Mistral expands Studio with managed agents and enterprise apps

Mistral AI expanded Studio on November 5, adding managed agents and apps alongside new workflow, enterprise-context, observability and evaluation capabilities. Most of the additions are previews, not generally available products.

The same Python agent code can run in-process during development, through a self-hosted worker or as a Mistral-managed deployment. According to Mistral’s managed-agent documentation, deployed agents support background execution, stored sessions and recovery from the last completed model or tool call after a crash. A tool call can also pause until a person or application approves it.

Developers can add skills, code-execution sandboxes, subagents and hooks while restricting the tools, data and actions an agent can access, Mistral said. The agent harness also supports connectors, Model Context Protocol servers, secrets and guardrails.

Studio workflows can combine fixed-rule steps with agent reasoning and human decisions. Mistral says they can resume interrupted runs, respond to events from enterprise systems, and expose generated code and a visual workflow graph for inspection.

Studio apps add hosting, authentication and access controls. They can use reusable, pre-approved components and move from local development to a dedicated link or Mistral Vibe. Deployment options vary by component: Mistral’s Studio overview says agents and workflows can run on customer-operated infrastructure, but apps run only on Mistral Cloud. Its self-hosting guide distinguishes self-hosting individual workloads from running the full Studio platform on premises.

The expansion also adds reusable context components—including connectors, MCP servers, document libraries, search indexes, prompts and skills—that can be shared across agents, workflows and apps. The November 5 release notes list a managed vector search index in public preview and HTTP API connectors in private preview.

Tracing and evaluation tools are part of the release as well. According to Mistral, managed-agent traces record agent, model and tool spans by default, with common secrets and personal data redacted on the worker. Studio can turn traces into evaluation datasets and run offline or live-traffic evaluations to compare versions and detect regressions. Mistral’s documentation limits observability and evaluations to Enterprise-tier organizations.

Mistral said ABANCA, Accenture, Alten, BNP Paribas, Capgemini, CMA CGM and Tata Consultancy Services already use Studio in production. The supplied evidence does not independently verify that adoption claim.

Managed Agents, the workflow builder, Apps CLI, managed search index, agent tracing, and offline and online evaluations are all labeled as public previews in the release notes. Mistral did not disclose pricing, service-level commitments or broader availability dates.