Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mnemis: Dual-Route Retrieval on Hierarchical Graphs for Long-Term LLM Memory

Zihao Tang, Xin Yu, Ziyu Xiao, Zengxuan Wen, Zelin Li, Jiaxi Zhou, Hualei Wang, Haohua Wang, Haizhen Huang, Weiwei Deng, Feng Sun, Qi Zhang

TL;DR

The paper addresses the challenge of long-horizon memory for LLMs by proposing Mnemis, a dual-route retrieval framework that fuses System-1 similarity search on a refined base graph with System-2 global selection over a hierarchical graph. The base graph stores granular memory components (Episodes, Entities, Edges, Episodic Edges) while the hierarchical graph enables top-down, layer-wise reasoning through Category Nodes and Category Edges under principled constraints. Through embedding+BM25 retrieval, RRF re-ranking, and a top-down global search, Mnemis achieves state-of-the-art performance on LoCoMo (93.9) and LongMemEval-S (91.6) with GPT-4.1-mini, outperforming RAG, Graph-RAG, and several memory baselines. The work demonstrates that combining complementary retrieval routes yields superior coverage and structural relevance for long-term memory, with practical implications for persistent AI agents and memory-intensive tasks; future work includes multimodal extensions and more flexible global-traversal planning.

Abstract

AI Memory, specifically how models organizes and retrieves historical messages, becomes increasingly valuable to Large Language Models (LLMs), yet existing methods (RAG and Graph-RAG) primarily retrieve memory through similarity-based mechanisms. While efficient, such System-1-style retrieval struggles with scenarios that require global reasoning or comprehensive coverage of all relevant information. In this work, We propose Mnemis, a novel memory framework that integrates System-1 similarity search with a complementary System-2 mechanism, termed Global Selection. Mnemis organizes memory into a base graph for similarity retrieval and a hierarchical graph that enables top-down, deliberate traversal over semantic hierarchies. By combining the complementary strength from both retrieval routes, Mnemis retrieves memory items that are both semantically and structurally relevant. Mnemis achieves state-of-the-art performance across all compared methods on long-term memory benchmarks, scoring 93.9 on LoCoMo and 91.6 on LongMemEval-S using GPT-4.1-mini.

Mnemis: Dual-Route Retrieval on Hierarchical Graphs for Long-Term LLM Memory

TL;DR

The paper addresses the challenge of long-horizon memory for LLMs by proposing Mnemis, a dual-route retrieval framework that fuses System-1 similarity search on a refined base graph with System-2 global selection over a hierarchical graph. The base graph stores granular memory components (Episodes, Entities, Edges, Episodic Edges) while the hierarchical graph enables top-down, layer-wise reasoning through Category Nodes and Category Edges under principled constraints. Through embedding+BM25 retrieval, RRF re-ranking, and a top-down global search, Mnemis achieves state-of-the-art performance on LoCoMo (93.9) and LongMemEval-S (91.6) with GPT-4.1-mini, outperforming RAG, Graph-RAG, and several memory baselines. The work demonstrates that combining complementary retrieval routes yields superior coverage and structural relevance for long-term memory, with practical implications for persistent AI agents and memory-intensive tasks; future work includes multimodal extensions and more flexible global-traversal planning.

Abstract

AI Memory, specifically how models organizes and retrieves historical messages, becomes increasingly valuable to Large Language Models (LLMs), yet existing methods (RAG and Graph-RAG) primarily retrieve memory through similarity-based mechanisms. While efficient, such System-1-style retrieval struggles with scenarios that require global reasoning or comprehensive coverage of all relevant information. In this work, We propose Mnemis, a novel memory framework that integrates System-1 similarity search with a complementary System-2 mechanism, termed Global Selection. Mnemis organizes memory into a base graph for similarity retrieval and a hierarchical graph that enables top-down, deliberate traversal over semantic hierarchies. By combining the complementary strength from both retrieval routes, Mnemis retrieves memory items that are both semantically and structurally relevant. Mnemis achieves state-of-the-art performance across all compared methods on long-term memory benchmarks, scoring 93.9 on LoCoMo and 91.6 on LongMemEval-S using GPT-4.1-mini.
Paper Structure (18 sections, 5 figures, 9 tables)

This paper contains 18 sections, 5 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Mnemis Methodology
  3. Base Graph
  4. Hierarchical Graph
  5. Memory Retrieval Mechanisms
  6. Experiments
  7. Experiment Setups
  8. Experiment Results
  9. Ablation Study
  10. Influence of System-1 and System-2 Routing
  11. Impact of top-$k$
  12. Effect of Backend Models
  13. Related Work
  14. Conclusion
  15. Case Study
  16. ...and 3 more sections

Figures (5)

  • Figure 1: Framework of Mnemis together with the workflow of base graph ingestion, hierarchical graph ingestion and search. Left is a real case from LoCoMo.
  • Figure 2: Mnemis Hierarchical Graph Overview.
  • Figure 3: Mnemis win case on LoCoMo benchmark. While similarity search fixates on the surface-level cause gastritis, Mnemis successfully identifies the underlying root cause, namely that Sam is overweight. This condition leads to the gastritis and motivates him to change his lifestyle.
  • Figure 4: LoCoMo result across different top-$k$ settings.
  • Figure 5: Mnemis win case on LongMemEval-S benchmark, where Mnemis successfully retrieve "Company's Annual Charity Soccer Tournament" from "Sports Events".