Feast
Open-source feature store for production machine learning
Feast is profiled here as a Feature Store tool for engineering teams. Read about features, pricing, and how it compares to related options in the tools directory.
Description
Feast is an open-source feature store that originated at Gojek in collaboration with Google Cloud in 2019 and now develops under the Linux Foundation AI and Data umbrella. It registers feature definitions as code, materializes values from offline stores into low-latency online stores, and serves them at inference time, which keeps training and serving data consistent. Recent releases extend the same registry to vector retrieval. Feast began as an open counterpart to the in-house feature platforms at large technology companies, and contributors now span dozens of organizations.
Key Capabilities:
Feature registry defined as code in Python
Offline store support for BigQuery, Snowflake, Redshift, and Spark
Online serving through Redis, DynamoDB, Bigtable, and Postgres
Point-in-time correct joins for training dataset generation
Feature server with HTTP and gRPC endpoints
Vector search support under Apache 2.0 licensing
Alternative tools
- Amazon SageMaker Feature Store
Managed feature store for machine learning on AWS
- Featureform
Open-source virtual feature store over existing infrastructure
- Tecton
Enterprise feature platform for real-time machine learning