Publications

OpsAgent: An Evolving Multi-agent System for Incident Management in Microservices

Yu Luo1, Jiamin Jiang1, Jingfei Feng1, Lei Tao1, Qingliang Zhang1, Xidao Wen2, Yongqian Sun1,*, Shenglin Zhang1, Tong Liu3, Wenjie Zhang3, Dan Pei4

1Nankai University   2Alibaba Cloud   3Lenovo   4Tsinghua University

ASE 2026

41st IEEE/ACM International Conference on Automated Software Engineering, Munich, Germany

* Corresponding author

Paper Code Story
OpsAgent teaser figure
OpsAgent refines heterogeneous observability data into textual evidence and coordinates specialized agents for transparent incident diagnosis.

Abstract

OpsAgent is a lightweight and self-evolving multi-agent system for incident management in microservices. It converts metrics, logs, and traces into structured textual evidence, coordinates specialized agents for anomaly detection, failure triage, and root cause localization, and improves through model refinement and accumulated operational experience. The system is designed to make incident diagnosis more auditable, transferable, and adaptive to real operational environments.

Project Overview

Training-Free Evidence

Heterogeneous observability data is distilled into compact textual evidence, making it easier for LLM agents to reason over operational signals.

Multi-Agent Diagnosis

An orchestrator coordinates task-specific agents so diagnosis becomes a structured collaboration rather than an unconstrained conversation.

Self-Evolution

The system combines refinement and accumulated operational experience, allowing the diagnostic workflow to improve from past incidents.

Method

OpsAgent method overview
The overview shows how processed evidence, coordinated agents, cross-review, and self-evolution form the OpsAgent workflow.

Citation

@inproceedings{luo2026opsagent,
  title={OpsAgent: An Evolving Multi-agent System for Incident Management in Microservices},
  author={Luo, Yu and Jiang, Jiamin and Feng, Jingfei and Tao, Lei and Zhang, Qingliang and Wen, Xidao and Sun, Yongqian and Zhang, Shenglin and Liu, Tong and Zhang, Wenjie and Pei, Dan},
  booktitle={Companion Proceedings of the 41st IEEE/ACM International Conference on Automated Software Engineering},
  year={2026}
}