First tagged release of Agent Memory Techniques — a hands-on, open-source guide to giving LLM agents memory.
What's inside
30 runnable Jupyter notebooks, each a self-contained technique with explanation and code:
- Foundations — conversation buffer, sliding-window, and summary memory
- Retrieval-based memory — vector stores and semantic search over past interactions
- Structured memory — knowledge graphs, episodic and semantic memory
- Tools & frameworks — Mem0, MemGPT, Letta, Zep, Graphiti
- Evaluation — LoCoMo long-conversation benchmarks
- Production patterns — putting memory into real agents
Get started
Clone the repo, open any all_techniques/<NN>_*/ notebook, and run it. Each technique stands on its own.
Free and open-source. Contributions and discussions welcome.