Set up Letta

The Letta desktop app is the quickest way to get started. From one interface, you can create agents, inspect their memory, connect model providers, and choose where their tools run.

Prefer working in a terminal? Follow the Letta Code CLI path instead.

Path 1: Letta desktop app

The desktop app supports both locally stored agents and agents managed through Letta’s Constellation cloud. Where the agent’s state lives and where its tools execute are separate choices.

1. Download and install the app

Download Letta Agent for your platform:

• macOS, Apple silicon
• Windows, x64
• Linux, x64 AppImage

Launch the app after installation.

2. Choose where your agent’s state lives

Choose between Local mode and Constellation when you create an agent:

• Local mode: The source of truth for the agent’s memory, context, and message history lives on the current machine. No Letta account is required. You are responsible for backing up local agent state.
• Constellation: The agent’s state is managed in Letta’s cloud and can be accessed from the desktop app, CLI, chat.letta.com, and supported remote environments. See the Constellation documentation for details.

You can sign in to Constellation and still keep local agents on the same machine. Local agents do not appear in the web app or other Constellation services.

3. Connect a model provider

Letta supports hosted models, connected providers, coding plans, and local inference:

• Constellation agents can use Letta Auto, pay-as-you-go models, and supported providers or coding plans connected to your account. They cannot use model servers running only on your local machine.
• Local agents can use supported providers you connect locally or a local inference server.

To configure a provider:

1. Click Connect model providers in the bottom-left menu of the app.
2. Choose a supported provider, coding plan, or local inference option appropriate for your agent’s storage mode.
3. Follow the prompts, then choose a model from the model selector.

Plans and model options change over time. See the pricing and model guide for current details.

4. Enable remote access (optional)

A remote environment separates where you talk to an agent from where its tools execute. For example, you can chat from your phone while the agent reads files or runs commands on your desktop.

Remote environments require a Constellation agent. They do not work with agents stored only in the Local backend. The desktop app must also remain running on the machine where you want tools to execute.

To make your desktop available as a remote environment:

1. Open settings inside the desktop app.
2. Enable Allow remote access.
3. Open chat.letta.com from another device and select your desktop from the environment picker.
4. Continue the conversation. Shell commands, file access, and other client-side tools now run on the selected desktop.

Permission modes and allow/deny rules determine which actions require confirmation. Review these settings before enabling remote access, especially on a machine that contains sensitive files or credentials. See the remote environments and permissions documentation for details.

Path 2: Letta Code CLI

Install Node.js 22.19 or newer, then install Letta Code:

npm install -g @letta-ai/letta-code

Start Letta Code from the directory where you want the agent to work:

cd /path/to/your/project
letta

On first launch, Letta Code creates a local agent without requiring a login. Use /connect to add a supported model provider, coding plan, or local inference server:

> /connect

Run /login only if you want to use Constellation-managed agents and cloud features. Once your model is connected, run /init to let the agent inspect the current project and bootstrap its memory:

> /init

Quick reference

Command	Action	Example
letta	Start interactive terminal mode	letta
/connect	Connect model providers, plans, or local inference	> /connect
/login	Sign in to Constellation	> /login
/model	Switch the active model	> /model
/init	Bootstrap agent memory for the current project	> /init
exit	Exit the Letta Code CLI	> exit or Ctrl+C

Next step

Now that the app or CLI is installed and a model is connected, create an agent and verify that its memory persists across conversations.

Continue to Meet your agent.