zenml-io/zenml

What it solves

ZenML provides a unified platform to operationalize AI and ML workflows. It bridges the gap between experimental code and production-ready systems by abstracting infrastructure complexity and automating the tracking, containerization, and deployment of pipelines for both classical ML models and AI agents.

How it works

ZenML uses a client-server architecture that allows users to write Pythonic workflows (pipelines) consisting of steps. These pipelines can be run on any infrastructure backend (stacks), allowing the same code to run in development and production. It integrates with existing tools like MLflow, LangGraph, and SageMaker, and provides an observable layer to track metrics, logs, and metadata for every run.

Who it’s for

It is designed for ML and AI Engineers working in company settings who need to manage the full MLOps lifecycle—from training and evaluation to deployment and monitoring—across traditional ML and LLM-based applications.

Highlights

• Infrastructure Agnostic: Run pipelines on any backend without rewriting code.
• Detailed Tracking: Automatically tracks code, metrics, logs, and metadata for individual runs.
• Extensible Integrations: Connects with popular tools such as MLflow, Langfuse, LangGraph, and GCP Vertex.
• Unified Framework: Manages both classical ML models and agentic loops in one place.
• Developer Tooling: Includes a web dashboard, a VS Code/Cursor extension, and an MCP server for natural language querying of pipelines.