feast-dev/feast

Health v2 · maintenance only

Adoption not part of the score

7230 stars · 1413 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Feast is an open-source feature store for machine learning that manages offline stores for historical training data and low-latency online stores for real-time inference, with a feature server for serving pre-computed features. It provides point-in-time correct feature retrieval to prevent data leakage and abstracts feature storage so models remain portable across data infrastructure.

Use cases

• serve features for real-time model inference with low latency
• generate point-in-time correct training datasets without data leakage
• manage feature definitions shared across training and serving pipelines
• build fraud detection or recommendation systems with real-time behavioral features
• retrieve documents via vector similarity search for RAG applications
• materialize features from data lakes like Apache Iceberg into online stores
• decouple ML models from underlying data infrastructure

When to choose

• you need consistent features across model training and online serving
• your team wants to productionize ML features on existing data infrastructure like Snowflake, Redis, or Iceberg
• you need point-in-time correct joins for training data
• you are building real-time ML applications such as fraud detection or personalization
• you want an open-source, vendor-neutral feature store with broad integrations

When to avoid

• you only need simple batch analytics without ML serving requirements
• you want a fully managed feature store and prefer a commercial SaaS like Tecton or SageMaker Feature Store
• your team cannot operate additional data infrastructure like online stores and feature servers
• you need heavy feature transformations computed at request time rather than pre-computed features

Facets

library · maturity stable

machine-learning data-science etl caching vector-database rag sdk cli machine-learning data-science big-data python cross-platform cloud self-hosted feature-store mlops online-store offline-store feature-serving point-in-time-correctness real-time-inference apache-iceberg llm-applications data-engineering docker kubernetes

Member repositories

For agents

markdown · JSON · MCP: product_card(name="feast-dev/feast")