Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Membox: Weaving Topic Continuity into Long-Range Memory for LLM Agents

Dehao Tao, Guoliang Ma, Yongfeng Huang, Minghu Jiang

TL;DR

Problem: existing LLM agent memories fragment discourse, harming narrative and causal flow. Approach: membox introduces Topic Loom to segment dialogue by topic and Trace Weaver to link boxes into long-range event timelines, preserving continuity. Contributions: formalizes topic continuity, designs a two-stage memory architecture, and demonstrates superior retrieval and temporal reasoning on LoCoMo with fewer context tokens. Significance: provides cognitively motivated, efficient memory for robust long-range reasoning in LLM agents.

Abstract

Human-agent dialogues often exhibit topic continuity-a stable thematic frame that evolves through temporally adjacent exchanges-yet most large language model (LLM) agent memory systems fail to preserve it. Existing designs follow a fragmentation-compensation paradigm: they first break dialogue streams into isolated utterances for storage, then attempt to restore coherence via embedding-based retrieval. This process irreversibly damages narrative and causal flow, while biasing retrieval towards lexical similarity. We introduce membox, a hierarchical memory architecture centered on a Topic Loom that continuously monitors dialogue in a sliding-window fashion, grouping consecutive same-topic turns into coherent "memory boxes" at storage time. Sealed boxes are then linked by a Trace Weaver into long-range event-timeline traces, recovering macro-topic recurrences across discontinuities. Experiments on LoCoMo demonstrate that Membox achieves up to 68% F1 improvement on temporal reasoning tasks, outperforming competitive baselines (e.g., Mem0, A-MEM). Notably, Membox attains these gains while using only a fraction of the context tokens required by existing methods, highlighting a superior balance between efficiency and effectiveness. By explicitly modeling topic continuity, Membox offers a cognitively motivated mechanism for enhancing both coherence and efficiency in LLM agents.

Membox: Weaving Topic Continuity into Long-Range Memory for LLM Agents

TL;DR

Problem: existing LLM agent memories fragment discourse, harming narrative and causal flow. Approach: membox introduces Topic Loom to segment dialogue by topic and Trace Weaver to link boxes into long-range event timelines, preserving continuity. Contributions: formalizes topic continuity, designs a two-stage memory architecture, and demonstrates superior retrieval and temporal reasoning on LoCoMo with fewer context tokens. Significance: provides cognitively motivated, efficient memory for robust long-range reasoning in LLM agents.

Abstract

Human-agent dialogues often exhibit topic continuity-a stable thematic frame that evolves through temporally adjacent exchanges-yet most large language model (LLM) agent memory systems fail to preserve it. Existing designs follow a fragmentation-compensation paradigm: they first break dialogue streams into isolated utterances for storage, then attempt to restore coherence via embedding-based retrieval. This process irreversibly damages narrative and causal flow, while biasing retrieval towards lexical similarity. We introduce membox, a hierarchical memory architecture centered on a Topic Loom that continuously monitors dialogue in a sliding-window fashion, grouping consecutive same-topic turns into coherent "memory boxes" at storage time. Sealed boxes are then linked by a Trace Weaver into long-range event-timeline traces, recovering macro-topic recurrences across discontinuities. Experiments on LoCoMo demonstrate that Membox achieves up to 68% F1 improvement on temporal reasoning tasks, outperforming competitive baselines (e.g., Mem0, A-MEM). Notably, Membox attains these gains while using only a fraction of the context tokens required by existing methods, highlighting a superior balance between efficiency and effectiveness. By explicitly modeling topic continuity, Membox offers a cognitively motivated mechanism for enhancing both coherence and efficiency in LLM agents.
Paper Structure (25 sections, 4 equations, 3 figures, 10 tables)

This paper contains 25 sections, 4 equations, 3 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Memory for LLM Agents
  4. External Knowledge Integration for LLMs
  5. Method
  6. Membox Construction: The Topic Loom
  7. Membox Linking: The Trace Weaver
  8. Trace Initialization (if $\mathcal{T}=\varnothing$).
  9. Event-to-Trace Voting.
  10. LLM Batch Verification.
  11. Secondary Trace Initialization.
  12. Retrieval
  13. Experiment
  14. Dataset and Evaluation
  15. Implementation Details
  16. ...and 10 more sections

Figures (3)

  • Figure 1: The Fragmentation-Compensation Paradigm: Disrupting Continuity with Ineffective Embedding-Based Recovery
  • Figure 2: Overview of the Membox architecture — the Topic Loom groups locally continuous dialogue into memboxes with event extraction, while the Trace Weaver links events across memboxes to capture long‑range topic recurrence.
  • Figure 3: Retrieval results with different top-k settings