Letta
Letta AI · 2023
Self-editing memory system for agents that remembers facts, relationships, and context across unlimited conversations. Best for building persistent AI companions and long-running automation.
Choose Letta when your agent needs to remember things. Most agents start each conversation with a blank slate. Letta gives agents a long-term memory they can read, write, and update — like a diary that persists across sessions. This is essential for personal assistants, companion agents, customer service bots, and any automation that needs continuity.
- Persistent memory. Letta stores facts, relationships, and preferences in a structured database. Ask it about a conversation from three months ago and it recalls the context, not just the transcript.
- Self-editing memory. Agents can reflect on their own memory, correct outdated facts, and consolidate redundant entries. The memory system improves over time rather than accumulating noise.
- Multi-agent support. Deploy fleets of agents with shared or isolated memory stores. One agent can delegate to another and retrieve the results later.
- Provider agnostic. Works with OpenAI, Anthropic, local models, and any OpenAI-compatible endpoint. The memory layer is independent of the model layer.
- REST API. Build applications on top of Letta's memory system without managing the infrastructure. The API handles authentication, memory storage, and agent lifecycle.
- Framework, not product. Letta is a building block. You need to write code or use the web UI to create agents. There is no consumer chat interface out of the box.
- Memory management overhead. Unbounded memory growth is a real problem. Letta has pruning and summarization, but you must monitor and tune memory policies for production deployments.
- Local setup required. The hosted version exists but the open-source version is where the value is. Expect Docker, Python, and database configuration.
- Not a coding agent. Letta provides memory and orchestration. It does not edit files, run tests, or integrate with git. Pair it with a coding agent for full workflows.
Developers building persistent AI applications, companion apps, customer service bots, research assistants, and any system where context continuity matters more than single-turn brilliance.
- Open Source (free): Full framework, local deployment, community support.
- Cloud (pay-as-you-go): Managed hosting with auto-scaling, monitoring, and enterprise SLA.
- Enterprise: Custom deployment, dedicated infrastructure, priority support.
Spec sheet
Open Source
Local
2023
Open source
Key features
- Persistent memory across sessions
- Self-editing memory (agents modify their own memory)
- Multi-agent orchestration
- REST API and SDK
- Supports any LLM provider
- Docker deployment
- Web UI for management