Autonomous Multi-Agent System for Integrated SRE and Self-Healing in Cloud-Native Environments

Abstract: Modern software systems, characterized by their complexity and distributed nature, pose significant challenges for Site Reliability Engineering (SRE) and DevOps teams. Traditional incident response processes are predominantly manual, reactive, and struggle to scale, leading to prolonged Mean Time to Resolution (MTTR) and increased operational overhead. This paper introduces an autonomous, self-healing framework designed to address these limitations. The proposed system leverages a multi-agent architecture, built using CrewAI and powered by advanced Large Language Models (LLMs) like DeepSeek-R1 and GPT-4o, to automate the entire incident management lifecycle. By integrating seamlessly with standard DevOps toolchains, including Kubernetes for orchestration, Middleware.io for monitoring, Jira for incident tracking, and GitHub for version control, the framework moves beyond simple alert triaging to proactive, automated remediation. The system demonstrated exceptional performance in a simulated production environment, achieving up to 96% accuracy in root cause analysis (RCA), a 73% success rate in automated code fix generation, and reducing the average end-to-end resolution time from hours to under 28 min. This work establishes an auditable, human-in-the-loop solution that significantly reduces manual effort, minimizes system downtime, and enhances the resilience of modern software operations.