Letta Review 2026

Stateful AI agent platform from the MemGPT team, providing long-term memory, tools, and a managed runtime for production agents.

Letta is the commercial platform built by the team behind the original MemGPT research and open-source project, which pioneered the idea of giving LLMs a virtual context-window manager so they can recall facts and conversations far beyond a single prompt. The Letta service offers a managed runtime for stateful agents: each agent has persistent memory blocks the model can read and rewrite, a tool registry, a per-agent ID and conversation log, and a REST and SDK API to interact with it. Developers can attach Python tools, connect external data sources, run agents in long-running background loops, and inspect every memory edit in the dashboard. The platform supports popular model providers (OpenAI, Anthropic, Google, OpenRouter, local Ollama) so memory and orchestration are decoupled from the underlying LLM. Letta has been integrating with the Model Context Protocol so MCP tools can be loaded directly into an agent, blending its own tool system with the broader MCP ecosystem. The Free tier is $0 with limited agents, Pro is roughly $20/mo for individuals, Team plans start near $100/mo, and Enterprise is custom. Common use cases include personal assistant agents that remember user preferences across sessions, customer support bots that retain account context, research agents that build a personal knowledge graph, and game NPCs with persistent backstory.