A Topology-Aware, Memory-Centric Architecture that Separates Root-Cause Derivation from Root-Cause Explanation

The article introduces O PSC ORTEX, a multi-agent prototype architecture designed to enhance fault diagnosis in microservice deployments by separating root-cause derivation from explanation. This system utilizes a structured operational memory that captures normal behavior and service dependencies, allowing for deterministic root cause analysis through a learned dependency graph, while a large language model assists in generating explanations and recommendations. This approach addresses the challenges of cascading failures in production environments, as validated by tests on an e-commerce benchmark with various failure scenarios, emphasizing the importance of operational memory in improving system observability and diagnosis.