llm-guardrail

Python SDK for LLM guardrails with safety classification, PII detection, prompt injection defense, and grounding checks. Protect AI applications locally with zero external APIs, streaming support, and plug-and-play integration for any LLM.

A discipline layer for Claude Code. Hooks at every tool boundary, a workflow that runs from intake to commit, and a small constitution the agent cannot bypass.

AANA is a pre-action control layer for AI agents: agents propose actions, AANA checks evidence/auth/risk, and tools execute only when the route is accept.

Real-time LLM hallucination guardrail — NLI + RAG fact-checking with opt-in claim-level streaming contradiction halt. Drop-in for any LLM backend.

LLM Guard - A pure Rust implementation guardrails for LLM input/output.

Container images built from upstream sources or custom Dockerfiles, with automated security scanning, release management, and SLSA provenance

ARCHIVED - A practical hybrid safety middleware for Large Language Models. It runs both a FastAPI production endpoint and the original interactive CLI with heartbeat. For AI labs, red/blue teaming, prompt engineering

Regex backstop for German HWG & EU 1924/2006 compliance. Flags health claims in LLM-generated marketing copy. 14 patterns, MIT licensed, zero deps.