Mistral expands Studio with managed agents, workflows and apps for enterprise AI

French AI company Mistral AI said on Thursday, November 5, that it has expanded its enterprise platform, Mistral Studio, with managed agents, apps and new workflow capabilities. According to the company's blog and official release notes, developers can use the Python Agents SDK (mistralai-agents) to define a model, instructions, tools and environment once, then run the same code in-process, self-host it with agents serve, or deploy it as a Mistral-managed service with agents deploy.

Long-running agents

The release notes say managed agents can run in the background for hours or days, resume from the last finished step after a crash, and wait for a person's approval before continuing. Mistral stores the session, so customers do not have to host their own database. Agents can be equipped with skills that capture team procedures, sandboxes for running code, subagents, connectors and MCP servers, secrets, guardrails and custom hooks. Tracing in Studio is turned on by default, with a span for each model call and tool call.

Mistral described an internal example built in Studio: when a production alert fires, an agent follows the team's runbook, asks for approval before sensitive actions and posts its findings in Slack, saving the team hundreds of hours each month, the company said.

Workflows, apps and deployment choices

A new workflow builder turns a plain-language description of a process into a workflow that Mistral's Vibe assistant writes, tests and deploys, with generated code that teams can inspect and edit. Workflows can resume after an interruption, start from events in enterprise systems and display their logic as a visual graph. New mistral apps commands scaffold an application, run it locally and deploy it to Studio in one step. Studio provides hosting, authentication and access controls, the company said.

Mistral said customers can run both the models and the platform in the public cloud, in their own virtual private cloud, on-premises, or across a mix of those environments. OpenTelemetry traces show model calls, tool use and approvals. An evaluation SDK lets teams test changes to prompts, models or tools against their own datasets before release.

Who is already using it

Mistral's blog named customers and partners already using Studio in production, including ABANCA, Accenture, Alten, BNP Paribas, Capgemini, CMA CGM and Tata Consultancy Services. Su Yang, head of artificial intelligence and IT innovation at BNP Paribas, said the bank runs Studio on its own infrastructure because the systems handle some of its most sensitive processes. CMA CGM said it has used Studio for cargo release, customer assistance and transport orders, moving new processes into production in weeks rather than months. Managed agents and several of the new features are in public preview.

Sources: Mistral blog, release notes