NemoClaw Knowledge Wiki

AI Agent Memory Types: CoALA Framework Overview

3 min read

🗂️ AI & Agents · View mindmap

Generated: 2026-05-27 · API: Gemini 2.5 Flash · Modes: Summary

AI Agent Memory Types: CoALA Framework Overview

Clip title: The Four Types of Memory Every AI Agent Needs Author / channel: IBM Technology URL: https://www.youtube.com/watch?v=BacJ6sEhqMo

Summary

This video provides a clear overview of the four main types of memory crucial for AI agents, drawing a helpful analogy to human memory. The speaker, Martin Keen from IBM, explains that just as humans rely on short-term memory, factual knowledge, learned skills, and personal experience, well-designed AI agents require similar capabilities. He introduces the CoALA framework (Cognitive Architectures for Language Agents) from a Princeton research team, which maps out these distinct memory types for AI systems, ranging from foundational to more complex, emerging areas.

The first two types discussed are Working Memory and Semantic Memory. Working memory is akin to a computer’s RAM or an AI’s “context window,” holding active, immediate information like the current conversation or system instructions. It’s fast and accessible but volatile and has limited size, much like short-term human memory. Semantic memory, on the other hand, functions as the agent’s persistent knowledge base, storing facts, rules, conventions, and documentation. Often implemented using markdown files or vector databases, this memory ensures the agent doesn’t repeat mistakes by providing foundational, persistent knowledge.

The video then delves into Procedural Memory and Episodic Memory. Procedural memory represents the agent’s learned skills – how to perform tasks, described in structured formats like skill.md files. This utilizes “progressive disclosure,” where the agent only loads detailed instructions for a skill when it’s actively needed, conserving working memory. Lastly, episodic memory is the agent’s record of past interactions, decisions, and lessons learned. Instead of saving every detail, this memory distills or compresses experiences, remembering what was useful for future conversations and acting as a form of genuine learning. The challenge here lies in deciding what information is valuable enough to retain and when to forget obsolete data.

Ultimately, the video concludes by illustrating how different AI agents leverage these memory types to varying degrees. A simple “reflex agent” like a thermostat might only need working memory. A customer support agent capable of resetting passwords would require working and procedural memory. A complex “coding agent” aiming for advanced tasks would benefit from all four types: working memory for immediate context, semantic memory for general knowledge, procedural memory for executing tasks, and episodic memory to learn from past experiences. This comprehensive memory architecture is what truly differentiates a responsive chatbot from a capable, evolving AI agent.

Video Description & Links

Description

Learn more about AI Agents here → https://ibm.biz/~OSlmklt3a

AI agents remember in more than one way. Martin Keen explains the four types of memory AI agents use, from context windows to learned experience. See how working, semantic, procedural, and episodic memory power real agentic systems.

AI news moves fast. Sign up for a monthly newsletter for AI updates from IBM → https://ibm.biz/~RcPTLwx76

aiagents aimemory contextwindow aiarchitecture

Tags

IBM, IBM Cloud

URLs

Graph View

Backlinks