MindMemOS: A Portable and Self-Evolving Memory Operating Layer for AI Agents

2026-08-12

AI Summary

This work addresses the challenge that existing AI agents lack memory systems capable of adaptive evolution after deployment. The authors propose a portable, self-evolving memory operating system that organizes open-world information using a unified entity–attribute–time structure, enabling scene-adaptive modeling and continual learning. Key innovations include the MindMemEvolve algorithm, which drives memory schema evolution through verification-guided evolutionary search, incorporates a “dreaming” mechanism for memory compression and conflict resolution, and leverages implicit human feedback to correct biases; and MindSkillEvolve, which distills execution trajectories into reusable and refinable skills. Experiments demonstrate that the approach achieves 94.03% accuracy on LOCOMO and 70.63% on PersonaMem, and improves task success rate by 9.2 percentage points over the baseline on SpreadsheetBench.

Abstract

Memory is a core component of AI agents, enabling them to accumulate experience, maintain personalization, and adapt over long-term interactions. However, existing memory systems often remain fixed after development, limiting their ability to adapt their memory models, organization strategies, and procedural knowledge through continued use. We present MindMemOS, a portable and self-evolving memory operating layer that organizes open-world information using a unified entity property timestructure. MindMemOS supports scenario-adaptive memory modeling, higher-order pattern discovery, autonomous memory refinement, and continuous skill evolution. Its MindMemEvolve algorithm employs validation-driven evolutionary search to optimize memory schemas for target scenarios, whiledreaming consolidates accumulated memories by merging redundant records and resolving conflicts. In addition, implicit corrective feedback serves as a human-in-the-loop signal for identifying and revising potentially inaccurate or misaligned memories. Its MindSkillEvolve algorithm further transforms agent execution trajectories into reusable and progressively refined skills. MindMemOS achieves 94.03% accuracy on LOCOMO and 70.63% on PersonaMem. MindSkillEvolve improves SpreadsheetBench success by 9.2 percentage points over the initial-skill baseline.

Problem

• AI agents
• memory systems
• self-evolving
• adaptive memory
• long-term interaction

Innovation

• self-evolving memory
• memory operating system
• evolutionary memory schema
• autonomous memory refinement
• skill evolution