
Adrian
secureagentics
Runtime security monitoring and control for AI agents. Catches 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引擎。它通过分析代理活动日志(工具调用、行为、输出)和推理轨迹来检测恶意、不一致或超出权限的行为,并在必要时进行干预。该项目使用Python编写,支持快速集成到LangChain代理中,仅需两行代码即可完成初始化和关闭操作。Adrian特别适合需要确保AI系统按照预期安全运行的场景,如企业级应用、自动化流程管理等,能够有效防止工具滥用、提示注入及策略漂移等问题。
Python
Apache License 2.0264
Stars
52
Forks
5
Watchers
4
Issues
Star 增长
今日0
近 7 天+118
近 30 天+206
综合评分75.17
默认分支main