ZenML

ZenML, developed by ZenML GmbH, is an Apache 2.0 open-source platform for writing workflows in Python and running them on the orchestrator and cloud infrastructure a team already uses. It targets ML and AI engineers working on traditional ML, LLM workflows, or agents in a company setting. The latest GitHub release listed is 0.97.0.

What It Is

ZenML lets you define workflows (pipelines) that run on any infrastructure backend (stacks). Any Pythonic logic can be embedded in a pipeline step, such as training a model or running an agentic loop. Existing code is wrapped in a @step, so tools like scikit-learn, PyTorch, LangGraph, LlamaIndex, or raw API calls can keep being used.

ZenML then containerizes and tracks the code, records runs with metrics, logs, and metadata, and abstracts infrastructure complexity.

Setup Path

Install with pip install "zenml[server]" (a slimmer client is available via pip install zenml), run zenml init, then zenml login to start a local server or connect to a remote one. It uses a client-server architecture with an integrated web dashboard. The repository includes examples such as a quickstart, agent architecture comparison, model and agent deployment, and end-to-end batch inference.

Integrations and Ecosystem

The README lists integrations with MLflow, Weights & Biases, LangGraph, Langfuse, SageMaker, GCP Vertex, and Kubernetes-based orchestrators including Kubeflow. A ZenML MCP Server lets MCP-compatible clients such as Claude Desktop or Cursor query pipelines, analyze runs, and trigger deployments in natural language. A VS Code/Cursor extension manages pipelines from the editor.

Kitaru for Agents

The README presents Kitaru as a separate open-source sister project for agents. It records agent runs, replays them against real code with recorded tool calls, and lets users fork a run with one change, such as a different model or prompt. Per the README, ZenML is for ML pipelines and Kitaru is for agents.