Letta (formerly known as MemGPT) is a platform for building stateful agents, positioned as "AI with advanced memory that can learn and self-improve over time." The project is under active development, and its current source code is located at letta-ai/letta-code, which contains the agent harness, interactive terminal UI, App Server, channels, and the runtime used by desktop and Web applications. (The V1 API server in the original letta-ai/letta repository has been retired and placed in an archive branch; existing tags and releases can still be used for reproduction, but active projects should use the current source code.)
Main capabilities and uses: centered on the core concept of "stateful," it gives agents persistent memory, allowing them to retain context and continuously improve over long-term interactions. It is suitable for scenarios that require maintaining agent memory, identity, and conversations across sessions and devices.
How to use: Letta can be installed from npm; the interactive terminal UI can be launched; the App Server can be run to support local or self-hosted agents. You can also directly use its ready-made entry points: desktop applications for macOS, Windows, and Linux; chat.letta.com in the browser (including mobile); Slack, Telegram, Discord, and custom channels; the Letta Agent SDK for building agents in TypeScript applications; and Letta Cloud for keeping agent memory, identity, and conversations available across computers.
Documentation and limitations: Current instructions for installation, development, and deployment can be found in the Letta documentation. The materials do not provide specific pricing, billing models, performance metrics, or effectiveness data; this content is not yet available.
