Letta

Letta is an open-source agent platform and research lab for building stateful AI agents that can learn from experience and retain context across interactions. It provides the Letta Agent SDK, Letta Code, managed Letta Cloud deployment, and a self-hostable App Server for coding agents, personal assistants, AI coworkers, and embedded application agents.

✨ Key Features

• Stateful agent runtime — Build agents whose state persists across conversations instead of treating every request as an isolated prompt.
• Persistent memory — Use core memory blocks and archival or recall memory to keep durable context available over time.
• Dynamic memory blocks — Attach, detach, update, and share structured memory blocks as an agent’s context or permissions change.
• Continual learning workflows — Let agents write and repair memory from experience, then evaluate how well they use that accumulated context.
• Letta Agent SDK — Create and operate agents from Python or TypeScript applications through the official clients and API.
• Letta Code — Run a memory-first coding agent in the desktop app or terminal, with local file and shell capabilities.
• Skills and tools — Extend agents with reusable skills, server-side tools, or client-side tools that execute in the calling environment.
• Schedules and channels — Automate recurring work and connect agents to channels such as Slack, Telegram, or Discord.
• Multi-model and BYOK support — Move agents between supported models and connect provider keys where the selected Letta plan permits it.
• Cloud or self-hosted deployment — Use managed Letta Cloud or run the complete Letta App Server yourself for application-specific control.

💳 Pricing

• Letta Code free access — The official quickstart says Letta Code can be used for free with existing provider API keys and supported coding subscriptions such as ChatGPT/Codex and zAI coding plans.
• Letta Cloud — Managed deployment is available, but the public pages reviewed for this candidate did not expose a stable plan table with current prices and quotas. Treat Cloud pricing as unverified until the live account or pricing flow confirms it.
• Self-hosted App Server — The runtime is open source and can be run by the customer; infrastructure and model-provider costs remain the customer’s responsibility.
• Enterprise and BYOK — Official documentation says BYOK for Letta API is available to enterprise customers; contact Letta for current terms.

Confirm current cloud plans, model charges, quotas, retention, and commercial terms on the official site before subscribing.

⚖️ Alternatives Compared

🎯 Use Cases

• Coding assistants — Give an agent durable project context, skills, and access to local or remote development environments.
• Personal work assistants — Maintain user preferences and work context across recurring conversations and tasks.
• AI coworkers — Connect agents to Slack, Telegram, or Discord for ongoing team workflows.
• Recurring automation — Use schedules to run prompts and operational tasks without manually starting each session.
• Embedded product agents — Add stateful agents to an application through the Python or TypeScript SDK.
• Long-running research — Preserve findings, instructions, and learned procedures across multiple sessions.
• Memory evaluation — Measure memory usage and memory generation in production-style agent workloads.
• Self-hosted deployments — Run the App Server with customer-controlled infrastructure and model providers.

Review agent actions, tool permissions, data retention, provider costs, and generated outputs before using Letta with sensitive or production data.