Feathr
github.com
- Free plan
- Linux
- Web
- API
- Self-hosted
What Feathr is
Feathr is an open-source platform for defining, registering, and sharing data and feature transformations. Its Python client supports customizable user-defined functions alongside PySpark and Spark SQL, while a unified API covers offline batch, streaming, and online processing. Point-in-time-correct joins connect computed features to training data to help avoid leakage. The project also lists time-based aggregations, sliding-window joins, lookup features, and derived features. An optional UI and registry support feature search, metadata and lineage review, access management, and sharing. Registry options include Azure Purview and ANSI SQL backends, with a REST API exposed by registry deployments. Documented integrations span Azure Synapse, Databricks, Azure Machine Learning, Jupyter Notebook, Azure Blob Storage, AWS S3, Snowflake, Kafka, EventHub, Redis, and Azure Cosmos DB. Deployment options include Azure, Databricks, and Azure Synapse; a Docker sandbox and locally installable Python client are also documented. The public project is under the Apache License 2.0 and the LF AI & Data Foundation. Its FAQ notes preprocessing currently seems to accept only one UDF function.
Who it suits
Feathr may suit data and machine-learning teams that build reusable features across batch, streaming, and online environments. Its FAQ says feature stores are typically useful for modeling entities such as users, accounts, or items, and may not be needed for regular image recognition.
Good to know, and what to watch
- Supports point-in-time-correct joins.
- Offers offline, streaming, and online processing.
- Registry supports search, lineage, and access controls.
- Integrates with listed storage, streaming, and ML tools.
- Apache License 2.0.
- Feature-level access control is not supported yet.
- Preprocessing currently seems to accept only one UDF.
- Role-based access control requires a SQL service.
The verdict
Feathr provides a broad feature-engineering workflow with APIs, multiple execution modes, and an optional registry. Teams should weigh its registry access-control limits and the stated preprocessing constraint against their needs.
What the maker publishes
- AI modeling
- Feathr computes feature transformations and joins them to training data using point-in-time-correct semantics to help avoid data leakage. source
- API
- The registry deployment exposes a REST API, and both the Feathr UI and Python client interact with it. source
- Cloud and compute integrations
- Documented integrations include Azure Synapse, Databricks, Azure Machine Learning, and Jupyter Notebook. source
- Community support
- The project directs users to its Slack channel and GitHub Discussions for questions and discussion. source
- Deployment
- The project documents Azure deployment and provides a self contained Docker sandbox, plus a locally installable Python client. source
- Execution modes
- Its unified data transformation API works in offline batch, streaming, and online environments. source
- Feature engineering
- It supports time based aggregations, sliding window joins, lookup features, and derived features with point in time correctness. source
- Feature registry
- The built-in registry supports searching features, viewing data sources and lineage, and managing access controls. source
- Founded
- 2017 source
- Installation and deployment
- The project documents installing its Python client with pip and deploying Feathr on Azure, Databricks, or Azure Synapse; its sandbox is distributed as a Docker container. source
- Integrations
- Listed integrations include Azure Blob Storage, ADLS Gen2, AWS S3, Azure SQL, Snowflake, Kafka, EventHub, Redis, Azure Cosmos DB, Databricks, and Azure Synapse. source
- Intended use
- The FAQ says a feature store is typically useful when modeling entities such as users, accounts, or items, and may not be necessary for regular image recognition. source
- Interfaces
- Feathr provides Pythonic APIs and customizable user defined functions with native PySpark and Spark SQL support. source
- License and availability
- The GitHub repository is public and its license file specifies Apache License 2.0. source
- ML tools
- The project lists Azure Machine Learning, Jupyter Notebook, and Databricks Notebook as machine learning platform integrations. source
- Notable limit
- The project FAQ says preprocessing currently seems to accept only one UDF function, subject to change based on requirements. source
- Processing modes
- Its unified transformation API supports offline batch, streaming, and online environments. source
- Project status
- The README says Feathr was open sourced in 2022 and is a project under the LF AI & Data Foundation. source
- Project stewardship
- The repository says Feathr is a project under the LF AI & Data Foundation and was open sourced in 2022. source
- Purpose
- Feathr is a data and AI engineering platform for defining, registering, and sharing data and feature transformations. source
- Purpose
- Feathr is a data and AI engineering platform for defining, registering, and sharing data and feature transformations. source
- Registry and governance
- The optional UI and registry let users search features, inspect metadata and lineage, manage access controls, and share features across teams. source
- Registry backends
- The feature registry supports Azure Purview and ANSI SQL backends; role based access control requires a SQL service to store its related information. source
- Scale
- The project says Feathr can process billions of rows and petabyte-scale data using optimizations such as bloom filters and salted joins. source
Checked against the maker's pages on 4 Oct 2026.