Letta Agent – Personalized Agent That Remembers and Learns
Letta Agent is an AI agent designed to maintain personalized, long-term memory, allowing it to remember user preferences and past interactions to improve its responses over time.
Background
Letta (formerly MemGPT) is an open-source project that gives large language model (LLM) agents a "virtual memory" system — letting them remember past conversations, learn user preferences over time, and maintain long-term context without hitting the token limits that normally cause AI chatbots to "forget" earlier parts of a session. The core technical insight is treating memory like an operating system's virtual memory: the agent automatically pages relevant information in and out of its limited context window. This matters because standard LLM agents (e.g., raw ChatGPT, many RAG systems) have no persistent memory — each new chat starts from scratch, and long conversations degrade in quality as the model's attention window fills up. Letta is developed by a startup of the same name (formerly backed by the Berkeley AI Research lab) and is used by developers building autonomous AI agents, personal assistants, and applications that require agents to retain user-specific knowledge across sessions.