Letta
by Letta AI (formerly MemGPT)
AI agent framework with persistent memory blocks — build agents that remember and evolve across sessions
Features
- Memory blocks — named text segments (human, persona, organization, observations, emotional state, relationships) inserted into the agent’s context window and updated per interaction
- Agent Development Environment (ADE) — interactive UI for building, testing, and inspecting agents and their memory
- Letta Code — coding-focused agent mode; integrates with Slack for digital coworker workflows
- Letta Channels — Slack integration where agents participate as full workspace members (web search, PR review, slide generation, screenshot analysis)
- Persistent deployment — Docker, tmux, or Mac Mini options for 24/7 agent availability; remote access via Letta Chat
- Skill system — add capabilities on demand (web search, code execution, slides, speech synthesis)
- MCP/API compatible — integrates with Claude and other LLMs as the reasoning backbone
Superpowers
Letta’s key insight is that memory architecture IS agent personality. By exposing memory as explicit, editable blocks rather than buried in system prompts, agents can evolve through interactions — accumulating observations, relationship context, and task history across sessions. The ADE makes memory design interactive: you add a “relationships” block, populate it during conversation, and the agent references it naturally in future sessions. This positions Letta as the production-grade solution for teams that need agents with institutional memory, not just stateless assistants.
Pricing
- Open source (self-hosted)
- Letta Cloud — managed hosting (check site for current pricing)