Mistral Workflows
#9 v Frameworky pro agentymistral · od 2026-04-29 · 11× · naposledy 01. 7. 2026
29
Momentum
Mistral Workflows is a production-grade orchestration platform for multi-step AI pipelines, available as part of Mistral Studio in public preview. It is built on Temporal's durable execution engine and allows developers to write workflows in Python combining LLM calls, tool use, external APIs, and human approvals. The system separates the control plane (Mistral-hosted) from the data plane (customer environment: Kubernetes, VMs, or local) and provides built-in durability, retries, OpenTelemetry observability, streaming, and human-in-the-loop primitives.
Vývoj momenta
04.04.03.07.
Vlastnosti
| Autonomy Level | Multi-agent orchestration with human-in-the-loop interruptions (wait_for_input), automatic retries, and deterministic workflow logic combined with non-deterministic activities |
| Channels | Triggered via Mistral API (POST /v1/workflows/{name}/execute), Studio UI, and Le Chat / Vibe Work (conversational workflows) |
| License | Python SDK package 'mistralai-workflows' available on PyPI under Apache 2.0 license |
| Modalities | Text/LLM calls, tool use, document processing (e.g., Document AI/OCR), structured data processing; streaming of LLM tokens |
| Platform | Part of Mistral Studio; hybrid deployment – control plane hosted by Mistral (Temporal cluster, Workflows API, Studio), workers on customer side (Kubernetes, VMs, on-prem/cloud/hybrid) via Helm chart |
| Price | Not publicly listed; custom enterprise terms via Studio/sales, no self-service tier mentioned |
| Release Date | April 27, 2026, launching as Public Preview |
| Tool/MCP Integrations | Connectors in workflows (e.g., Linear connector with on-behalf-of auth), integration with Mistral Agents, Judges, Datasets; native OpenTelemetry tracing |
