The database hands your agent its own spec

You give an agent database access by giving it two URLs. The first is the database. The second is GET /api/v1/sqlite/openapi.yaml: the service's own machine-readable spec, every endpoint, parameter, and response shape, published by the database itself.

The agent reads the spec, then operates the full surface: transactions, key-value writes, history queries, rollbacks. No tool definitions to hand-write, no wrapper library to maintain, no drift between what your integration notes claim and what the API accepts: the spec ships with the service and is versioned with it. When the surface grows, the agent's knowledge grows on the next fetch.

You didn't integrate the database with your agent. The database introduced itself.

The database introduced itself.