About

The verdict before the query

Veto is a deterministic safety and cost gate for the SQL that AI coding agents now write and run. It exists because the fix for a probabilistic system isn't a better prompt — it's a check that can't be talked out of a block.

Why we built it

AI agents are increasingly trusted to write and execute database migrations and queries. The moment you give a non-deterministic system the ability to act on production, the range of its possible actions includes the destructive ones — and "ask the model to double-check itself" just adds more of the same probabilistic judgment. We wanted the safety layer to be the reliable part: external to the agent, deterministic, and auditable. So we built one for SQL. There's a longer version of this argument on the blog.

What Veto is

An MCP server whose analyze_sql tool returns a structured ok / warn / block verdict for Postgres SQL — catching destructive operations, locking risk, correctness traps, and query cost before a statement runs. There is no LLM in the core: the same input always yields the same verdict, and it never connects to your database. See the docs to connect it.

Who builds it

Veto is built and operated by MetaWear s.r.o., an independent software studio based in Ostrava, Czech Republic (IČ 14338017, registered at the Regional Court in Ostrava, file no. C 88655). We build small, sharp developer tools and ship them ourselves — so support comes straight from the people who wrote the code.

Get in touch

Questions, feedback, or a tricky verdict you want to discuss? Email info@metawear.cz or visit metawear.cz.