Intelligent Mixture-of-Models Router for Efficient Heterogeneous LLMs Inference

LLM Semantic Router: Intelligent Mixture-of-Models (MoM) System with Privacy Preservation and Prompt Guard. The semantic router intelligently directs OpenAI compliant API requests to the most suitable backend models based on semantic understanding of request content.

Lightweight, Self-Hosted AI Guardrails Model based on ModernBERT.

Local DeepSeek V4 proxy: reduces hallucinations strengthens system prompt following stops reasoning loops repairs empty tool-call replies — with built-in session analysis and GUI. / 本地 DeepSeek V4 代理：降低幻觉、增强 system prompt 遵循、防思维循环、修复空回复 自带会话分析与 GUI。

Pure-Rust prompt-injection detector with 1.5MB embedded MLP classifier. 98.40% accuracy, p50 14ms CPU inference, bindings for Python/JS/Go. Apache-2.0/MIT alternative to Rebuff (archived) and Lakera Guard.

A Input Prompt Guard Library with Bert model trained on prompt injection dataset

A web app for testing Prompt Guard, a classifier model by Meta for detecting prompt attacks.

Fullstack local agentic system. Prompt guard & Lua code gen/validation. Ollama-hosted models. REST, SSE. Rich CLI / NextJs Clients. Built for True Tech Hack 2026 by OpenTeam2026 (Tim K. - Lead).

Advanced AI Agent playground with Gemini/GPT integration, supporting mocked/production RAG, history compression, and detailed data provenance for logic validation.

AI-powered tool for detecting and classifying prompt safety, identifying potential injections and jailbreaks in both raw text and code inputs.

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

Fork for LLM security evaluation, prompt guard patterns, and trusted prompt/runtime safety research.