Letta: Stateful AI Agents with Persistent Memory

Letta (formerly MemGPT) is a framework for building stateful AI agents with persistent memory, enabling LLMs to manage their own memory hierarchy — including long-term recall, working context, and external data — for truly autonomous, memory-aware agents.

Features

Self-Editing Memory

Agents autonomously manage their own memory using tool calls, deciding what to remember, forget, and retrieve — implementing a virtual context management system inspired by operating system memory hierarchies.

Tiered Memory Architecture

Multi-tier memory system with core memory (always-available context), recall memory (searchable conversation history), and archival memory (long-term storage) for efficient context management within LLM token limits.

Memory Tools

Built-in memory manipulation tools that agents use to read, write, search, and update their own memory stores, enabling unbounded conversation history and knowledge accumulation.

Agent Server

Production-ready agent server with REST API for deploying, managing, and interacting with stateful agents, supporting multi-user environments and concurrent agent execution.

Multi-Agent Orchestration

Support for multi-agent systems where agents can communicate with each other, share memory contexts, and collaborate on complex tasks with coordinated memory management.

Tool Integration

Extensible tool system allowing agents to interact with external APIs, databases, and services while maintaining memory of past tool interactions and results.

Key Capabilities

Unbounded Context: Manage conversations far beyond LLM context window limits
Persistent State: Agent memory persists across sessions and restarts
Model Agnostic: Works with OpenAI, Anthropic, local models, and more
ADE (Agent Development Environment): Visual interface for building and debugging agents
Composable Memory: Mix and match memory backends (local, cloud, custom)
Letta Cloud: Managed service for deploying stateful agents at scale
Open Source: Core framework available under Apache 2.0

Best For

Developers building AI agents that need to remember across long interactions
Teams creating personal assistants with evolving user knowledge
Researchers exploring memory-augmented language models
Applications requiring agents that manage their own context autonomously
Organizations deploying multi-agent systems with shared memory

Performance Statistics

~21.7k GitHub stars (as Letta/MemGPT)
Pioneered the self-editing memory paradigm for LLM agents
Active research and production community

Access

Open-source framework on GitHub
Letta Cloud for managed deployments
ADE (Agent Development Environment) for visual agent building
Python SDK and REST API

Last built with the static site tool.