Mistral Workflows hits preview with Python SDK

This is Mistral leaning harder into the enterprise AI platform stack, not just model APIs. The interesting part is less “workflow automation” as a buzzword and more the durable execution, observability, and human-in-the-loop plumbing that teams usually end up building themselves.

• –The Temporal-backed execution model makes this more credible than a thin agent wrapper; durability and retries are the actual hard part in long-running AI systems
• –Python-first workflow authoring lowers adoption friction for teams already shipping backend services and agent pipelines
• –Studio-native tracing and audit trails matter for regulated or ops-heavy use cases like compliance, support triage, and document review
• –The split control plane / customer-run workers model is aimed at teams that need data residency and deployment control
• –Public preview plus a v3.0 SDK suggests Mistral wants developers to start treating Workflows as a real production primitive, not an experimental add-on