SQLSage: a local PostgreSQL SELECT reviewer for Supabase query investigations

I'm developing SQLSage, a free MIT CLI for reviewing PostgreSQL SELECT queries. I'm looking for feedback from Supabase builders who have a query they are already investigating.

It works with local schema DDL/catalog files and optional saved PostgreSQL plans. It explains the result, flags supported correctness/performance risks, and separates predictions from observed evidence. No Supabase service-role key, database connection, or query execution is needed for the file workflow.

Here is a small reproducible example for a public-schema table:

schema.sql

CREATE TABLE public.orders (
  id bigint PRIMARY KEY,
  customer_id bigint NOT NULL,
  status text NOT NULL,
  created_at timestamptz NOT NULL
);

query.sql

SELECT id
FROM public.orders
WHERE status = 'paid'
ORDER BY created_at DESC
LIMIT 50;

With Node.js 22.18+:

npx --yes sqlsage@0.2.1 analyze --query query.sql --schema schema.sql --format markdown

I ran this exact published-version example. It flags that ordering only by created_at does not fully order tied rows; ties at the limit boundary can change which rows are returned. It also labels its execution description as an offline prediction and makes no speedup claim. You can start with npx --yes sqlsage@0.2.1 demo instead.

This is not an RLS/security audit, server workload monitor, or automatic SQL rewriter. Schema/query support is limited; it does not establish that a Supabase policy is correct. The current feedback trial covers single-query reports. Plan comparison is excluded pending a query-identity repair.

If you have 15–20 minutes to try one real supported SELECT locally, the opt-in and feedback issue is here. I'm trying to learn what the report changes in your investigation, where it is wrong/confusing, and whether you use it for another real query.

Please keep SQL, schemas, plans, reports, credentials, and customer/employer data local. The public issue asks only for broad use cases and plain-language observations. No purchase is required; the MIT CLI remains free.