MemoWM: How World Models Change What Agents Need to Remember

2026-10-07

AI Summary

This study addresses the escalating storage demands of long-lived agents caused by continuous experience accumulation. To mitigate this, we propose a world model-based framework for memory compression and conditional allocation. Specifically, the approach leverages world models to capture reusable regularities, enabling information compression and reconstruction through shared predictions. Furthermore, a task-aware memory allocation strategy is designed to dynamically balance reconstruction error against storage costs, precisely retaining high-value downstream information that transcends prior knowledge. Our experiments validate the effectiveness of employing strong models to reduce per-experience storage overhead. The proposed method achieves a 2.62% improvement in accuracy while reducing storage requirements by 53.9%, demonstrating simultaneous optimization of both performance and efficiency.

Abstract

Long-term agents face growing storage demands as they accumulate experience. World models capture reusable regularities that can reduce the information stored for each experience. We formulate the problem of memory allocation conditioned on a world model and introduce MemoWM, a framework that uses shared predictions to compress retained information and reconstruct omitted content. Its task-aware allocation rule balances the expected impact of reconstruction errors against storage cost, retaining information with downstream value beyond the predictive prior. Across five long-term agent-memory benchmarks, MemoWM achieves 42.42\% average answer accuracy, exceeding the strongest baseline by 2.62 percentage points, while reducing average experience-specific storage by 53.9\% relative to MIRIX, the most storage-efficient baseline. Further analysis shows that stronger world models reduce per-experience storage at comparable task quality. Accounting for model parameters reveals a trade-off between shared model capacity and recurring storage costs, with the capacity that minimizes total storage increasing as more interactions are retained. Our code is available at https://github.com/Feld-maxiu/MemoWM.

Problem

• long-term agents
• memory allocation
• world models
• storage demands
• memory compression

Innovation

• World Models
• Memory Allocation
• Long-term Agents
• Information Compression
• Storage Optimization