Guide
Showcase
Blog
Version
Guide
Showcase
Blog

CLI

Agent TARS comes with a powerful CLI that provides following commands:

Agent TARS CLI

All commands

To view all available CLI commands, run the following command in the project directory:

agent-tars -h

The output is shown below:

Usage:
  $ agent-tars <command> [options]

Commands:
  [start]       Run Agent TARS in interactive UI
  serve         Launch a headless Agent TARS Server
  request       Send a direct request to an LLM provider
  run           Run Agent TARS in silent mode and output results to stdout
  workspace     Manage Agent TARS workspace

Common flags

Agent TARS CLI provides several common flags that can be used with most commands:

FlagDescription
-c, --config <path>Path to configuration file(s) or URL(s), supports multiple values
--port <port>Port to run the server on (default: 8888)
--openOpen the web UI in the default browser on server start
--logLevel <level>Specify the log level (debug, info, warn, error)
--debugEnable debug mode (show tool calls and system events)
--quietReduce startup logging to minimum
--model.provider <provider>LLM provider name
--model.id <model>Model identifier
--model.apiKey <apiKey>Model API key
--model.baseURL <baseURL>Model base URL
--streamEnable streaming mode for LLM responses
--thinkingEnable reasoning mode for compatible models
--toolCallEngine <engine>Tool call engine type (native, prompt_engineering, structured_outputs)
--workspace <path>Path to workspace directory
--browser.control <mode>Browser control mode (mixed, browser-use-only, gui-agent-only)
--planner.enabledEnable planning functionality for complex tasks
-h, --helpDisplay help for command

agent-tars [start]

The default command (or explicitly using agent-tars start) runs Agent TARS with an interactive UI.

Usage: agent-tars [start] [options]

Options:
  # All common flags are supported

You can also run the interactive UI by simply using:

npx agent-tars

This command will start a web server and display a welcome message with the URL to access the UI.

agent-tars serve

The agent-tars serve command launches a headless Agent TARS Server.

Usage: agent-tars serve [options]

Options:
  # All common flags are supported

Currently, this command is similar to the interactive UI mode but intended for headless operation.

agent-tars request

The agent-tars request command allows you to send direct requests to LLM providers.

Usage: agent-tars request [options]

Options:
  --provider <provider>     LLM provider name (required)
  --model <model>           Model name (required)
  --body <body>             Path to request body JSON file or JSON string (required)
  --apiKey [apiKey]         Custom API key
  --baseURL [baseURL]       Custom base URL
  --stream                  Enable streaming mode
  --thinking                Enable reasoning mode
  --format [format]         Output format: "raw" (default) or "semantic"

Example:

# Using a JSON file for request body
npx agent-tars request --provider openai --model gpt-4 --body ./request.json

# Using a JSON string for request body
npx agent-tars request --provider anthropic --model claude-3-sonnet --body '{"messages":[{"role":"user","content":"Hello"}]}' --format semantic

agent-tars run

The agent-tars run command executes Agent TARS in silent mode and outputs results to stdout.

Usage: agent-tars run [options]

Options:
  --input [...query]        Input query to process (required)
  --format [format]         Output format: "json" or "text" (default: "text")
  --include-logs            Include captured logs in the output (for debugging)
  # All common flags are also supported

Example:

# Simple text query
npx agent-tars run --input "What is Agent TARS?"

# Multiple word query with JSON output
npx agent-tars run --input What is the weather today in New York? --format json

# With custom model
npx agent-tars run --input "Explain quantum computing" --model.provider openai --model.id gpt-4

agent-tars workspace

The agent-tars workspace command helps you manage Agent TARS workspaces.

Usage: agent-tars workspace [options]

Options:
  --init                    Initialize a new workspace
  --open                    Open workspace in VSCode

Create a global workspace

The --init option creates a new Agent TARS workspace with configuration files:

npx agent-tars workspace --init

During initialization, you'll be prompted to select:

  • Configuration format (TypeScript, JSON, or YAML)
  • Default model provider
  • Whether to initialize a git repository

Opening global workspace

The --open option opens your workspace in VSCode if available:

npx agent-tars workspace --open

This command will open the global workspace located at ~/.agent-tars-workspace.

Configuration files

Agent TARS automatically looks for these configuration files in the current directory:

  • agent-tars.config.ts
  • agent-tars.config.yml / agent-tars.config.yaml
  • agent-tars.config.json
  • agent-tars.config.js

You can specify a custom configuration file using the --config flag:

npx agent-tars --config ./my-config.json

Multiple configuration files can be specified and will be merged sequentially:

npx agent-tars --config ./base-config.yml --config ./override.json

Remote configuration URLs are also supported:

npx agent-tars --config https://example.com/config.json

Environment variables

Agent TARS supports environment variables for sensitive information like API keys. When you specify an environment variable name (in all caps) as a value, Agent TARS will use the value of that environment variable:

npx agent-tars --model.provider openai --model.apiKey OPENAI_API_KEY --model.baseURL OPENAI_BASE_URL

In this example, Agent TARS will use the values of the OPENAI_API_KEY and OPENAI_BASE_URL environment variables.