Mistral turns Studio into a managed platform for AI agents and apps

Mistral has expanded Studio from a model deployment layer into a broader platform for building and operating enterprise AI systems. Developers can now create managed agents, coordinate them through workflows and publish internal apps on shared infrastructure.

Managed agents can use reusable skills, code sandboxes, subagents and lifecycle hooks. Studio runs the underlying execution environment, while administrators set boundaries around tools, data and sensitive actions. OpenTelemetry traces and evaluations let teams inspect runs and test changes before release.

Workflows combine fixed rules, model reasoning and human approvals. Interrupted runs can resume, and new event triggers connect the automation to enterprise systems. Apps add hosting, authentication and access controls, so a tool built with an agent or workflow can be distributed across a company without a separate deployment stack.

The main differentiator is deployment flexibility. Mistral says models and the Studio platform can run in its cloud, in a customer VPC, on-premises or across a hybrid setup. That positions Studio as an alternative for companies that want agent infrastructure without moving sensitive processes outside their own perimeter.

Mistral says customers including BNP Paribas, Capgemini, CMA CGM and TCS already use Studio in production. The release is built around Mistral Large 4, although the practical news is the managed runtime and governance layer rather than another model benchmark.