I’ve been working on something called Guardrail Layer, an open-source system that sits between your databases and the tools (APIs, LLMs, automations) that query them.

It’s basically a data safety layer, it lets you connect to MySQL or Postgres, automatically redacts sensitive fields, and lets you run or even ask natural-language queries safely.

It’s built with Node.js, TypeScript, Prisma, Postgres, and Docker, and you can run it with a single docker compose up.

It lets you add metadata to help the AI understand your structure better. Also allows you to redact rows in your DB to keep them safe and secure from users and the AI itself.

GitHub: https://github.com/tyoung1996/guardrail-layer

It’s still early (alpha stage) but already functional, I’d love feedback from folks building with LLMs or connecting internal databases to AI tools.

Thanks!