ZenML

ZenML is a unified orchestration platform for ML and AI agent workflow pipelines, functioning as a metadata layer on existing infrastructure. Its core approach is 'define once, deploy anywhere': write a Python pipeline once and deploy it on Kubernetes, Vertex AI, SageMaker, or locally without code modifications. The platform manages artifact versioning, model registry with complete lineage, and smart caching to eliminate redundant computation. ZenML includes Kitaru, its AI agent runtime that records executions in replayable traces, enabling evaluation of variants (model, tool, prompt) without re-executing agents and converting fixes into automated regression tests for CI/CD.

ZenML does not provide a French-language interface and remains dominated by English-speaking technical users. A documented REST API enables integration with third-party tools, and the freemium model (free self-hosted open-source, $999/month cloud managed, Enterprise on request) makes the platform accessible for small projects. Data and compute remain in your VPC without vendor lock-in risk thanks to the modular architecture nature. The learning curve requires good understanding of modern MLOps/LLMOps patterns and a technical team capable of managing underlying infrastructure or opting for the managed cloud version.