Agent Platform Feature Store

What Agent Platform Feature Store is

Agent Platform Feature Store is a managed, cloud-native feature store for managing machine-learning features and serving them online from BigQuery data sources. BigQuery tables and views serve as its offline store, so the service does not provision a separate offline store. For online predictions, it serves the latest feature values at low latencies. The service supports online and offline stores, point-in-time joins, feature monitoring, and both serving modes.

Teams can optionally register data sources and feature columns in the Feature Registry. Registration supports choosing columns, combining sources, and monitoring statistics and drift. Monitoring can run on a schedule or manually to retrieve statistics and detect anomalies. Online serving options include Bigtable, which supports large data volumes but not embeddings, and optimized online serving, which is deprecated and scheduled to sunset on February 17, 2027. Bigtable supports near-real-time continuous sync from BigQuery; optimized serving supports scheduled sync only. IAM policies can manage access to feature groups, online store instances, and feature views. Resources must share a region or multi-region with their BigQuery source, and dual-region bucket sources are unsupported. Online-store retention has no limit, though BigQuery retention limits or quotas may apply to offline data.

Who it suits

It suits teams that manage ML features in BigQuery and need online serving for predictions alongside an offline store. It is also relevant to teams that need feature monitoring, point-in-time joins, or IAM-controlled access to feature resources.

Limits

The pay-as-you-go plan is usage-based: online operations are charged per node-hour to the minute, and offline operations use BigQuery pricing. Optimized online serving supports scheduled sync rather than continuous sync and is scheduled to sunset on February 17, 2027; Bigtable serving does not support embeddings.

Getting started

1. Open the Agent Platform Feature Store documentation website.
2. Use the web interface or API to manage the service.
3. Choose Bigtable or optimized online serving for online features.
4. Use BigQuery tables or views as the offline data source.
5. Keep Feature Store resources in the same region or multi-region as the BigQuery source.

Verdict

Choose Agent Platform Feature Store if your feature data is in BigQuery and you need both online serving and offline feature management. Consider another serving approach if embeddings are central, or if you need continuous sync but plan to use optimized online serving.