AtomMem: Building Simple and Effective Memory System for LLM Agents via Atomic Facts

2026-06-18

AI Summary

Large language models are constrained by fixed context windows, hindering their ability to effectively accumulate and reuse long-term information across multi-turn dialogues. Existing memory systems often suffer from coarse representations and unstable updates. To address these limitations, this work proposes a long-term memory system tailored for LLM-based agents, which employs a Fact Executor to extract high-value atomic facts as memory units, constructs hierarchical event structures and temporal user profiles, and leverages an associative memory graph for dynamic cross-session retrieval and evolution. Evaluated on the LoCoMo benchmark, the proposed method achieves state-of-the-art performance, substantially enhancing multi-turn reasoning capabilities and offering an efficient, stable, and scalable solution for personalized agent memory.

Abstract

Large language models (LLMs) demonstrate strong reasoning and generation abilities, but their fixed context windows limit long-term information accumulation and reuse across multi-session interactions. Existing memory-augmented systems often construct memory in a coarse and unstable manner, relying on inefficient memory representations or unstable unconstrained updates. To address these challenges, we propose AtomMem, a long-term memory system designed for value-dense storage and stable memory evolution. AtomMem introduces a Fact Executor, which selectively extracts high value atomic facts from long form interactions to serve as highly efficient memory representations. Subsequently, AtomMem organizes these facts into hierarchical event structures and temporal profiles, capturing coherent episodic contexts and tracking dynamically evolving user attributes over time. During retrieval, the system activates an associative memory graph to connect fragmented memories. Experiments on the LoCoMo benchmark confirm that AtomMem achieves state-of-the-art performance across various reasoning tasks, offering a scalable and economically viable solution for deploying intelligent personalized agents.

Problem

• long-term memory
• large language models
• memory representation
• multi-session interaction
• context window limitation

Innovation

• Atomic Facts
• Memory System
• Fact Executor
• Associative Memory Graph
• Hierarchical Event Structure