LightMem-Ego: First-Person Memory for AI Glasses

aigclink · x · 2026-07-20

The ZJU NLP team open-sourced LightMem-Ego, enabling AI glasses to continuously record what the user sees and hears into queryable memory. It connects to Rokid glasses via an Android app, browser frontend, and online backend, streaming camera and microphone data to build a memory bank in real time.

The system splits continuous experience into three layers: Current memory (ongoing scene), Short-term memory (recent events & conversations), Long-term memory (habits, preferences, semantic facts). Queries are dynamically routed to the appropriate memory layer. The post includes GitHub and paper links, demonstrating a deployable egocentric memory solution.

Related event: Zhejiang University Open-Sources LightMem-Ego for AI Glasses(2 posts)→

Original post →

More from Research

Research channel →