Letta

What is Letta?

Letta is an open-source agent development platform built around a memory-first architecture, originating from the MemGPT research project at UC Berkeley. It provides an Agents API that handles persistent state, context management, tool calling, and memory automatically — so developers can focus on building agents that genuinely learn and improve over time rather than resetting with every session. Letta supports self-hosted deployment via Letta Server or cloud usage, and works with any LLM provider. Its flagship consumer product, Letta Code, is a memory-first coding agent accessible from terminal or any device.

Key Features

• Persistent Memory Architecture

  Agents maintain in-context memory blocks, recall memory for interaction history, and archival memory storage across sessions — all queryable and editable via REST API.

• Sleep-Time Compute

  Background memory subagents continuously refine prompts, consolidate knowledge, and improve agent context between active sessions without user intervention.

• Agent Development Environment (ADE)

  No-code visual interface to inspect an agent's memory blocks, reasoning steps, and tool call chains — making debugging and behavior tuning transparent and direct.

• Multi-Agent Orchestration

  Built-in support for agents to message each other synchronously or asynchronously, enabling both supervisor-controlled pipelines and peer-to-peer agent collaboration.

• Extensive Tool Integration

  Supports over 7,000 tools including custom-defined functions, MCP-compatible tools, and pre-built integrations executable in cloud sandboxes.

• Model-Agnostic & Portable

  Memory blocks and conversation history are portable across different LLM providers, preventing vendor lock-in and allowing model swaps without losing agent state.

Use Cases

• Personalized Coding Assistants : Developers can run Letta Code as a persistent coding agent that remembers project context, coding preferences, and past decisions across every session.
• Customer-Facing Service Agents : Enterprises can deploy long-running agents that accumulate knowledge about individual users over time, enabling deeply personalized support experiences.
• Deep Research Agents : Agents can autonomously process large document sets in the background, distill findings into memory, and surface relevant insights on demand — as demonstrated in Letta's collaboration with 11x.
• Recommendation Systems : Stateful agents can track user behavior and preferences over time to power continuously improving recommendation engines, as deployed by BILT Rewards.
• Multi-Agent Pipelines : Engineering teams can compose networks of specialized agents that share state and communicate to handle complex, long-horizon workflows beyond single-agent capability.

FAQs