feast

Feast addresses the challenge of managing machine learning features consistently across training and serving environments by providing a unified feature store that handles both offline and low-latency online storage. It generates point-in-time correct feature sets to prevent data leakage, a common source of model performance issues, and abstracts feature retrieval from storage infrastructure, aiding model portability. It is primarily targeted at ML platform teams needing to operationalise feature management without tightly coupling models to specific data systems. With nearly 7,000 stars and active commits as of April 2026, Feast shows steady maturity but lacks notable recent growth signals. It is not ideal for teams looking for a lightweight or embedded feature store within a specific cloud provider's ecosystem, where alternatives like AWS SageMaker Feature Store or Tecton may offer more integrated experiences and managed services.

Choose Feast when you require a consistent, open source feature management layer that supports both batch and real-time ML workflows across diverse data infrastructures. Consider AWS SageMaker Feature Store or Tecton if you prefer a managed, cloud-native feature store tightly integrated with your cloud provider's ML stack.

Feast is typically used in teams building ML platforms that serve models requiring both historical feature datasets for training and low-latency feature retrieval for online inference. Common deployments include fraud detection pipelines, demand forecasting systems, and recommendation engines where feature consistency and correctness across environments are critical.

Latest commit: 2026-04-26. Activity data is based on daily RepoPi snapshots of the GitHub repository.