Overview | Kubeflow

Kubeflow Hub (formerly known as Model Registry) is the central component in Kubeflow for managing the full lifecycle of ML models. It brings together two complementary capabilities under one project:
Together, Model Registry and Model Catalog cover the full model-management workflow: discover a model in the catalog, register it in the registry, version and govern it through the ML lifecycle, and deploy it to production.
The Model Registry is a central index for ML model developers to manage models, versions, and ML artifacts metadata. It fills a gap between model experimentation and production activities, providing a central interface for all stakeholders in the ML lifecycle to collaborate on ML models.
The Model Registry supports every stage of the ML lifecycle:
DevOps, Data Scientists, and developers need to collaborate with other users in the ML workflow to get models into production. Data scientists need an efficient way to share model versions, artifacts and metadata with other users that need access to those models as part of the MLOps workflow.
The Model Catalog is a read-only discovery service for ML models across multiple catalog sources. It acts as a federated metadata aggregation layer, allowing users to search and discover models from various external catalogs through a unified interface.
The catalog service operates as a metadata aggregation layer that:
The catalog does not store model weights or artifacts — it provides metadata and references to where models can be found in their source repositories.
For the API specification, see the Model Catalog REST API reference.
This section describes Model Registry use-cases in the context of a MLOps Platform with Model Training, Experimentation, and Deployment.
A company, ACME Inc., is developing a machine-learning model for predicting customer churn. They require a centralized model registry for their MLOps platform (based on Kubeflow) for managing their ML model development lifecycle, including training, experimentation, and deployment. They want to ensure model governance, reproducibility, and efficient collaboration across data scientists and engineers.
The Data Scientist uses Kubeflow Notebooks to perform exploratory research and trains several types of models, with different hyperparameters and metrics. The Kubeflow Model Registry is used to track those models, in order to make comparisons and identify the best-performing model. Once the champion model is selected, the Data Scientist shares the model with the team by maintaining the appropriate status flag on the registry. The Data Scientist also tracks the lineage of training data sources and notebook code.
The Data Scientist after identifying a base model, uses Kubeflow Pipelines, Katib, and other components to experiment model training with alternative weights, hyperparameters, and other variations to improve the model’s performance metrics; Kubeflow Model Registry can be used to track data related to experiments and runs for comparison, reproducibility and collaboration.
The MLOps Engineer uses Kubeflow Model Registry to locate the most recent version for a given model, verify it is approved for deployment, understand model format, architecture, hyperparameters, and performance metrics to configure the serving environment; once deployed, Model Registry is used to continue monitoring and track deployed models for performance and mitigate drift.
The Business Analyst uses Kubeflow Model Registry to audit deployed models, monitor model performance by integrating with observability tools to track key metrics and identify when model is drifting or needs re-training; capabilities of model lineage enable identifying all related artifacts such as training which was used or the original training data.
By implementing a model registry, ACME Inc. can significantly enhance their MLOps platform’s functionality, enabling efficient model training, experimentation, and deployment. The Model Registry empowers Data Scientists, MLOps engineers, and Business analysts to collaborate effectively and make informed decisions based on reliable data and insights.
Was this page helpful?
Thank you for your feedback!
We're sorry this page wasn't helpful. If you have a moment, please share your feedback so we can improve.
© 2026 The Kubeflow Authors.Documentation distributed under CC BY 4.0
The Linux Foundation® (TLF) has registered trademarks and uses trademarks. For a list of TLF trademarks, see Trademark Usage.
Website Privacy Policy