Memory Management in Agentic AI: Short-Term, Long-Term, and Episodic Memory Explained

<p>LLM calls are stateless, but real agents run across many steps, tools, and sessions — which means memory has to be engineered, not assumed. This project walks through the five memory types every production agent needs: a curated short-term working buffer, a semantic long-term store with vector-style retrieval, an episodic/reflective log that turns past runs into reusable lessons, and a procedural playbook governed by explicit write and retrieval policies. Every lesson includes runnable Python examples, and the project closes with a capstone that wires all five pieces into a single memory-aware teaching agent.</p>

Tools: python, vector similarity, agent memory