PostgreSQL

PostgreSQL

Connection string (host, port, database, user, password). Reads use list rows, search (SQL or structured syntax), or execute — not colon smart IDs.

Data Model

• Tables — relational storage for records.
• Parameterized reads — list rows, search, or execute with bound SQL.
• SQL execution — ad hoc DDL/DML, including batched statement bundles.
• Schema intelligence — foreign keys, routines, triggers, and dependency graphs.
• Performance — plans, statistics, index health, contention signals, and DBA tuning (top queries, health checks, hypopg).
• Data quality — profiling, duplicate detection, constraint validation, anomaly scans.

Operational order (unfamiliar schema)

1. List tables or AI schema narration — table and column names before SQL.
2. Inspect one table — table='schema.table' for column detail, keys, and indexes.
3. List rows, search, or execute SQL — never guess column or parameter names.

Gotchas

• Single-row reads: postgres_list_rows with where on the primary key, or postgres_execute — not legacy postgres:row:… handles.
• Read-only paths vs mutating SQL: Prefer guarded read paths for exploration; wrap destructive statements in explicit transactions.
• Change rehearsal: Dry-run style previews exist for risky DML—use them before committing bulk updates.
• Local databases: Reach localhost through the desktop proxy tunnel or a connection string that targets loopback from the execution environment.