Mistral Studio adds managed agents, hosted apps and OpenTelemetry traces

Mistral said on 5 November 2026 that it has expanded Mistral Studio, its enterprise AI platform, with managed agents and apps, and with new features for workflows, enterprise context, agent observability and evaluations. The post is the company's own account. It names seven customers and partners it says already run Studio in production: ABANCA, Accenture, Alten, BNP Paribas, Capgemini, CMA CGM and Tata Consultancy Services.

The pitch is one platform for three things a business process needs. In Mistral's words, an agent investigates and acts, a workflow coordinates execution and approvals, and an app gives employees a way to use both. All three connect to company data and systems through connectors and MCP servers.

What is new

• Managed agents. Developers give an agent skills that capture a team's procedures, sandboxes to run code, subagents to delegate to, and hooks that add custom logic at set points in a run. Mistral says Studio runs them on the infrastructure behind "millions of Mistral Vibe sessions each week", and that developers set limits on the tools, data and actions an agent can reach.
• Workflows. A workflow defines which steps follow fixed rules, where an agent reasons and when a person decides. Interrupted runs resume where they stopped. The release adds event triggers from enterprise systems and a visual graph of each workflow's logic.
• Apps. Studio supplies hosting, authentication and access controls. Mistral says teams can build from pre-approved, reusable components to cut the amount of new code that needs security review. An app is published at its own link or inside Mistral Vibe.
• A Mistral MCP server. Developers can build with Mistral Vibe or the coding agent they already use, test locally, and deploy to their Studio environment with one command.

Where it runs, and what you can see

Mistral says customers can deploy both the models and the platform in their own cloud, a VPC or on-premises, use Mistral-hosted deployments, or mix them. BNP Paribas's head of AI, Su Yang, is quoted saying the bank runs Studio on its own infrastructure because the systems it builds run some of its most sensitive processes.

For oversight, the post says customers choose the models, govern access to data and tools, and decide which actions need human approval. OpenTelemetry traces show each step of a run, including model calls, tool use and approvals. Evaluations let a team test a change to a prompt, model or tool against its own datasets before release. Mistral says the traces and evaluations remain the customer's.

The numbers are Mistral's own

The post gives three internal examples. Mistral's engineers built an agent that starts a production-incident investigation when an alert fires: it follows the team's runbook, asks for approval before sensitive actions and posts findings in Slack. Mistral says this saves "hundreds of hours each month". A product-operations workflow that classifies bugs and feature requests and routes them by rule saves "about 35 hours a week". A staffing app that shortlists people for customer projects comes with no figure.

The one customer claim about speed is from CMA CGM, whose VP of AI, Julien Smadja, says each new process the shipping group automates on Studio is "production-ready in weeks rather than months". The other customer quotes describe intent or deployment choices, not results.

What the post leaves out

• Price. There is no figure for Studio, for managed-agent runtime or for self-hosted licences.
• Release status. The post says "today" and "now" but does not say whether the features are generally available or in preview, or which are available in self-hosted deployments.
• Baselines. The hours saved are not set against what the work took before, and there is no error rate for the triage workflow or the incident agent.
• Model choice. The post is written around Mistral Large 4. It says customers choose the models but does not say whether non-Mistral models can run in Studio.

Teams comparing agent platforms can put this next to the managed runtimes from the US clouds. Mistral's distinguishing claim is that the whole platform, not only the model, can sit inside your perimeter. Ask the sales team the four questions above before treating that as settled.