Memory Beyond Recall: A Dual-Process Cognitive Memory System for Self-Evolving LLM Agents

2026-06-08

AI Summary

Current memory systems in large language model (LLM) agents are largely confined to surface-level recall and struggle to support implicit, evolving user personalization—such as belief revision, causal reasoning, and cross-domain abstraction. This work proposes the Dual-Process Cognitive Memory (DCPM) system, which for the first time integrates dual-process cognitive theory into LLM memory architecture. DCPM organizes memories hierarchically—from raw inputs to cross-domain patterns—via a synergistic mechanism of daytime synchronous writing and nighttime asynchronous reasoning. It employs a dual-link structure to trace belief evolution and leverages a nighttime inference engine to induce abstract patterns and detect cross-domain conflicts, thereby constructing high-level core schemata. Evaluated on LongMemEval, PersonaMem, and PersonaMem-v2 benchmarks, DCPM significantly enhances implicit cross-session reasoning performance (up to +5.20 points), while offering limited gains on pure episodic recall tasks.

Abstract

Long-term memory for an LLM agent is more than retrieving the right passage at the right time. Current memory systems collapse belief revision, causal coupling, and cross-domain abstraction into a single retrieval surface tuned for surface recall, and consequently struggle on implicit personalisation that requires reasoning over how a user has evolved. We propose DCPM, which reorganises agent memory along a cognitive capability hierarchy ascending from raw inputs and atomic facts, through diachronic belief trajectories and identity, to domain schemas, latent intentions and cross-domain patterns. The hierarchy is driven by two processes inheriting the architectural split of dual-process theory: a synchronous daytime writer (System1) that records belief revisions as doubly linked supersedes chains, and an asynchronous nighttime engine (System2) that induces schemas and intentions and sweeps for cross-domain collisions abstracted into higher-level core schemas. On LongMemEval, PersonaMem and PersonaMem-v2, enabling System2 contributes most where the benchmark rewards implicit cross-session inference (up to +5.20 on PersonaMem-v2) and least on span recall, matching the architectural prediction.

Problem

• long-term memory
• implicit personalisation
• belief revision
• cross-domain abstraction
• LLM agents

Innovation

• dual-process memory
• cognitive hierarchy
• belief revision
• cross-domain abstraction
• self-evolving LLM agents