Mistral Studio adds managed agents and apps to put Le Chonk to work

Felix Laurent 5 Nov 2026 4 min read

AI agents

Mistral expanded Mistral Studio, its enterprise AI platform, on November 5, 2026. The update adds managed agents and apps, plus new tools for workflows, enterprise data, observability and evaluations. Mistral pitches it as the place to put Mistral Large 4, the model it nicknames Le Chonk, to work.

The model came first. Mistral launched Large 4 in preview on October 6 and released its open weights later that month. Studio is how Mistral wants companies to turn that model into running systems, on Mistral's cloud or on their own servers.

What Mistral added to Studio

Mistral splits a business process into three parts. An agent investigates and acts, a workflow coordinates steps and approvals, and an app gives employees a way to use both. Studio now hosts all three on shared infrastructure, connected to company data through connectors and MCP servers.

Most of the new features ship in public preview, according to Mistral's release notes from the same day. The table lists the main ones.

Feature	What it does	Status
Managed agents	Hosts multi-step agents built with the Agents SDK	Public preview
Workflow builder	Mixes fixed rules, agent steps and human approvals	Public preview
Apps CLI	Deploys internal apps with hosting, sign-in and access control	Public preview
Tracing for managed agents	Records each step of an agent run	Public preview
Offline and online evaluations	Tests prompts, models and tools on company datasets	Public preview
HTTP API connectors	Connects agents to systems that do not use MCP	Private preview
Studio apps in Vibe Code Web	Opens Studio apps inside Mistral's coding tool	Generally available

Only the Vibe Code integration is generally available. Everything else is still a preview, which matters for companies planning production use.

How managed agents work in Mistral Studio

Developers build agents in Python with Mistral's Agents SDK. They can give an agent skills that capture a team's procedures, sandboxes to run code, and subagents to hand off parts of a task. Hooks add custom logic at set points in a run.

The same agent can run three ways: inside the developer's own process, on the company's servers, or on Mistral's cloud with one deploy command. Mistral says the runtime is the one behind millions of weekly sessions of Mistral Vibe, its coding agent. Developers set limits on which tools, data and actions each agent can reach.

Mistral gave one example from its own engineering team. Its incident agent starts investigating when an alert fires, follows the team's runbook, asks for approval before sensitive actions and posts findings in Slack. Mistral says it saves the team hundreds of hours a month.

Workflows handle the processes around such agents. They decide which steps follow fixed rules, where an agent reasons and where a person signs off. A run that stops can resume where it left off. Mistral's product operations team uses one to sort internal ideas and customer feedback, saving about 35 hours a week, by the company's count.

Apps are the third piece. Studio provides hosting, sign-in and access control, so an internal tool built with a coding agent can meet company security rules faster. Mistral's own staffing app recommends people for customer projects and explains how their experience fits.

Why Mistral sells control

The common thread is control. Companies can run both the models and Studio in their own cloud, a private VPC or on their own servers. They can also mix these with Mistral-hosted deployments. Studio's documentation notes that apps run on Mistral Cloud for now, while agents and workflows can also run on a company's own infrastructure.

Teams choose the models, set which actions need human approval and keep their own records. OpenTelemetry traces show every model call, tool use and approval in a run. Evaluations let teams test changes against their own datasets before release. Mistral says those traces and evaluations stay with the customer, along with the knowledge built up through use.

Several large customers already run Studio in production, including ABANCA, BNP Paribas and shipping group CMA CGM. BNP Paribas runs it on its own infrastructure. Its head of AI, Su Yang, said the systems they build "run some of our most sensitive processes."

Consulting firms Accenture, Capgemini, Alten and Tata Consultancy Services also build on the platform. That partner list shows who Mistral is targeting: large European companies with strict data rules.

Is Mistral Studio ready for enterprise agents

For pilots, yes. Studio now takes a model all the way to a working system, with agents, workflows, apps, tracing and evaluations. All of it can run on a company's own hardware. Banks and a global shipping group already use it in production, and their processes are among the most sensitive a company has.

For broad rollout, not yet. Managed agents, apps, workflows and evaluations are all public previews, and HTTP connectors are a private preview. Observability and evaluations are limited to Enterprise-tier customers, according to Studio's documentation. Companies that need on-premises control and an open model will find a strong offer here. Until Mistral marks these features as generally available, though, they are signing up for preview software.