ToolsNews
Mistral Studio: managed agents, workflows and apps
Mistral expanded its Studio enterprise platform with managed agents, stronger workflows and apps. Pair it with Le Chonk (Mistral Large 4) and deploy in the cloud, a VPC or on-prem while keeping control of data and infrastructure. ABANCA, BNP Paribas and others already run it in production.
· 2 min read
What changed?
Mistral expanded Studio today with managed agents and apps, plus new workflow features, enterprise context, agent observability and evaluations. The pitch is faster paths from agent, process or internal app idea to production without building the whole runtime yourself.
On the model side, Le Chonk (Mistral Large 4) is the headline pairing for coding, defensive cybersecurity and complex documents. Studio adds connectors and MCP servers so agents can reach enterprise data and systems.
Agents, workflows and apps
Managed agents get skills, sandboxes, subagents and hooks. The runtime rides the same infrastructure that serves millions of Mistral Vibe sessions each week. Developers set the boundaries for tools, data and actions. Mistral’s own incident investigation agent follows a runbook, asks for approval before sensitive steps, then posts findings to Slack.
Workflows spell out where rules apply, where an agent reasons and where a human decides. Interrupted runs can resume; event triggers come from enterprise systems; and there is a visual graph. Their product ops team built a workflow that routes bugs and feature requests, saving about 35 hours a week.
Apps get hosting, authentication and access controls. You can test locally, then deploy with one command to a dedicated link or inside Mistral Vibe. A new Mistral MCP server lets teams build from Vibe or another coding agent and push into their Studio environment with that same command.
Sovereignty and control
Models and the platform can run in Mistral’s cloud, your VPC, on-premises or a mix. OpenTelemetry traces cover model calls, tool use and approvals. Evaluations run on your own datasets before you ship prompt, model or tool changes. Those traces and evals stay yours.
Customers and partners already in production include ABANCA, Accenture, Alten, BNP Paribas, Capgemini, CMA CGM and Tata Consultancy Services. For us the draw is the packaged observability and eval layer next to the agent stack, plus choosing where it runs. We covered Large 4 separately; this story is the platform, not the parameter count.
Produced with the help of AI, reviewed by an editor. How we make our articles
Reviewed and edited by the Neural Daily desk. About
