Quick Start

This chapter walks you through the full process — from installing the CLI to having your first successful conversation. Once you’ve finished, you’ll be ready for day-to-day interactive use.

Overview

The complete first-launch flow consists of the following steps:

Install the CLI and verify that the unitarylab command is available.
Run unitarylab and follow the wizard to configure your LLM, log in, and start a session.
Have your first conversation.

If you’ve already completed some of these steps, feel free to jump straight to the relevant section.

Installation

We provide one-click installation scripts for each platform:

macOS / Linux: shell script, or install via Homebrew
Windows: PowerShell script

macOS / Linux (Shell)


curl -fsSL https://assets.unitarylab.com/install.sh | bash

macOS / Linux (Homebrew)


brew install unitarylab/unitarylab-agent/unitarylab

Windows (PowerShell)


irm https://assets.unitarylab.com/install.ps1 | iex

Once installation is complete, run the following in any terminal:


unitarylab --help

If the output includes a list of subcommands such as configure, login, logout, and whoami, unitarylab has been added to your PATH correctly and you can proceed to the next step.

Launch and Initialization

Run:


unitarylab

When you run this command, the CLI automatically inspects your environment. If you are not logged in or have not yet configured an LLM, the wizard will guide you through the following setup in order:

Step A: Log In to Your Account

After the LLM is configured, the wizard prompts you to log in:

The wizard asks for your account (phone number or email) and then your password. If you log in with a phone number, prefix it with an international country code such as +86.
If you already know your account details, you can pass them directly on the command line to skip the interactive prompts (see the Account Guide for details).
For more account-related commands, see the Account Guide.

Step B: Configure the LLM

The wizard first prompts you to configure the LLM service the CLI will call:

Provider: Choose from DashScope, OpenAI, or Custom (any OpenAI-compatible service, such as SiliconFlow or a self-hosted gateway).
- For DashScope or OpenAI, the API Base is preset and does not need to be entered.
- For Custom, you must manually enter the API Base (the root path of the OpenAI-compatible endpoint).
API Key and model validation: Enter the API Key for the chosen provider. Input is not displayed in plain text.
- The system does not validate the key immediately. Instead, it first tries to fetch the model list. If the fetch fails (because the API key is incorrect, or because the provider does not support automatic fetching), the system prompts you to enter the model list manually.
- The API Key and the chosen model are not actually validated until after you finalize the model list and pick a specific model from it.
Model list: Once the API Key has been entered, the CLI displays all currently available models.
- Use ↑ and ↓ to move the cursor.
- Press Space to toggle a model’s selection state, and press Enter to confirm.
- Multi-select is supported; pressing Space again on a selected model deselects it.
- For long lists, type a model name directly to filter the results.
- Previously used models are pre-selected, and the most recently used model is automatically pinned to the top.
- Press Space to select Add custom model ID, then Enter to open the manual-entry page, where you can add a custom model not covered by the list. Follow the on-screen format guidance to enter a valid model ID.
Preparing for model switching: After configuration, if you need to change models in a later session, simply enter the /model command in the interactive interface. The system will display your configured model list for quick switching.

Step C: Start a Session

After configuration and login, the wizard asks the following questions in order:

Conversation mode: Choose from agent (full tool chain, the default), ask (retrieval-based Q&A), or plan (planning-oriented).
Model: Choose from your configured model list. If there is only one model, this step is skipped automatically. If there are multiple, a fuzzy-matching selector opens, and you can type keywords to filter.
Session: Start a new session or continue from one of the eight most recent sessions in history. You can also enter a session_id to specify a particular historical session. When continuing an existing session, the CLI automatically restores the working directory that session was last using.
Working directory: The root directory the agent uses for reading and writing files. If a working directory was restored in the previous step, this step is skipped automatically. If the directory does not exist, the wizard will offer to create it.
Tool approval: Whether to show a confirmation prompt before each call to a non-safe tool.

Once these are answered, you’re in the conversation. The prompt looks like this:


[agent│deepseek-v4-pro] ›

For session and working directory details, see the Session and Working directory Guide. For the meaning of each part of the prompt, see the Interactive Conversation Guide.

Your First Conversation

Type your question or instruction at the prompt and press Enter to send:


[agent│deepseek-v4-pro] › Help me automatically solve a PDE system

In agent mode, when the agent calls a tool (for example, downloading a file to the working directory), it follows the approval policy you set in the wizard. Generated visualization files (SVG, PNG, JPG, PDF, and so on) are automatically opened with the system’s default application; for other file types, only the saved path is printed in the terminal.

You can type /help at any time in a session to see all available commands and their descriptions. See the Interactive Conversation Guide for details. For example:

/mode – Switch the conversation mode (agent, ask, plan).
/model – Switch the model.
/cd <path> – Change the working directory.
/config – Reconfigure the API, and so on.

Changing Your Configuration

During day-to-day use, you can run /config at any time to update your configuration or switch accounts. Run this command in the terminal at any time to reconfigure your provider, API Base, or API Key, and to update your model list. (See the Model Configuration Guide for details.)

Logging In Again

To switch users or after your login session expires, run /login to obtain new access credentials. If you already know your account details, you can pass them directly on the command line to skip the interactive prompts (see the Account Guide for details).

Exiting

You can exit the conversation in any of the following ways:

Type exit, quit, /logout, or :q and press Enter.
Press Ctrl-D.
Press Ctrl-C.

Of these, /logout clears your local login state, so you’ll need to enter your account and password again the next time you log in. The other methods do not require you to log in again as long as your account session is still valid. After you exit, your session information and working directory are saved locally so that you can pick up where you left off the next time you launch the CLI.