ARTFEED — Contemporary Art Intelligence

H-Mem: Hybrid Memory Mechanism for LLM Agents

ai-technology · 2026-05-18

Researchers propose H-Mem, a novel memory mechanism for LLM-based agents like OpenClaw and Manus. It uses a hybrid structure combining a temporal and semantic tree with a knowledge graph to model memory evolution and improve retrieval. The tree allows short-term memory to evolve into long-term memory, while the knowledge graph captures relationships. This addresses poor performance in memory utilization for question-answering tasks. The work is published on arXiv under ID 2605.15701.

Key facts

  • H-Mem is a memory mechanism for LLM-based agents.
  • It uses a hybrid structure with a temporal and semantic tree and a knowledge graph.
  • The tree enables short-term memory to evolve into long-term memory.
  • The knowledge graph captures relationships between memory data.
  • It aims to improve memory retrieval and utilization for QA tasks.
  • The paper is on arXiv with ID 2605.15701.
  • OpenClaw and Manus are examples of LLM-based agents.
  • Previous works lacked principled memory modeling and retrieval.

Entities

Institutions

  • arXiv

Sources