Letta

Letta (formerly MemGPT) is a framework for building stateful AI agents with 23k+ GitHub stars — a LangChain alternative focused on persistent memory and self-improving agents.

Key features

Memory management

• Persistent memory across conversations and sessions
• Editable memory blocks for agent state
• Automatic memory consolidation and summarization
• Multi-tier memory (core, archival, recall)
• Memory search and retrieval

Agent capabilities

• Self-improving agents that learn from interactions
• Multi-step reasoning and planning
• Tool use and function calling
• Multi-agent orchestration
• Human-in-the-loop workflows

Developer experience

• Python SDK with type hints
• REST API for language-agnostic access
• Pre-built agent templates
• Integration with OpenAI, Anthropic, and local models
• Web UI for agent management

Production features

• Conversation persistence and history
• Agent versioning and rollback
• Usage analytics and cost tracking
• Role-based access control
• SSO and enterprise authentication

At a glance

License	Apache-2.0
Stack	Python, TypeScript, PostgreSQL
Self-hosted	Yes — Letta OSS
Cloud	letta.com (managed)
Memory	Persistent, multi-tier

Self-hosting

pip install letta

Letta can be embedded in your application or run as a standalone server. The cloud version provides managed infrastructure and team features.