| Feathr | Open-source | Point-in-time. Supports various timestamp formats. | Supports most major sources and file formats(csv, parquet, avro, orc, delta lake) | Native transformation support with declarative framework •Row-level transformation, window aggregation transformation. •Supports offline, streaming and online transformations. | Supports feature materialization via both Python API and configuration files + CLI Redis, CosmosDB, AeroSpike, SQL | Scales. Performant. with built-in, low-level Spark optimizations | Tensor type (for deep learning/ML) + Primitive Types |
| Databricks Feature Store | Proprietary | Only time-travel (No point-in-time support). | Limited. Delta Lake tables for offline and Amazon Aurora for online. | No native transformation support. •Only general data processing with PySpark notebook. •Users must know PySpark. •No online feature transformation. •Vendor locked to Spark. | Manually managed by notebook | Doesn’t have Spark optimizations but still scales because of Spark | Primitive Types |
| Feast | Open-source | •Point-in-time and requires a fixed timestamp format. •Timestamp is always required even for non-time-series data. | Supports most major sources. Doesn’t support CSV. | Only row-level transformation with Pandas (Python library) | Supports feature materialization via CLI | Single node. In-memory. Doesn’t scale. | Primitive Types |