Nothing to hand? Load the — a schema with the classic type, key and RLS mistakes — or the , a mostly-clean events schema where the problem is indexing and pagination.
Paste the SQL
A schema, a migration, one slow query, or all of it. The instant prescan reads it for free while you type and lists what it mechanically found: varchar(n), naive timestamps, float money columns, int primary keys, OFFSET pagination, unindexed foreign keys, unwrapped auth.uid() in policies, and the tables and workload signals in your paste.
The AI reviews it
Severity-ranked findings, each quoting the SQL it concerns and carrying a runnable fix; the missing indexes as complete CREATE INDEX statements; and an honest health check across schema design, indexing, query performance, security & RLS, and operations. Every prescan hit is confirmed or explicitly set aside — including the false positives.
Take the fixes and re-review
Greenfield? Take improved.sql, the corrected rewrite of your own SQL. Already live? Take fixes.sql — every finding’s corrective statement in one runnable file, high severity first — plus the missing indexes, optionally rewritten as CONCURRENTLY so they build without blocking writes. Then press Use as new input & re-review: the next review is diffed against this one, so you see exactly what your edits resolved, what is still open, and what you introduced. History is saved to your account, not just this browser.
Derived from the @affaan-m/postgres-patterns skill (based on the Supabase agent skills).