OpsMem: Dual-Memory Reasoning with Cross-Memory Resonance for Failure Diagnosis

OpsMem introduces a dual-memory framework for iterative software-failure diagnosis. Short-term memory tracks the evolving diagnostic state, while long-term memory stores reusable operational experience. A cross-memory resonance mechanism retrieves state-relevant experience, conditions multi-agent diagnosis on both memory types, and consolidates reusable knowledge from resolved incidents. On a real-world Huawei microservice failure-diagnosis dataset, the abstract reports improvements of up to 46.88% in Match and 18.39% in Relevant over the strongest baseline. The supplied information does not include detailed methodology, ablations, cost analysis, or statistical significance results.