A Multi-Agent Agentic Framework for Autonomous Cloud Infrastructure Monitoring, Anomaly Detection, and Self-Healing

Abstract

Modern cloud infrastructure management demands continuous vigilance across distributed services, log streams, and resource metrics at a scale that overwhelms human operators and rule-based systems alike. This paper presents a novel agentic, multi-tenant platform that deploys seven specialized AI agents—Log Intelligence, Crash Diagnostic, Resource Opti-mization, Anomaly Detection, Recovery, Recommendation, and Cost Optimization—coordinated by a central orchestrator. Each agent is backed by large language models (LLMs) accessed through LangChain, supporting pluggable providers including Google Gemini, OpenAI GPT, Anthropic Claude, and Groq. Asynchronous inter-agent communication is realized via Rab-bitMQ message queues, real-time state propagation through Redis Pub/Sub, and persistent storage on MongoDB. A statistical log-filtering pipeline reduces raw CloudWatch log volume by up to 98.8% before LLM inference, making the system economi-cally viable at production scale. Confidence-gated decision logic governs autonomous recovery actions: high-confidence diagnoses trigger immediate auto-healing, while low-confidence scenarios escalate to collaborative multi-agent analysis or human review. Experimental results demonstrate 93.9% anomaly detection pre-cision, 85% recovery action accuracy, and a full-pipeline median latency of 20.3 seconds from log ingest to completed remediation, establishing our framework as a practical foundation for next-generation AIOps platforms.

Introduction

The text presents a research framework for an AI-driven multi-agent AIOps system designed to manage the growing complexity of cloud-native infrastructures such as microservices, containers, and serverless applications. These environments generate massive volumes of logs and telemetry, making traditional monitoring approaches (dashboards, threshold alerts, manual debugging) inefficient due to alert fatigue and slow incident resolution.

To address this, the paper proposes an autonomous, LLM-powered multi-agent platform that uses technologies like Node.js, React, LangChain, MongoDB, Redis, and RabbitMQ. The system introduces a structured set of seven agents responsible for different stages of the incident lifecycle, including log analysis, anomaly detection, diagnosis, recovery, and cost optimization. A key feature is a confidence-gated collaboration mechanism, where agents consult each other when their certainty is low, improving reliability and reducing false alerts. It also includes a log-filtering pipeline that significantly reduces noise before LLM processing, improving scalability and cost efficiency, and supports multiple LLM providers for flexibility.

The system is built using a three-tier architecture (frontend, backend orchestration layer, and infrastructure layer) with strong support for multi-tenancy, ensuring each organization has isolated and secure environments. The backend coordinates agents using asynchronous messaging and shared state systems.

The framework extends prior work in AIOps, anomaly detection, and self-healing systems by combining statistical methods with LLM-based reasoning, enabling more contextual, adaptive, and explainable decision-making.

Conclusion

We have presented a multi-agent agentic platform that autonomously monitors, diagnoses, and heals cloud infrastruc-ture through coordinated AI agents backed by large language models. The proposed architecture introduces a confidence-gated collaboration protocol that balances autonomous action against human oversight, a tiered log-filtering pipeline that achieves up to 98.8% noise reduction before LLM inference, and a pluggable multi-provider LLM layer that decouples AI capabilities from vendor commitments. Experimental results on simulated workloads demonstrate 93.9% anomaly detection precision, 85% recovery action accuracy, and a full-pipeline median latency of 20.3 seconds from log ingest to completed healing action. Our frame-work establishes a practical, extensible foundation for next-generation AIOps systems that blend statistical rigor, LLM-powered reasoning, and safety-first autonomous action.

References

[1] Y. Dang, Q. Lin, and P. Huang, “AIOps: Real-world challenges and research innovations,” in Proc. IEEE/ACM ICSE-SEIP, 2019, pp. 4–5. [2] P. He, J. Zhu, Z. Zheng, and M. Lyu, “Drain: An online log parsing approach with fixed depth tree,” in Proc. IEEE ICWS, 2017, pp. 33–40. [3] R. Vaarandi and M. Pihelgas, “LogCluster—a data clustering and pattern mining algorithm for event logs,” in Proc. CNSM, 2015, pp. 1–7. [4] M. Du, F. Li, G. Zheng, and V. Srikumar, “DeepLog: Anomaly detection and diagnosis from system logs through deep learning,” in Proc. ACM CCS, 2017, pp. 1285–1298. [5] Q. Wu et al., “AutoGen: Enabling next-gen LLM applications via multi-agent conversation,” arXiv preprint arXiv:2308.08155, 2023. [6] T. Lorido-Botran, J. Miguel-Alonso, and J. Lozano, “A review of auto-scaling techniques for elastic applications in cloud environments,” J. Grid Comput., vol. 12, no. 4, pp. 559–592, 2014. [7] S. Yao et al., “ReAct: Synergizing reasoning and acting in language models,” in Proc. ICLR, 2023. [8] J. Audibert, P. Michiardi, F. Guyard, S. Marti, and M. Zuluaga, “USAD: UnSupervised anomaly detection on multivariate time series,” in Proc. ACM KDD, 2020, pp. 3395–3404. [9] J. Kephart and D. Chess, “The vision of autonomic computing,” IEEE Comput., vol. 36, no. 1, pp. 41–50, 2003. [10] H. Chase, “LangChain,” GitHub repository, 2023. [Online]. Available: https://github.com/langchain-ai/langchain

Copyright

Copyright © 2026 Diddi Siddarth, Marupaka Suchith Kumar, Madishetty Vasanth, Merugu Vamshi Vishwanath, Surisetti Balraj, Mr. P. Subbarao . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.