LightMem: Lightweight and Efficient Memory-Augmented Generation (Oct 2025)

Title: LightMem: Lightweight and Efficient Memory-Augmented Generation (Oct 2025)
Link: http://arxiv.org/abs/2510.18866v1
Date: October 2025

Summary:
Despite their remarkable capabilities, Large Language Models (LLMs) struggle to effectively leverage historical interaction information in dynamic and complex environments. Memory systems enable LLMs to move beyond stateless interactions by introducing persistent information storage, retrieval, and utilization mechanisms. However, existing memory systems often introduce substantial time and computational overhead. To this end, we introduce a new memory system called LightMem, which strikes a balance between the performance and efficiency of memory systems. Inspired by the Atkinson-Shiffrin model of human memory, LightMem organizes memory into three complementary stages. First, cognition-inspired sensory memory rapidly filters irrelevant information through lightweight compression and groups information according to their topics. Next, topic-aware short-term memory consolidates these topic-based groups, organizing and summarizing content for more structured access. Finally, long-term memory with sleep-time update employs an offline procedure that decouples consolidation from on-line inference. Experiments on LONGMEMEVAL with GPT and Qwen backbones show that LightMem outperforms strong baselines in accuracy (up to 10.9% gains) while reducing token usage by up to 117x, API calls by up to 159x, and run-time by over 12×. The code is available at https://github.com/zjunlp/LightMem.

Key Topics:
LLM Memory Systems
Efficient Memory
Human Memory Model
Sensory Memory
Short-Term Memory
Long-Term Memory
Token Efficiency
API Call Reduction
Runtime Optimization
Memory-Augmented Generation
Dialogue Agents

Chapters:
00:00 - Welcome & Introduction
00:34 - LightMem: Quick Summary
02:01 - Why Existing Memory Fails
04:00 - Human Memory Inspiration
04:35 - Sensory Memory: Filtering
05:31 - Topic Segmentation
06:30 - Short-Term Memory: Efficiency
08:17 - Long-Term Memory: Offline
10:25 - Impressive Performance Results
11:43 - Future Research & Impact

Stock video credits:
Stas Knop - https://www.pexels.com/@stasknop
Pachon in Motion - https://www.pexels.com/@pachon-in-mot...
Colin Jones - https://www.pexels.com/@larchmedia
StefWithAnF - https://www.pexels.com/@stefwithanf-1...
Engin Akyurt - https://www.pexels.com/@enginakyurt
Pixabay - https://www.pexels.com/@pixabay
Oleg Gamulinskii - https://www.pexels.com/@oleg-gamulins...
Charlie Mounsey - https://www.pexels.com/@charlie-mouns...
Trippy Lagoon - https://www.pexels.com/@trippy-lagoon...
Mikhail Nilov - https://www.pexels.com/@mikhail-nilov
Pressmaster - https://www.pexels.com/@pressmaster
Pavel Danilyuk - https://www.pexels.com/@pavel-danilyuk
Yaroslav Shuraev - https://www.pexels.com/@yaroslav-shuraev
KATRIN BOLOVTSOVA - https://www.pexels.com/@ekaterina-bol...
crazy motions - https://www.pexels.com/@crazy-motions...
Dan Cristian Pădureț - https://www.pexels.com/@paduret
Kelly - https://www.pexels.com/@kelly
Anthony 🙂 - https://www.pexels.com/@inspiredimages
cottonbro studio - https://www.pexels.com/@cottonbro
Kindel Media - https://www.pexels.com/@kindelmedia
José Alfredo Munguía Lira - https://www.pexels.com/@rectorretro
@svetjekolem - https://www.pexels.com/@svetjekolem
Danil Shostak - https://www.pexels.com/@danil-shostak...
Silviu Din - https://www.pexels.com/@silviu-din-16...
Soumya - https://www.pexels.com/@soumya-1446957