Memory is Reconstructed, Not Retrieved: Graph Memory for LLM Agents
Shuo Ji, Yibo Li, Bryan Hooi
Why It Matters
What makes this one worth your time
This approach could improve the efficiency and accuracy of LLM agents in tasks requiring long-term memory reasoning, potentially reducing computational costs.
MRAgent dynamically reconstructs memory access for LLM agents using an associative graph structure.
Summary
The paper introduces MRAgent, a framework that enhances memory retrieval for LLM agents by using an associative memory graph and an active reconstruction mechanism, allowing dynamic adaptation of memory access during reasoning.
Key contributions
- Proposed a Cue-Tag-Content graph structure for memory representation.
- Developed an active reconstruction mechanism for dynamic memory retrieval.
- Demonstrated significant improvements on LoCoMo and LongMemEval benchmarks.
Notable insights
- Integrating LLM reasoning directly into memory access allows for iterative exploration and pruning of retrieval paths.
- Associative tags in a Cue-Tag-Content graph serve as semantic bridges, enhancing memory retrieval.
Possible limitations
- Not stated in the abstract
Abstract
arXiv:2606.06036v1 Announce Type: new Abstract: 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.