AlloyDB PostgreSQL for agents: Preview isolated query compute

Google Cloud documents PostgreSQL for agents in AlloyDB as a Preview feature (page last updated 2026-09-24 UTC): a decoupled path so AI agents can query live transactional data with full isolation from core business systems—and without affecting the primary production database—while agent compute can start in seconds and scale back to zero when idle (docs).

Preview only

This product is subject to Google’s Pre-GA Offerings Terms and Additional Terms for Generative AI Preview products—available “as is” with limited support. It is not generally available; Google has not published a GA date, SKUs, regions, or pricing on this public page. Access is via Google’s access request / join the Preview flow; deeper technical docs sit behind that gate (docs).

What Google states

• Agents query live transactional data with full isolation so they don’t affect core business systems or the primary production database
• Decoupled architecture framed for many collaborating agents that start in seconds
• Compute can automatically scale back to zero when idle

Google also markets “thousands” of collaborating agents and “lakehouse analytics” on live operational data—vendor framing only on a thin public page. Stick to isolation and scale-to-zero as written; unpublished compute internals stay out of this brief.

Who should care

Teams running AlloyDB who want agent analytics on fresh OLTP state without hammering the primary should start at the AlloyDB PostgreSQL for agents docs—keep every claim in Preview, and treat access-gated details as out of scope until Google publishes them publicly.