New course on Letta with DeepLearning.AI
LLMs as Operating Systems: Agent Memory
We are excited to announce the launch of our brand new course in collaboration with Andrew Ng's DeepLearning.AI - LLMs as Operating Systems: Agent Memory.
In this course, we go over the fundamentals of adding memory to LLM agents:
- Why is memory important for LLM agents? Memory is key to personalizing agents - without memory, agents are less engaging, less personalizable, and are unable to perform multi-step reasoning (breaking down a task into many subtasks).
- Why do LLMs not have memory? LLMs are stateless compute units: they hold information in their weights, but they do not remember their previous inputs and outputs (for example prior conversations) unless the prior inputs are explicitly feed back to the LLM. This means that if we want to use LLMs to build powerful agentic systems, we have to add memory outside of the LLM.
- What is the concept of self-editing memory? With self-editing memory, you can give LLM agents the ability to learn over time by making modifications to their persistent state via tool calling.
- How does the concept of memory management (or "context management") in LLM agents resemble memory management in an operating system? In a traditional computer, the operating system (OS) moves data back and forth between an unlimited "virtual memory" and a limited "physical memory". You can think of the role of the context management system for LLM agents as that of an "LLM OS", where the LLM OS must move data back and forth between a "virtual context" (all the data available to the LLM agent) and the "physical context" (the actual context window of the LLM input). This concept of the "LLM OS" context manager was first introduced in the MemGPT research paper.
The course includes around 1.5 hours of material, including video lectures and Python notebooks. In one notebook, you learn to implement a basic self-editing agent (in the style of MemGPT) entire from scratch.
If you ever read the MemGPT research paper or have used the Letta framework, and were curious about how the engineering works under the hood, you can take this course to find out!
To read more about the course and enroll for free, visit the course page on DeepLearning.AI's website.
Traditional LLMs operate in a stateless paradigm—each interaction exists in isolation, with no knowledge carried forward from previous conversations. Agent memory solves this problem.
As AI agents become more sophisticated, understanding how to design and manage their context windows (via context engineering) has become crucial for developers.
Today we're announcing Letta Filesystem, which provides an interface for agents to organize and reference content from documents like PDFs, transcripts, documentation, and more.
We've releasing new client SDKs (support for TypeScript and Python) and upgraded developer documentation
Introducing Agent File (.af): An open file format for serializing stateful agents with persistent memory and behavior.
Introducing the Letta Agent Development Environment (ADE): Agents as Context + Tools
Letta v0.6.4 adds Python 3.13 support and an official TypeScript SDK.
Letta v0.5.2 adds tool rules, which allows you to constrain the behavior of your Letta agents similar to graphs.
Letta v0.5.1 adds support for auto-loading entire external tool libraries into your Letta server.
We're excited to announce the Letta Leaderboard, a comprehensive benchmark suite that evaluates how effectively LLMs manage agentic memory.
Memory blocks offer an elegant abstraction for context window management. By structuring the context into discrete, functional units, we can give LLM agents more consistent, usable memory.
