zenml-io/zenml

ZenML 🙏: One AI Platform from Pipelines to Agents. https://zenml.io.

ZenML – An Open‑Source MLOps Platform for Pipelines, Agents & LLMs

What it is – ZenML is a Python‑based framework that lets AI/ML engineers define pipelines (workflows) as reusable, versioned steps and run them on any infrastructure (local, Docker, Kubernetes, SageMaker, etc.). It adds automatic containerisation, artifact tracking, metadata logging and a web dashboard, while integrating with the tools you already use (MLflow, Weights & Biases, LangChain/Graph, Langfuse, Sagemaker, GCP Vertex, etc.).

Why it matters – Modern AI projects combine classic ML models, large‑language‑model (LLM) services, and autonomous agents. ZenML orchestrates the whole lifecycle—from data‑prep and training to evaluation, deployment, monitoring, and even conversational querying of runs—so teams can move from experiment to production without rewriting code.

Core Features (as described in the README)

  1. Pipeline‑as‑code – Write pipelines with @step decorators; any Python logic (training, inference, agent loops) can be wrapped.
  2. Infrastructure‑agnostic stacks – Switch back‑ends (local Docker, Kubernetes, Kubeflow, SageMaker, GCP Vertex, etc.) via a simple config.
  3. Automatic containerisation & versioning – ZenML builds reproducible containers for each step and tracks code, parameters, and outputs.
  4. Run tracking & observability – Metrics, logs, and metadata are stored and visualised; integrates with MLflow, W&B, Langfuse.
  5. Extensible integrations – Built‑in connectors for popular tools and a plug‑in system for custom integrations.
  6. Web dashboard & client‑server architecture – A dedicated UI (via the zenml-dashboard repo) for monitoring pipelines; can run locally or as a remote server.
  7. LLMOps & Agent support – Guides and examples for RAG pipelines, agentic workflows, and evaluation loops.
  8. MCP (Model‑Context‑Protocol) chat interface – Query pipelines in natural language through Claude Desktop, Cursor, etc.
  9. Open‑source sister project Kitaru – Replay‑based evaluation for AI agents, tightly coupled with ZenML.

Typical Use Cases

Scenario How ZenML helps
Classic ML model training Define data‑prep, training, validation steps; run on local Docker or a Kubernetes cluster; track experiments with MLflow/W&B.
LLM fine‑tuning & RAG pipelines Orchestrate data ingestion, embedding generation, retrieval, generation, and evaluation; use the LLMOps guide for best practices.
Agentic workflows Wrap LangGraph/LangChain agents in steps; compare architectures; replay runs with Kitaru for safe testing.
Batch inference & monitoring Build end‑to‑end pipelines that pull new data, run inference, store results, and push metrics to monitoring dashboards.
Production deployment Deploy models or agents as scalable services; ZenML handles versioned artifacts and can trigger deployments via the MCP chat interface.
Cross‑team collaboration Centralised metadata, dashboards, and chat‑based queries let data scientists, engineers, and product owners stay aligned.

Ecosystem & Community

  • Integrations – MLflow, Weights & Biases, LangChain/Graph, Langfuse, SageMaker, GCP Vertex, Kubernetes, Kubeflow, and many more (see src/zenml/integrations).
  • Extensions – VS Code/Cursor extension for pipeline management directly from the editor.
  • Related projects – zenml-dashboard (UI), mcp-zenml (chat interface), kitaru (agent replay).
  • Adoption – Used by companies such as Airbus, AXA, JetBrains, Rivian, WiseTech Global, Brevo, Leroy Merlin, Koble, Playtika, NielsenIQ, Enel, etc.
  • Community – Open‑source on GitHub, Apache‑2.0 license, contribution guide, good‑first‑issues, public roadmap, Slack community, blog, and regular releases.

Getting Started (from the README)

# Install the full client‑server package
pip install "zenml[server]"   # plain `pip install zenml` gives a slimmer client

# Initialise a ZenML repo
zenml init

# Start a local server or connect to a remote one
zenml login

Then explore the examples/ folder – the quickstart example walks through pipelines, steps, artifacts, snapshots, and deployments.

License

  • Apache License 2.0 – free to use, modify, and distribute.

TL;DR

ZenML is a production‑ready, open‑source MLOps framework that lets you write, version, and run AI pipelines (including LLM and agent workflows) on any cloud or on‑prem infrastructure, with built‑in observability, integrations, and a web UI. It’s aimed at engineers who want a unified platform for both classic ML and modern generative‑AI workloads.

Related

  • Project
    mlrun/mlrun

    An open-source AI orchestration platform that automates the delivery of ML and generative AI pipelines, from data management and training to serverless deployment and monitoring.

  • Project
    clearml/clearml

    An open-source MLOps and LLMOps suite that provides experiment tracking, orchestration, and data management to streamline the deep learning development lifecycle.

  • Project
    mlflow/mlflow

    MLflow is an Apache‑2.0‑licensed open‑source platform for managing the full lifecycle of machine‑learning and generative‑AI projects. It provides experiment tracking, model registry, deployment tools, and a new GenAI stack (tracing, evaluation, prompt registry, optimization, and an AI gateway) with integrations for dozens of LLM providers and agent frameworks.

  • Project
    416rehman/DeepZero

    An automated vulnerability research pipeline engine that uses YAML-defined workflows to orchestrate data ingestion, filtering, and LLM-based security assessments.

  • Project
    Netflix/metaflow

    A human-centric framework for building and manage real-life AI and ML systems, streamlining the transition from prototyping to production deployment.