**[Is Shared Memory Destroying Multi-Agent Systems? DecentMem Improves Accuracy by 24% While Halving Token Usage]** According to monitoring by Beating, a team from the University of Cambridge and the University of Chicago has open-sourced a multi-agent memory framework called DecentMem, which replaces global shared memory with decentralized private memory. Traditional systems commonly use shared memory, but when agents read the same context, they often converge on similar decision paths, causing the advantages of specialization to disappear. The core idea of DecentMem is that collaboration must rely on cognitive diversity, and retaining private memory is essential to maintaining complementary paths. DecentMem enables agents to maintain their own dual-pool memory: the Experience Pool (E-pool) stores historical experiences and reflection records, while the Exploration Pool (X-pool) continuously generates new candidate ideas. An online decision-maker dynamically adjusts the weighting of the two pools based on periodic scores from a large language model, helping agents autonomously balance exploitation and exploration. Theoretically, self-evolutionary search is modeled as a random walk on a graph, with global reachability ensuring agents can escape local optima. In tests on AutoGen, DyLAN, and AgentNet, DecentMem achieved an average improvement of 8.6% compared to the strongest centralized memory baselines, with relative improvements of up to 23.8% in the best scenarios, while halving token consumption. The research found that the more collaboration relies on improvisational discussion rather than fixed processes, the more pronounced the advantages of decentralization. In the DyLAN framework, which emphasizes free negotiation, the number of iterations required to achieve equivalent performance decreased by approximately 60%, and convergence speed improved by about 2.5 times. [Original Link]