Show HN: An open-source LLM guardrail powered by Jev

Category: security

Tags: llm, security, guardrail, owasp, semantic-analysis

Score: 5.8/10 (Innovation: 6, Technical: 5, Documentation: 7, Utility: 5)

Jev Guard is a contextual LLM guardrail that classifies text as user input, retrieved content, tool output, or model output, returning an allow/review/block decision alongside OWASP LLM Top 10 risk scores. It wraps the Jev semantic scoring API with a versioned decision policy, a Node.js demo, an npm-ready library, and a CLI. The project is notable for its detailed OWASP-aligned risk taxonomy, quotation-scope clarification logic, and explicit distinction between textual risk signals and calibrated vulnerability probabilities.

Target audience: AI/ML engineers and security-conscious backend developers building LLM applications

Repository: https://github.com/gulbaki/jev-llm-guard · JavaScript · MIT · 14 stars

View on Hacker News