Agent Platform Feature Store
Agent Platform Feature Store is a managed cloud service for managing machine-learning features and serving them online from BigQuery data.
What Agent Platform Feature Store is
Agent Platform Feature Store is a managed cloud service for managing machine-learning features and serving them online from BigQuery data. BigQuery tables and views form the offline store, so Feature Store does not create a separate offline store. It serves the latest feature values from BigQuery for online predictions at low latencies. Teams may optionally register data sources and columns in the Feature Registry to choose columns, combine sources, and monitor statistics and drift. Monitoring can run manually or on a schedule to retrieve statistics and flag anomalies such as feature drift. 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. Google recommends Vector Search for embeddings. IAM policies can govern access to feature groups, online store instances, and feature views. Resources must be in the same region or multi-region as the BigQuery source; dual-region bucket sources are unsupported. The online store has no data retention limit, while BigQuery limits or quotas may affect offline data.
Good
- BigQuery tables and views provide the offline store
- Serves latest feature values for online predictions
- Feature monitoring can be scheduled or run manually
- IAM policies control access to feature resources
Watch out
- Optimized online serving is deprecated
- Bigtable serving does not support embeddings
- Resources must match the BigQuery source region
- Dual-region bucket sources are unsupported
Who it suits
This service is for teams managing machine-learning features in BigQuery and serving feature values for online predictions. It may suit teams that need feature monitoring or IAM controls over feature resources.
Verdict
Agent Platform Feature Store links BigQuery-based offline data with online feature serving and optional monitoring. Review the serving-mode limits and regional constraint, particularly if you need embeddings or use dual-region bucket sources.
Plans and prices
From Agent Platform Feature Store's own pricing page, checked 4 Oct 2026.
| Plan | Price | What it gives you |
|---|---|---|
| New Agent Platform Feature Store (pay-as-you-go) | — | Data processing node: $0.08 / 1 hour · Optimized online serving node: $0.30 / 1 hour · Bigtable online serving node: $0.94 / 1 hour · Bigtable online serving storage: $0.000342466 / 1 gibibyte hour · Offline operations: BigQuery pricing · Usage-based; online operations are charged per node-hour, to the minute. Offline operations use BigQuery pricing. |
The facts
Each one read from Agent Platform Feature Store's own pages, with the date it was checked.
What it does
- Data retention
- The online store has no data retention limit, while BigQuery retention limits or quotas may apply to the offline data.checked 4 Oct 2026
- Data sync limit
- Optimized online serving supports scheduled data sync but not continuous data sync; Bigtable online serving supports near-real-time continuous sync from BigQuery.checked 4 Oct 2026
- Embeddings
- Google recommends its purpose-built Vector Search to efficiently store and serve embeddings.checked 4 Oct 2026
- Feature registry
- Registering data sources and feature columns in the Feature Registry is optional; registration supports selecting columns, combining sources, and monitoring statistics and drift.checked 4 Oct 2026
- IAM access control
- IAM policies can control access to feature groups, online store instances, and feature views, with bindings for users, groups, domains, and service accounts.checked 4 Oct 2026
- Location constraint
- Feature Store resources must be in the same region or multi-region as the BigQuery source, and dual-region bucket sources are unsupported.checked 4 Oct 2026
- Monitoring
- Feature monitoring can run on a schedule or manually to retrieve statistics and detect anomalies such as feature drift.checked 4 Oct 2026
- Offline data
- BigQuery tables and views form the offline store, so Feature Store does not provision a separate offline store.checked 4 Oct 2026
- Online predictions
- It serves the latest feature values from BigQuery for online predictions at low latencies.checked 4 Oct 2026
- Purpose
- It is a managed, cloud-native feature store for managing ML features and serving them online from BigQuery data sources.checked 4 Oct 2026
- Serving options
- Bigtable online serving supports large data volumes but not embeddings; optimized online serving is deprecated and scheduled to sunset on February 17, 2027.checked 4 Oct 2026
- Support and learning
- The documentation links to Google Cloud Support, community forums, release notes, and tutorials with Colab and GitHub examples.checked 4 Oct 2026
