Build a personal coding assistant that remembers your coding style and past projects across sessions.
Create a customer support bot that recalls ticket history and customer preferences to provide better service.
Develop an AI companion that learns about the user's interests and personality over time.
Run a local AI agent in your terminal that maintains context and improves from repeated interactions.
Requires setting up a persistent memory backend (likely database) and configuring API endpoints before agents can demonstrate learning behavior.
Letta (formerly known as MemGPT) is a platform for building AI agents that have persistent memory, meaning they can remember past conversations, learn from interactions, and improve over time rather than starting fresh with every session. Most AI chatbots forget everything the moment a conversation ends; Letta solves that problem by giving agents a memory system that persists across sessions. The way it works is you create an agent through the Letta API or SDK, define what it should know about itself and the people it talks to using "memory blocks" (basically named notes the agent can read and update), and then send messages to it. The agent reads and writes to its memory as it works, so future conversations can build on earlier ones. You would use Letta when building an AI assistant that needs to know who you are across multiple sessions, for example, a personal coding assistant, a customer support bot that remembers ticket history, or an AI companion. It can also run locally in your terminal via a command-line tool. Letta is written in Python and supports both Python and TypeScript/Node.js SDKs. It works with any AI model (model-agnostic) and exposes a REST API for integration into applications.
Generated 2026-05-18 · Model: sonnet-4-6 · Verify against the repo before relying on details.