AI memory systems · Building Long-Term Memory for AI Agents with MEMS Zero and Python

Out of the box, an LLM doesn't know you.
Store facts in a memory layer.
Recall them later.

Without memory	With memory
Forgets	Remembers name, preferences

Do this in your own tools, step by step. Ticks are saved on this device.

Chat without memory.
Add a fact.
Recall it.

Default LLM memory?

None across sessions.

Solution?

A memory layer.

0:00 In this tutorial, I'll show you exactly how you can build AI agents that remember everything from simple facts to entire conversation histories using a straightforward Python approach that anyone can implement. And now in this video, we're going to work with MEMS zero, which is an open source framework that you can use to add memory to your agents. And now I'm not just going to run you through the quick start. I'm really going to do a deep dive and show you how I would implement this into a real world AI application.

0:23 And for that, I've created the following GitHub repository in my AI cookbook. So it's under the MEMS zero folder, how to build long-term memory for agents with me zero. And now, as always, you can find all the resources and the link to the repository in the description for if you want to follow along. So throughout this video, I will be walking you through this example over here. And of course, we'll dive into some code examples as well.

0:46 And now for those of you that are new here, welcome. My name is Dave Abalar. I'm the founder of Data Luminina, and we've been building custom data and AI solutions for our clients for the past 6 years already. So way before the chat GPT boom, I have a bachelor and a master's degree in AI. And my goal really with these videos is to share my knowledge and to help you become a better AI engineer. So let's dive in.

1:10 All right, so let's cover a little bit of context first before diving straight into the code examples. But let's talk about the importance of memory management and where MEMS zero can help and also why it's different from some of the other approaches that are already out there. So if you've been