secureagentics

Adrian

secureagentics

Open-source runtime AI agent security tool - monitors and controls AI agents, catching malicious tool use, prompt injection, and policy drift in real time, before the agent acts.

AI 简介

Adrian 是一个开源的 AI 代理运行时安全监控与控制工具,用于实时检测并干预恶意工具调用、提示注入和策略偏移等风险行为。它通过分析代理的日志(如工具调用、动作序列)和推理轨迹实现深度可观测性,支持 LangChain/LangGraph 生态的轻量级集成(仅需两行代码),提供审计或主动阻断两种策略模式。适用于企业级 AI 应用部署、合规敏感场景(如金融、医疗)及需要对 LLM 代理行为进行细粒度治理的开发团队。

Python
Apache License 2.0
382
Stars
78
Forks
8
Watchers
2
Issues

Star 增长

今日0
近 7 天0
近 30 天+58
综合评分51.49
默认分支main