Conversations

A conversation is a message thread with an agent. One agent can have many conversations, and every conversation uses the same agent identity and long-term memory.

Use separate conversations to keep tasks independent while preserving what the agent has learned about you and your work.

Start and resume conversations

When you run letta without arguments, the CLI resumes the default conversation with your last-used agent.

Start a new conversation from the CLI:

letta --new

Or start one during an interactive session:

> /new

Use /resume to browse previous conversations, or pass a conversation ID directly:

> /resume
letta --conv conv-abc123

Fork a conversation

Use /fork to branch the current conversation, including its in-context history. The new conversation can take a different direction without changing the original thread.

Use /btw for a side question. It forks the conversation in the background and returns the answer without interrupting your main task.

Conversation context and agent memory

A conversation contains the messages needed for its current thread. As it grows, older messages are compacted so the active context remains within the model’s limit.

Long-term knowledge belongs to the agent’s MemFS memory, not one conversation. Durable lessons saved there remain available across conversations.