opensearch-project/agent-health

特工健康

开源AI代理评估与可观察性

Agent Health 帮助您评估、监控和优化各种 AI 代理——编码助手、运维/RCA 代理、客户支持代理、数据分析代理、检索/发现代理以及多代理工作流。它提供实时执行流、基于大型语言模型的评估（带轨迹比较）、批实验以及通过 OpenTelemetry 追踪进行深度可观察性——开箱即用本地存储，或在您需要共享集群时存储在 OpenSearch。

网站 • Slack（闲暇） • 推特/X • 演示视频 • 文献资料 • 更新日志

什么是Agent Health？• 人工智能技能 • 安装 • 特色 • 配置 • 贡献

对代理评估运行进行通过率、准确性、成本和绩效指标的并排比较。

---

什么是健康特工？

代理健康是一个基于OpenSearch构建的AI代理评估与可观察性框架。

Open-source AI Agent Evaluation & Observability

Agent Health helps you evaluate, monitor, and optimize AI agents of any kind — coding assistants, ops/RCA agents, customer-support agents, data-analysis agents, retrieval/discovery agents, and multi-agent workflows. It provides real-time execution streaming, LLM-based evaluation with trajectory comparison, batch experiments, and deep observability through OpenTelemetry traces — stored locally out of the box, or in OpenSearch when you want a shared cluster.

Website •
Slack •
Twitter/X •
Demo Video •
Documentation •
Changelog


    


Side-by-side comparison of agent evaluation runs with pass rate, accuracy, cost, and performance metrics over time.

---

What is Agent Health?

Agent Health is an evaluation and observability framework for AI agents, built on OpenSearch. It helps you measure agent performance through “Golden Path” trajectory comparison — where an LLM judge evaluates agent actions against expected outcomes — and provides deep observability into agent execution via OpenTelemetry traces.