Mistral Studio adds managed agents, workflows and apps
Multi-step agents with built-in sandboxes and subagents, a workflow builder with human approval steps, and OpenTelemetry tracing, all deployable in your own VPC or on-premises.
Mistral expanded its Studio platform on 5 November with managed agents and apps, plus new workflow, enterprise-context and observability tools, all running on its Large 4 model. Managed agents are multi-step agents that Studio runs for you, with sandboxes, subagents, skills and custom hooks built in, rather than developers wiring that infrastructure together themselves. Workflows combine fixed rules, agent reasoning and human approval steps, with event triggers and a visual graph to inspect what ran. A new observability layer exports OpenTelemetry traces covering model calls, tool use and approvals, and an evaluations tool lets teams test a prompt, model or tool change against their own datasets before shipping it.
Deployment stays flexible: Studio runs in Mistral’s cloud, a customer’s VPC, fully on-premises, or a mix of the three. Mistral says ABANCA, Accenture, BNP Paribas, Capgemini, CMA CGM and Tata Consultancy Services already run Studio agents in production, with one customer’s incident-investigation agent said to save “hundreds of hours” a month.
Why it matters: this moves Mistral from selling a model to selling the scaffolding around it, agents, workflows, evaluation and tracing, aimed at the same enterprise buyers Microsoft and Google are chasing with their own agent platforms. For teams already using Mistral’s models, it’s a reason to look at Studio before building that orchestration layer in-house.
The caveat: Mistral doesn’t say whether these features are generally available or still in preview, and the time-saved figures are self-reported by its own customers, not independently measured.