expanding Studio so companies can create and deploy agents, workflows, and AI applications connected to their own data and systems. The platform runs in the cloud, in a VPC, on local servers, or across a combination of environments.

The update arrives alongside Mistral Large 4, known internally as Le Chonk, a model designed for enterprise tasks such as programming, defensive cybersecurity, and complex document analysis.

A platform for moving from models to real work

Studio brings together three components that are often kept separate:

• Agents, which research and carry out multistep tasks.
• Workflows, which coordinate rules, AI actions, and human approvals.
• Applications, which provide an interface for employees to use those systems.

Companies can connect these components to their documents, internal tools, and databases through connectors and MCP servers. MCP is a standard that allows models to communicate with external tools and information sources.

Developers can also create and test solutions locally with Mistral Vibe or other coding agents, then deploy them to Studio with a single command. The platform handles hosting and execution, while the team retains access to the code and the system's operation.

Agents that investigate and act

The new managed-agent feature lets teams build systems capable of completing complex tasks. They can follow internal procedures, run code in isolated environments, and delegate parts of the work to subagents, which are agents specialized in a specific task.

They also include controls for deciding which data, tools, and actions are available to them. Developers can add their own logic at specific points in the execution and require human approval before sensitive operations are performed.

Mistral says its team uses one of these agents to investigate production incidents. When an alert is triggered, the agent follows the internal playbook, gathers information, requests authorization for sensitive actions, and posts its conclusions in Slack. According to the company, the system saves the engineering team hundreds of hours each month.

Workflows with rules and oversight

Studio workflows can combine automated steps with AI decisions and human reviews. If an execution is interrupted, it can resume from the point where it stopped.

The platform now includes triggers from enterprise systems and a visual graph for reviewing each workflow's logic. Teams can describe a process with Mistral Vibe, generate a first version, test it, and modify the code before publishing it.

One internal example classifies employee suggestions and customer feedback. The model identifies whether each entry concerns a bug or a product request, and the rules route it to the appropriate team. Mistral says this workflow saves the product operations team around 35 hours per week.

Applications ready for the enterprise

Creating an AI application is only part of the problem. To use it within an organization, you also need authentication, permissions, security, and a reliable way to maintain it.

Studio provides those capabilities and lets companies reuse components they have approved. Applications can query documents, use agents and workflows, and share context across different teams. After testing them locally, developers can publish them with a single command and make them accessible through a dedicated link or from Mistral Vibe.

Mistral's Applied AI team created an application to assign people to projects. The system compares a job's requirements with profiles and experience spread across documents and tools, then provides a list of candidates with evidence explaining why they are a good fit.

Control over data and AI

One of Studio's central questions is where the solution runs. A company can choose infrastructure managed by Mistral, its own cloud, a VPC, or local servers. This allows sensitive data and the processes built on it to remain within the company's perimeter.

The platform also includes OpenTelemetry traces, which record every step of an execution: model calls, tool usage, and approvals. Evaluations let teams compare changes to prompts, models, or tools using their own datasets before putting them into production.

In practice, this means a company can find out why a process slowed down or failed, and check whether a new version genuinely improves the results. It does not have to treat the agent's responses as a black box.

ABANCA, Accenture, Alten, BNP Paribas, Capgemini, CMA CGM, and Tata Consultancy Services are already using Studio in production, according to Mistral. The platform is being applied to use cases such as penetration testing and customer onboarding processes involving know-your-customer controls, known as KYC.

For you, the important change is not just having a more capable model. It is being able to turn that model into a tool connected to a company's processes, with permissions, oversight, and records. The next challenge will be showing that these automations maintain their accuracy and security when they move from an internal test to critical, repeated operations.