Module 1 · How agents remember · Lesson 1
Memory types for agents
Working, episodic, semantic and procedural memory, what each one is in a real agent system, and how to decide where each fact about a 12-week case should live.
By the end of this lesson you can
- Define working, episodic, semantic and procedural memory in the CoALA sense and name a concrete mechanism for each
- Classify the facts a long-running case produces by memory type, owner and lifetime
- Explain why the context window is working memory, not long-term memory, and what that implies for multi-week cases
- Compare full-history and memory-based designs by tokens per turn, and explain why cost is not the main argument