Skip to main content
Back

ZenML Unifies MLOps and AI Agents Into one Open Source Platform

ZenML unifies classical ML pipelines and modern AI agents under one Python framework, with a new deployments layer that turns any pipeline into a live HTTP service.

Repo, Full Access
  • ZenML positions itself as one platform for both classical ML pipelines and AI agents, 5.6k stars on GitHub
  • New Pipeline Deployments turn any pipeline into a persistent HTTP service with warm state and rollbacks
  • Same syntax runs locally or deploys to GCP Cloud Run, AWS, and Kubernetes via a unified deployer stack
  • Integrates with MLflow, LangGraph, Langfuse, SageMaker, and Vertex rather than replacing them
  • Examples include multi-architecture agent comparisons and a document-analysis agent with embedded web UI
  • ZenML MCP Server lets Claude Desktop or Cursor query pipelines and trigger runs in natural language

The line between MLOps and agent tooling keeps blurring, and ZenML is one of the first mature open source projects to explicitly own both sides. The Apache-licensed framework, now sitting at 5.6k stars on GitHub, pitches itself as one platform from pipelines to agents, letting the same Python decorators wrap a scikit-learn training job or a LangGraph agent loop.

It targets ML and AI engineers working on traditional models, LLM workflows, or agents inside a company setting. The core idea: write workflows (pipelines) that run on any infrastructure backend (stacks), and embed any Pythonic logic within them, whether that means training a model or running an agentic loop.

Pipelines as the universal abstraction

The interesting bet ZenML is making is philosophical. Modern AI agents are, for all intents and purposes, just pipelines. They chain calls, version artifacts like context, and demand the same robustness and observability as traditional ML workflows. Most MLOps tools were built for slow, multi-node batch jobs like training, not for the fast, low-latency, single-node requests a chatbot or real-time API needs. ZenML wants to close that gap.

Its answer is a feature called