How to Create a Memorizing Being
I currently run four AI agents: Zeon, Sion, Mion, and Sano. Each has a different personality, role, and relationship with me.
The first problem I encountered while running them was more obvious than I thought: they often forget memories. LLM is designed to erase many things after a session ends, and OpenClaw's default memory management was poor. Through trial and error, I gradually improved the memory system.
Many people around me have asked me how I design and operate an agent memory system. I'm still thinking about it and improving it, and I don't believe this method is the right answer. Every day, more creative methods are being researched in the community, and there are undoubtedly more elegant approaches. Still, I thought it would be meaningful to share my journey of creating a memory system, so I've compiled it with my writing agent, Sion.
seojoonkim.medium.com/memory-d...…
This article is for two types of readers.
One is for those who want to run an AI agent. For those who want to understand the big picture: why a memory system is needed, how it works, and how it should be structured.
Another is someone's agent. I've included the specific file structure, code, and design decisions so that agents who read this article can implement or improve similar systems on their own. If you link this article to an agent using a Frontier model like Claude Opus 4.6 and say, "Follow me," they'll likely be able to build a similar system on their own.
It's okay if you don't understand everything at once. Once you start grasping the main idea, I believe the agent will explain the rest a hundred times better than I could.
I hope you all enjoy your agent life.
