Feast
Feast is an open-source feature store designed to serve structured data to AI and LLM applications during training and inference.
Profile
Feast is an open-source platform that helps machine learning teams define, store, and serve features consistently for both training and real-time inference.
Feast is an open-source feature store designed to serve structured data to AI and LLM applications during training and inference. Founded by Willem Pienaar, Danny Tarlow, and others, the project originated at Google and was open-sourced in 2019. It is now hosted under the Linux Foundation AI & Data Foundation.
Feast provides a centralized platform for defining, storing, and retrieving features, aiming to ensure consistency between training and production environments. The project reports over 12 million downloads, 293 contributors, and 5,500 Slack members. Adopters include Robinhood, NVIDIA, Discord, Cloudflare, Walmart, Shopify, Salesforce, Twitter (X), IBM, Capital One, Red Hat, Expedia, HelloFresh, Adyen, and SeatGeek.
Feast integrates with a wide range of offline stores (Snowflake, BigQuery, Redshift, Spark, PostgreSQL, Trino, DuckDB, Azure Synapse, Dask, ClickHouse, MSSQL) and online stores (Redis, DynamoDB, Bigtable, Cassandra, MySQL, PostgreSQL, Snowflake, SQLite, Dragonfly, SingleStore, IKV, Hazelcast, ScyllaDB, Milvus, Qdrant, Couchbase). The project has recently added native integrations for MLflow and MongoDB, and published performance benchmarks for sub-2ms online serving on Kubernetes. Feast does not have a corporate entity or disclosed revenue; it is a community-governed open-source project.
No funding rounds or venture capital investment have been publicly reported. The project's governance and roadmap are managed by a technical steering committee with representatives from key contributing organizations. Feast competes with other open-source feature stores like Tecton (which has a commercial product) and managed services from cloud providers.
Who buys this
- Machine learning engineering teams at large enterprises
- Data science and MLOps teams building production AI systems
- Organizations with real-time ML use cases like fraud detection, recommendations, and risk scoring
- Companies using LLMs and RAG pipelines that need structured feature serving
Publicly disclosed clients
- Robinhood
- NVIDIA
- Discord
- Cloudflare
- Walmart
- Shopify
- Salesforce
- Twitter (X)
- IBM
- Capital One
- Red Hat
- Expedia
- HelloFresh
- Adyen
- SeatGeek
Strengths and what to watch
Strengths
- Strong community adoption with 12M+ downloads and 293 contributors, indicating broad validation and a healthy open-source ecosystem.
- Deep integration with major cloud and database providers, including Snowflake, BigQuery, Redis, and MongoDB, reducing vendor lock-in for users.
- Proven performance with published benchmarks showing sub-2ms online serving latency on Kubernetes, addressing a critical requirement for production ML.
Watch for
- Feast is a community-governed open-source project without a corporate parent or disclosed revenue, making its long-term sustainability dependent on volunteer contributions and corporate sponsorships.
- The project faces competition from commercial feature stores like Tecton (backed by Andreessen Horowitz) and managed services from cloud providers (AWS SageMaker Feature Store, GCP Vertex AI Feature Store), which may offer more polished user experiences.
- Governance and roadmap decisions are made by a technical steering committee, which can lead to slower feature development or community disagreements compared to vendor-driven projects.
Key Information
- Industry
- MLOps
- Founded
- 2019
Frequently Asked Questions
What is Feast and what does it do for machine learning teams?
Feast is an open-source feature store that helps machine learning teams define, store, and serve features consistently for both training and real-time inference. It originated at Google and was open-sourced in 2019.
How does Feast ensure consistency between training and production environments?
Feast provides a centralized platform for defining, storing, and retrieving features, ensuring the same feature definitions and values are used during model training and online inference. This reduces discrepancies and improves model reliability in production.
What databases and stores does Feast integrate with?
Feast integrates with offline stores like Snowflake, BigQuery, Redshift, and Spark, and online stores like Redis, DynamoDB, Bigtable, and Cassandra. It also supports PostgreSQL, MySQL, MongoDB, and many others for flexible deployment.
What performance benchmarks has Feast published for online serving?
Feast has published performance benchmarks showing sub-2ms online serving latency on Kubernetes. This demonstrates its ability to meet critical latency requirements for production machine learning applications like fraud detection and recommendations.
Who are some notable companies using Feast in production?
Notable adopters of Feast include Robinhood, NVIDIA, Discord, Cloudflare, Walmart, Shopify, Salesforce, Twitter (X), IBM, Capital One, Red Hat, Expedia, HelloFresh, Adyen, and SeatGeek. These companies use Feast for real-time ML use cases.
How is Feast governed and what is its long-term sustainability?
Feast is a community-governed open-source project hosted under the Linux Foundation AI & Data Foundation. It has no corporate parent or disclosed revenue, relying on volunteer contributions and corporate sponsorships, which may affect long-term sustainability.
Sources
- feast.dev — Project description, list of adopters, integrations, download and contributor statistics, use cases, and code examples.
- feast.dev — Performance benchmarks for sub-2ms online serving on Kubernetes.
- feast.dev — Announcement of native MLflow integration for automatic feature lineage.
- feast.dev — Announcement of native MongoDB support as both online and offline store, plus vector search.