Mistral AI Launches Workflows — an Orchestration Layer for Enterprise AI

Mistral AI announced the public preview of Workflows — an orchestration layer for enterprise AI that ensures reliable operation of AI processes in production. The platform addresses common issues: silent pipeline failures, long processes hanging due to timeouts, lack of pause and resume mechanisms, and inability to verify after deployment. Workflows is embedded in Studio, so orchestration and components (inference, agents, connectors, observability) work as a single unit. Developers write workflows in Python, publish them to Le Chat, and each step is tracked and audited in Studio. Use cases include: automating cargo release in maritime shipping — document checks, pause for human approval (wait_for_input()), and full execution history; KYC document verification — fast processing with OpenTelemetry traces; support request routing — intent and urgency categorization visible in Studio. Features include: reliable execution with state tracking, observability with a full timeline, human-in-the-loop with a single line of code, integration with Studio, role-based access control (RBAC), deployment in cloud/on-prem/hybrid. It is based on Temporal (used by Netflix, Stripe, Salesforce), extended for AI workloads. The control plane runs on Mistral, while data and processing stay in the client's environment (Kubernetes via Helm chart). Python SDK v3.0 is now available.