Mistral Studio adds managed agents with human-approval gates, workflows and MCP deployment

Mistral Studio is Mistral's enterprise platform for building and running agents. This update adds:

- Managed agents with built-in human-approval gates, which pause before sensitive actions until a person signs off.

- Visual workflow graphs for orchestration.

- Prebuilt apps with access controls.

- One-step deployment of Model Context Protocol (MCP) servers.

- The Mistral Apps CLI.

Agents, workflows and the CLI are in public preview; HTTP API connectors are in private preview.

Deployment is the differentiator. Studio runs on Mistral Cloud, in a customer's private VPC or fully on-premises. That is Mistral's sovereignty pitch to European governments, banks and defense customers who cannot send data to US hyperscalers. It pairs with Mistral Large 4, the open-weight model released October 29 at $2.10 per million output tokens and pitched for agentic coding and security work. Analysts say Large 4 and Kimi K3.1 Pro now match closed frontier models on finance workloads.

Human approval gates are now standard for agent platforms. AWS's Claude Code on AgentCore uses pull-request review, and Okta and AgentCore policies control which agent can act for which user. Mistral's version is built into the agent runtime itself.

Caveats: most features are still in preview, and Mistral is competing against OpenAI, Anthropic and AWS agent platforms with far larger ecosystems. Watch for general availability dates and named European reference customers.