ConMem: Structured Memory-Guided Adaptation in Training-Free Multi-Agent Systems

2026-06-07

AI Summary

This work addresses the limitations of existing large language model–based multi-agent systems, which suffer from sensitivity to noisy trajectories, inadequate modeling of memory-skill relationships, and reliance on additional training or high-quality supervision. To overcome these challenges, the authors propose ConMem, a novel framework that introduces, for the first time, a training-free, relation-aware structured memory mechanism. ConMem constructs structured experience units via memory card distillation, explicitly models skill dependencies through a relation-aware memory graph, and dynamically integrates historical strategies using task-driven retrieval and graph coordination. Evaluated across multiple benchmarks and mainstream architectures, ConMem significantly outperforms current approaches, reducing reasoning candidate sets by over 50% and cutting planning overhead by more than 80%, thereby substantially enhancing the efficiency and adaptability of multi-agent collaboration.

Abstract

Recent advances have improved the adaptive capabilities of LLM-based multi-agent systems (MAS) through memory-, skill-, and learning-based approaches, yet these approaches remain challenged by noisy trajectories, insufficient modeling of memory-skill relations, and reliance on additional training or high-quality supervision. To address these limitations, we propose ConMem, a relation-aware and training-free framework that enables efficient multi-agent adaptation through cross-experience coordination. Specifically, ConMem distills historical interaction trajectories into structured memory cards to capture reusable strategies and cues, organizing them into a relation-aware memory graph. At runtime, ConMem retrieves cards according to task needs and coordinates them through the card graph to resolve strategy conflicts and recover their dependencies. Combined, these modules yield structured and relation-aware guidance, enabling robust, lightweight adaptation in multi-agent systems without additional training. Extensive experiments across multiple benchmarks and mainstream MAS architectures show consistent gains over existing memory architectures, with improved inference-time efficiency through pruning more than 50% of expanded candidates and reducing planning overhead by over 80%. Our codes are available at https://anonymous.4open.science/r/ConMemCode

Problem

• multi-agent systems
• memory-skill relations
• training-free adaptation
• noisy trajectories
• adaptive capabilities

Innovation

• structured memory
• relation-aware graph
• training-free adaptation
• multi-agent systems
• memory-guided coordination