arxiv:2606.06036

Memory is Reconstructed, Not Retrieved: Graph Memory for LLM Agents

Published on Jun 4

· Submitted by

Yibo Li on Jun 15

#2 Paper of the day

National University of Singapore

Upvote

Authors:

Shuo Ji ,

Yibo Li ,

Abstract

MRAgent combines associative memory graphs with active reconstruction to enable dynamic memory access during reasoning, improving long-horizon memory reasoning while reducing computational costs.

Generated by Qwen/Qwen2.5-Coder-32B-Instruct

Despite recent progress, LLM agents still struggle with reasoning over long interaction histories. While current memory-augmented agents rely on a static retrieve-then-reason paradigm, this rigid pipeline design prevents them from dynamically adapting memory access to intermediate evidence discovered during inference. To bridge this gap, we propose MRAgent, a framework that combines an associative memory graph with an active reconstruction mechanism. We represent memory as a Cue-Tag-Content graph, where associative tags serve as semantic bridges connecting fine-grained cues to memory contents. Operating on this structure, our active reconstruction mechanism integrates LLM reasoning directly into memory access, allowing the agent to iteratively explore and prune retrieval paths based on accumulated evidence. This ensures that memory retrieval is dynamically adapted to the reasoning context while avoiding combinatorial explosion caused by unconstrained expansion. Experiments on the LoCoMo benchmark and LongMemEval benchmark demonstrate significant improvements over strong baselines (up to 23%), while substantially reducing token and runtime cost, highlighting the effectiveness of active and associative reconstruction for long-horizon memory reasoning.

View arXiv page View PDF GitHub 75 Add to collection

Community

liushiliushi

Paper author Paper submitter 3 days ago

avahal

about 24 hours ago

the most interesting part is treating memory as a cue-tag-content graph and weaving active reconstruction into the reasoning loop rather than a static retrieve step. by using forward and reverse traversal to expand or prune candidate memories and a lightweight toolkit for tool-based queries, they keep search cost in check while letting the llm guide what to fetch. i’d like to see how robust the cue/tag extraction is to noisy dialogs or long-tail prompts, since bad tags could mislead the reconstruction. btw, arxivlens had a solid breakdown that helped me parse the method details, especially the reconstruction loop and how evidence accumulates: https://arxivlens.com/PaperView/Details/memory-is-reconstructed-not-retrieved-graph-memory-for-llm-agents-7839-d129fc11