Skip to main content
ZenML’s integration system allows you to extend the core framework with support for popular cloud platforms, ML tools, and orchestration systems. Each integration provides specialized stack components that seamlessly work together.

​
How Integrations Work

Integrations are optional extensions that add new flavors of stack components to ZenML. When you install an integration, you get:
  • Stack Components: Orchestrators, artifact stores, experiment trackers, etc.
  • Service Connectors: Authentication and resource management for cloud platforms
  • Materializers: Custom serialization for integration-specific data types
  • Step Operators: Execute individual steps on specialized infrastructure

​
Installation

Install integrations using the zenml[integration-name] syntax: 
# Install a single integration
pip install "zenml[aws]"

# Install multiple integrations
pip install "zenml[aws,gcp,mlflow]"

# Install all integrations
pip install "zenml[all]"
You can also use the ZenML CLI: 
zenml integration install aws
zenml integration list

​
Available Integrations

AWS

Amazon Web Services integration with SageMaker, S3, and ECR

Google Cloud

GCP integration with Vertex AI, GCS, and GCR

Azure

Microsoft Azure with AzureML and Blob Storage

Kubernetes

Native Kubernetes orchestration and deployment

Kubeflow

Kubeflow Pipelines orchestration

MLflow

MLflow experiment tracking and model registry

W&B

Weights & Biases experiment tracking

SageMaker

AWS SageMaker orchestration and step operators

Vertex AI

GCP Vertex AI pipelines and training

​
Integration Categories

​
Cloud Platform Integrations

Provide comprehensive support for major cloud providers:
  • AWS: S3 artifact storage, ECR container registry, SageMaker orchestration
  • GCP: GCS artifact storage, GCR container registry, Vertex AI orchestration
  • Azure: Blob Storage, AzureML orchestration and training

​
Orchestrator Integrations

Execute pipelines on specialized orchestration platforms:
  • Kubernetes: Native k8s orchestration with full Pod customization
  • Kubeflow: Kubeflow Pipelines (KFP) v2 orchestration
  • SageMaker: AWS SageMaker Pipelines with distributed training
  • Vertex AI: GCP Vertex AI Pipelines with AutoML support

​
Experiment Tracking Integrations

Track experiments and compare runs:
  • MLflow: Open-source experiment tracking with model registry
  • Weights & Biases: Cloud-based tracking with collaboration features

​
Stack Components by Integration

Each integration provides different types of stack components:
IntegrationOrchestratorArtifact StoreContainer RegistryExperiment TrackerStep Operator
AWS✓-✓-✓
GCP✓✓-✓✓
Azure✓✓--✓
Kubernetes✓---✓
Kubeflow✓----
MLflow---✓-
W&B---✓-

​
Service Connectors

Most cloud integrations include service connectors for authentication: 
from zenml.client import Client

# Create an AWS service connector
Client().create_service_connector(
    name="aws-connector",
    type="aws",
    auth_method="secret-key",
    configuration={
        "aws_access_key_id": "AKIAIOSFODNN7EXAMPLE",
        "aws_secret_access_key": "wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY",
        "region": "us-east-1",
    },
)
Service connectors can be shared across multiple stack components, making it easy to manage credentials centrally.

​
Using Integration Components

After installing an integration, register its components with your stack: 
# Register a SageMaker orchestrator
zenml orchestrator register sagemaker-orch \
    --flavor=sagemaker \
    --execution_role=arn:aws:iam::123456789:role/SageMakerRole \
    --region=us-east-1

# Register an MLflow experiment tracker
zenml experiment-tracker register mlflow-tracker \
    --flavor=mlflow \
    --tracking_uri=https://mlflow.example.com

# Create a stack with these components
zenml stack register prod-stack \
    -o sagemaker-orch \
    -e mlflow-tracker \
    -a s3-artifact-store

​
Integration Settings

Many integrations support step-level settings for fine-grained control: 
from zenml import step
from zenml.integrations.aws.flavors.sagemaker_orchestrator_flavor import (
    SagemakerOrchestratorSettings,
)

@step(
    settings={
        "orchestrator": SagemakerOrchestratorSettings(
            instance_type="ml.p3.2xlarge",
            volume_size_in_gb=100,
        )
    }
)
def train_model() -> None:
    # This step runs on a GPU instance
    ...

​
Best Practices

Integration dependencies can be large. Only install integrations you actively use to keep your environment lean:
# Good: Install specific integrations
pip install "zenml[aws,mlflow]"

# Avoid: Installing all integrations unnecessarily
pip install "zenml[all]"  # Only for testing
Service connectors provide centralized credential management and automatic authentication:
  • Store credentials once, use across multiple components
  • Support multiple authentication methods (keys, roles, service accounts)
  • Automatic credential refresh for cloud platforms
  • Audit trails for credential usage
Use different stacks for development and production:
# Development: local execution, fast iteration
zenml stack register dev -o local -a local

# Production: cloud execution, scalability
zenml stack register prod -o sagemaker -a s3 -e mlflow
Pin integration versions in production environments:
# requirements.txt
zenml[aws]==0.55.0
zenml[mlflow]==0.55.0

​
Next Steps

Cloud Integrations

Learn about AWS, GCP, and Azure integrations

Orchestrators

Explore orchestration options

Experiment Tracking

Track and compare experiments

Service Connectors

Set up authentication

Build docs developers (and LLMs) love

Get started for freeTalk to us