LightMem-Ego: Your AI Memory for Everyday Life

2026-07-13

AI Summary

This work addresses the challenge that existing personal AI assistants on mobile and wearable devices struggle to efficiently accumulate, organize, and retrieve users’ long-term, multimodal daily experiences in response to queries. The paper proposes the first lightweight, streaming multimodal memory architecture tailored for everyday life, which continuously captures experiences through temporally aligned visual and audio streams. It constructs a hierarchical memory structure comprising current, short-term, and long-term components, and introduces a dynamic retrieval routing mechanism that adaptively activates the relevant memory tier based on query content, fusing multimodal evidence to generate responses. Implemented to run in real time on resource-constrained smartphones and AI glasses, the system effectively supports personalized assistance tasks such as object finding, conversation recall, life summarization, and behavioral pattern discovery.

Abstract

Personal AI assistants on mobile and wearable devices continuously perceive users' daily lives through visual and audio streams. However, answering queries about past experiences requires lightweight multimodal memory that can continuously accumulate, organize, and retrieve long-term experiences, which remains challenging. To address this challenge, we present LightMem-Ego, a lightweight streaming multimodal memory system for everyday-life assistance. The system continuously captures egocentric visual and audio streams, aligns them on a shared timeline, and organizes them into a hierarchical memory consisting of current, short-term, and long-term memory. Given a user query, LightMem-Ego dynamically routes retrieval to the appropriate memory level and generates answers grounded in multimodal evidence. The demonstration can be deployed on smartphones and AI glasses, supporting object finding, conversation recall, life summarization, routine discovery, and personalized assistance. Code is available at https://github.com/zjunlp/LightMem-Ego.

Problem

• lightweight multimodal memory
• continuous memory accumulation
• long-term experience retrieval
• egocentric perception
• personal AI assistant

Innovation

• lightweight multimodal memory
• egocentric streaming
• hierarchical memory architecture
• dynamic retrieval routing
• on-device AI assistant