Agent Memory

Agents that remember across conversations

🔑 Key Concepts

Short-term memory — Current conversation. Buffer window (last N messages) or summary (compress older context).
Long-term memory — Facts, preferences, learned patterns stored in a vector DB. Retrieved when relevant.
Episodic memory — Past interactions as episodes. 'Last time user asked about X, we found Y.' Retrieved on similar situations.
Memory tools — Give agents 'save_memory' and 'recall_memory' tools. Let them decide what's worth remembering.

💡 Practice: Try implementing each concept yourself before moving on. Reading about RAG and building RAG are very different things.