Stateful agents

A stateful agent is a persistent AI identity with its own memory, model configuration, tools, and message history. You can return to the same agent across sessions, computers, and interfaces instead of starting from scratch each time.

What belongs to an agent

Each agent has:

• A name and description
• A personality and system prompt
• Long-term memory stored in MemFS
• Model and tool configuration
• One or more conversations

Memory is shared across an agent’s conversations. A preference learned in one conversation can therefore improve how the agent works in another.

Switching agents

Use /agents in the CLI to browse and switch agents. Pin an agent with /pin or the favorite control in the desktop app so it is easier to find later.

> /agents
> /pin

Use separate agents when you want independent identities or memory. Use separate conversations when you want parallel threads with the same identity and memory.

Where agents are stored

When you sign in with Letta, your agents are backed up to the cloud and available through the desktop app, CLI, and web app. You can run the same agent on any connected computer.

Agents created without a Letta account are stored only on local disk. They cannot be accessed remotely through the web app and may require manual backups. See Local setup.