LiCoMemory: Lightweight and Cognitive Agentic Memory for Efficient Long-Term Reasoning

2025-11-03

AI Summary

Large language model (LLM) agents face fundamental limitations due to finite context windows and the absence of persistent memory. Existing external memory architectures predominantly rely on flat graph structures, leading to tight coupling between semantic content and topological relationships—resulting in inefficient, redundant retrieval. To address this, we propose LiCoMemory, a lightweight cognitive agent memory framework featuring a novel hierarchical CogniGraph structure that explicitly decouples semantic indexing from topological connectivity. We further design a time- and hierarchy-aware two-stage retrieval mechanism, augmented by a re-ranking strategy to enhance retrieval accuracy and response coherence. Evaluated on the LoCoMo and LongMemEval benchmarks, LiCoMemory significantly outperforms state-of-the-art baselines across temporal reasoning, cross-session consistency, and retrieval efficiency. Notably, it reduces memory update latency by an order of magnitude.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Large Language Model (LLM) agents exhibit remarkable conversational and reasoning capabilities but remain constrained by limited context windows and the lack of persistent memory. Recent efforts address these limitations via external memory architectures, often employing graph-based representations, yet most adopt flat, entangled structures that intertwine semantics with topology, leading to redundant representations, unstructured retrieval, and degraded efficiency and accuracy. To resolve these issues, we propose LiCoMemory, an end-to-end agentic memory framework for real-time updating and retrieval, which introduces CogniGraph, a lightweight hierarchical graph that utilizes entities and relations as semantic indexing layers, and employs temporal and hierarchy-aware search with integrated reranking for adaptive and coherent knowledge retrieval. Experiments on long-term dialogue benchmarks, LoCoMo and LongMemEval, show that LiCoMemory not only outperforms established baselines in temporal reasoning, multi-session consistency, and retrieval efficiency, but also notably reduces update latency. Our official code and data are available at https://github.com/EverM0re/LiCoMemory.

Problem

• Addresses LLM agents' limited context windows and lack of persistent memory
• Resolves redundant representations and unstructured retrieval in graph-based memory
• Improves efficiency and accuracy in long-term reasoning and knowledge retrieval

Innovation

• Hierarchical graph structure with semantic indexing layers
• Temporal and hierarchy-aware search with reranking
• Lightweight framework for real-time memory updates