AI-Agent-Automation: A Multi-Agent-Based AIOps Automated Operation and Maintenance Platform

This article introduces the open-source multi-agent AIOps and platform engineering automation system AI-Agent-Automation, maintained by imtarget05 and released on GitHub (2026-05-30). The system integrates LangGraph orchestrator, local LLM (Ollama), RAG knowledge base, and n8n visual workflow to achieve automatic fault detection, root cause analysis, and repair for Kubernetes and Prometheus infrastructures, with a core multi-agent collaboration architecture.

Under modern cloud-native architectures, the complexity of Kubernetes clusters, the explosion of Prometheus monitoring data, and the fault propagation chain of microservices make traditional manual operation and maintenance unsustainable, with long fault localization times. The rise of LLM brings possibilities for operation and maintenance automation, but integrating it into workflows is an industry challenge. AI-Agent-Automation was born in this context to build a complete intelligent operation and maintenance agent system.

The system adopts a five-layer architecture:

1. Orchestration Layer: LangGraph framework, which defines agent interactions with graph structures, supports loops, conditional branches, and state management, and flexibly handles different fault processes.
2. Inference Layer: Prioritizes support for local LLM (Ollama integration) to ensure privacy and compliance in data-sensitive environments, while retaining scalability for cloud-based models.
3. Knowledge Layer: RAG system that encodes fault records and Runbooks into a knowledge base, automatically retrieves similar cases to assist decision-making.
4. Execution Layer: n8n visual workflow engine that connects AI decisions with operation and maintenance actions (service restart, scaling, etc.) without requiring extensive code.
5. Monitoring Layer: Real-time dashboard displays metrics such as agent status and task queues, with multi-layer Guardrails mechanisms to ensure operation controllability.

The system supports three types of scenarios:

1. Intelligent Fault Response: After a Prometheus alert is triggered, the detection agent confirms the fault, the root cause analysis agent collects logs/metrics for reasoning, RAG provides repair suggestions, executes the repair, and records the process.
2. Preventive Maintenance: Regularly analyzes cluster resource trends, predicts capacity bottlenecks, triggers scaling suggestions or strategies to avoid service interruptions.
3. Knowledge Precipitation and Inheritance: Automatically extracts information to update the knowledge base after fault handling, shortens the learning curve for new engineers, and reduces service fluctuations caused by experience differences.

The project's technology stack selection balances practicality and forward-looking:

• LangGraph instead of self-developed orchestration: Leverages the mature framework's concurrency control and state management capabilities to reduce development complexity.
• Local LLM priority: Meets enterprise data compliance requirements and reduces API costs.
• n8n as the execution layer: Uses its rich integration ecosystem to quickly connect to various infrastructures.
• Modular design: Loosely coupled components for easy replacement or expansion.

The current project is in the early stage and faces challenges:

1. Model Hallucination: LLM may produce incorrect conclusions in root cause analysis, requiring manual review.
2. Context Window Limitation: Large volumes of complex fault logs may exceed the model's processing capacity.
3. Action Security: Automated operations carry risks, requiring more fine-grained permission control. Future directions: Introduce multi-modal processing of monitoring charts, combine reinforcement learning with operation and maintenance feedback, and develop more intelligent predictive maintenance algorithms.

AI-Agent-Automation is an important exploration in the AIOps field, combining LLM reasoning capabilities with a multi-agent collaboration architecture to build an autonomous operation and maintenance system. Although there is still a gap from fully autonomous "unmanned operation and maintenance", it provides a reference architecture paradigm and is an open-source solution worth attention for teams exploring operation and maintenance intelligence.