AWS shows how Postgres can give AI agents business context
Its EDB walkthrough separates the meaning of database fields from permission to read them.
The AWS walkthrough connects EDB Postgres AI with Bedrock AgentCore for customer-data questions. In its example, a comment explains that a purchase price means the whole order, preventing confusion with a unit price. AWS also reruns a question under reduced database permissions. These are the provider’s examples, not tests we performed.
EDB’s schema guide describes an index of table, view and column definitions and comments, rather than the stored customer records. Foreign keys describe relationships between tables. Creation requires an embedding model, which converts descriptions into searchable numerical representations. The database schema is crawled immediately when the knowledge base is created; this initial work is separate from handling later changes.
For those later changes, the processing reference specifies Disabled as the default: no automatic updates occur, so an explicit refresh is needed after schema changes. Live mode uses database triggers; Background mode processes queued changes through a worker.
The role guide separates password and OAuth access. Password access requires a login role and a matching password-checking rule; trust and peer are refused on this route. OAuth bearer access instead needs an oauth rule. Membership in aidb_users enables tool listing and calls; schema and table privileges must also be granted. EDB recommends grouping privileges into profile roles. Revoking a profile role takes effect on the next call, which opens a fresh connection. Setting NOLOGIN prevents the agent from connecting.