🧠 LLM Agent Memory with Letta

A hands-on course exploring the implementation and application of memory systems for Large Language Model (LLM) agents using the Letta framework.

Description

This repository contains a series of Jupyter notebooks that guide you through building increasingly sophisticated memory systems for LLM agents. From implementing basic self-editing memory from scratch to orchestrating multiple agents with shared memory, this course provides a comprehensive exploration of how to enhance LLM capabilities through effective memory management.

Prerequisites

• Python 3.10.12
• OpenAI API key (stored in .env file)
• Tavily API key (for web search functionality in Lab 5)

Features

📚 Structured Learning Path

The course is organized into progressive labs:

• Lab 1: Implementing self-editing memory from scratch

  • Breaking down the LLM context window
  • Adding memory to the context
  • Modifying memory with tools
  • Implementing an agentic loop

• Lab 3: Building Agents with memory

  • Creating a simple agent with memory
  • Understanding agent state
  • Working with core memory
  • Working with archival memory

• Lab 4: Programming Agent Memory

  • Understanding memory blocks
  • Working with ChatMemory
  • Defining custom memory modules
  • Creating agents with custom memory

• Lab 5: Agentic RAG & External Memory

  • Loading data into archival memory
  • Connecting data via tools
  • Creating custom tools
  • Loading tools from Langchain
  • Creating research agents with web search capabilities

• Lab 6: Multi-Agent Orchestration

  • Creating shared memory blocks between agents
  • Orchestrating multiple agents
  • Adding an orchestrator agent
  • Building a recruiting workflow with multiple specialized agents

Setup Guide

1. Clone this repository:

   git clone https://github.com/corticalstack/llm-agent-memory-letta.git
   cd llm-agent-memory-letta

2. Install dependencies using Poetry:

   poetry install

   Or using pip:

   pip install -r requirements.txt

3. Create a .env file with your OpenAI API key:

   OPENAI_API_KEY=your_api_key_here
   

4. For Lab 5, you'll also need a Tavily API key:

   TAVILY_API_KEY=your_tavily_api_key_here
   

5. Launch Jupyter to run the notebooks:

   jupyter notebook

Architecture

The repository is structured around the Letta framework, which provides tools for building LLM agents with different types of memory:

• Core Memory: Persistent information about the human and agent
• Archival Memory: Long-term storage for detailed information
• Recall Memory: Short-term memory for recent conversations
• Custom Memory Blocks: User-defined memory structures for specialized tasks

The labs progressively build on these concepts, introducing more complex memory systems and agent interactions.

Resources

• Letta Documentation