Show HN: Guardrail Layer – self-hosted AI data layer for secure DB chat (github.com)

Guardrail Layer is an open-source, self-hosted backend that sits between your database and any LLM, automation, or dashboard to enforce privacy and access rules at query time. It automatically validates connections to Postgres/MySQL, introspects schemas to create AI-ready metadata, and applies safe defaults like automatic redaction (emails, IDs, sensitive values), row-level access, and policy-based filters. It also offers natural-language → SQL translation for safe DB chat and can expose read-only APIs, all deployed via Docker Compose for on-prem or controlled-cloud setups. For the AI/ML community this matters because it tackles a common risk: leaking sensitive database fields when exposing data to LLMs or analytics bots. By inserting a policy‑aware “bodyguard” layer, teams can prototype chatbots, analytics UIs, or LLM integrations without sacrificing compliance. Technically, Guardrail Layer is built with Node.js/Express, Prisma, TypeScript, PostgreSQL, and a React/Tailwind frontend (Vite), and is designed to be extensible. It’s early-stage (Apache‑2.0) and seeking contributors—useful for practitioners wanting a practical, self-hosted solution to secure DB-LLM interactions.