security tool · postgres · row-level security
The prompt can't guarantee tenant isolation. The database can.
An AI SQL agent over Postgres where row-level security is enforced by the connection, not the model — with every query audited and a hard token cap.
read-only by defaultRLS enforced by the connectionquery audit loghard token cap
free vs. paid
same code. one line, hardened.
The free core proves the model — the paid version is what you put in front of real tenants.
free · MIT · on GitHub
free, MIT, on GitHub
- NL → SQL over Postgres with RLS on
- Read-only, single tenant model
- Query audit log + hard token cap
- The RLS-bypass test that proves the leak
$0 · MIT license · clone it today
view on GitHub →paid · one-time
the $149 version
- Multi-tenant isolation across many tenants/policies
- Write queries behind an explicit confirmation flow
- Cost dashboard (per-tenant token & query spend)
- The full RLS-bypass test suite — every attack from the article, red-teamed on every run
$149 · one-time · same code, hardened shelf
get it when it ships →the proof
I attacked my own agent — 5 ways an LLM bypasses row-level security.
RLS scopes rows by a setting the query can change. The prompt can't stop it — the connection has to. Five real bypasses, and the fix for each — the same bypasses the $149 suite tests for, automatically, every run.
product waitlist
leave your email — get it when it ships
No spam. One email when the $149 hardened version is ready.