Agent Memory Techniques
Agent Memory Techniques is a collection of 30 runnable Jupyter notebooks that teach every major memory technique for LLM-based agents, from simple conversation buffers to production-grade frameworks like Mem0 and Letta. It solves the problem of agents forgetting context across sessions by providing hands-on, code-first tutorials. It is for developers building AI agents who need to implement or choose the right memory approach.
✨ Key features
- 30 runnable Jupyter notebooks covering all memory techniques
- Six families: short-term, long-term, cognitive architectures, retrieval, frameworks, evaluation
- Hands-on code for each technique, ready to use
- Includes production frameworks: Mem0, Letta, Zep, Graphiti
- Learning paths and a decision tree to guide selection
- Comparison matrix for side-by-side technique evaluation
🎯 Use cases
- Implement conversation buffer or sliding window memory in a chatbot
- Build long-term memory with vector stores or knowledge graphs for personalized assistants
- Design cognitive architectures with hierarchical or reflective memory for complex agents
- Evaluate and benchmark memory systems for production deployment
- Integrate memory frameworks like Mem0 or Letta into an agent
🚀 Usage
Clone the repository and open any notebook, e.g., all_techniques/01_conversation_buffer_memory/conversation_buffer_memory.ipynb, in Jupyter. Each notebook is self-contained and runnable.
❓ FAQ
What is agent memory?
Agent memory is the set of techniques that let an LLM-based agent remember information across turns, sessions, and tasks. Without it, an agent re-derives context every time and cannot personalize or maintain coherence.
How many techniques are covered?
The repository covers 30 distinct memory techniques, grouped into six families: short-term context management, long-term storage, cognitive architectures, retrieval and multi-agent patterns, batteries-included frameworks, and production deployment patterns.
Which technique should I start with?
If you are new, start with 01 Conversation Buffer Memory, as almost every other technique builds on it. Alternatively, use the learning paths or decision tree in the README to pick based on your goal.
Are the notebooks runnable?
Yes, the README states that there are 30 runnable Jupyter notebooks, each with real code you can use today.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.