Getting Started

Get up and running with AgentLoop in minutes. This guide covers installation and your first steps with the interactive mode.

Requirements

Node.js >= 22.0.0
npm or yarn
Podman (optional, for container sandboxing)

Installation

Install AgentLoop globally:


npm install -g @trygentic/agentloop

After installation, you can run:


agentloop

Quick Start

Start Interactive Mode

Launch AgentLoop in interactive mode:


agentloop

This opens the interactive TUI where you can communicate with agents using natural language.

Initialize Your Project

In the interactive terminal, initialize AgentLoop for your project:


> /init

This detects your project type and sets up the AgentLoop configuration.

Start Conversing

Simply type your request in natural language—no special syntax required:


> Break down "user authentication" into tasks and add them to the board

AgentLoop will use the Product Manager agent to analyze your request and create tasks on the kanban board.

View Your Tasks

Open the interactive kanban board to see your tasks:


> /orchestrator kanban

Run the Orchestrator

Start processing tasks:


> /orchestrator run

Running Modes

AgentLoop supports three primary running modes:

Interactive Mode

Interactive Mode is the primary way to use AgentLoop. It provides a full TUI experience with conversational interaction.


agentloop

In this mode:

Chat naturally with agents
Use slash commands for system operations
View the kanban board
Monitor agent progress in real-time

See Interactive Mode for a complete guide.

Headless Query

Query Mode sends a single query and exits. Useful for scripts and automation.


agentloop query "Analyze the architecture of this project"

The response is printed to stdout, making it suitable for piping to other commands.

Headless Orchestrator

Orchestrator Mode runs the orchestrator directly without the interactive TUI.


agentloop orchestrator

This processes all pending tasks until completion, useful for CI/CD pipelines or background execution.

Authentication & Model Selection

When you first run AgentLoop, you’ll go through a two-step setup process.

AgentLoop requires an active subscription. On first launch:

Select “Sign in via Web Browser”
Your browser opens to the AgentLoop authentication page
Log in with your account credentials
Return to the terminal — authentication completes automatically

Step 2: Choose Your Model

After signing in, select your AI model. You can start using AgentLoop immediately—no separate LLM provider subscription required.

Tier	Models	Description
Free Models	OpenCode Zen	Start immediately — no additional cost
AgentLoop Subscription	GLM 4.7	Premium model included with your subscription
Your Own API Keys	Claude, GPT-4o, Gemini, etc.	Connect providers you already have access to

Connecting Anthropic (Claude Models)

To use Claude models with your Anthropic Pro/Max subscription:

Select “Connect your own API key” from the model selection
Choose Anthropic as the provider
Select “Claude Code CLI (recommended)”

If not already installed, run:


npm install -g @anthropic-ai/claude-code

Authenticate with Claude:
```
claude
```
Follow the prompts to log in with your Anthropic account

Important: After completing Claude CLI authentication, restart AgentLoop:


agentloop

Your Claude models will then appear in the “Available Models” section.

Which Model Should I Choose?

New users: Start with a Free Model to explore AgentLoop without any additional setup
Subscribers: Use GLM 4.7 for a premium experience included with your subscription
Claude users: Claude Sonnet 4.5 is recommended for most tasks — it’s fast and capable. Use Claude Opus 4.5 for complex reasoning tasks where quality matters more than speed.
Power users: Connect multiple providers (OpenAI, Google, OpenRouter) for flexibility

Auth Commands


# Check authentication status
agentloop auth status
 
# Log out
agentloop auth logout
 
# Switch LLM provider or account
agentloop auth switch

Configuration

AgentLoop stores configuration in TOML files:

Location	Purpose
`./.agentloop/config.toml`	Project-level config
`~/.config/agentloop/`	User-level config, database, and credentials

See Configuration for the complete reference.

Next Steps

Interactive Mode - Master the TUI experience
Commands vs Slash Commands - Understand the two interaction types
Orchestrator - Learn about task scheduling
Agents - Explore the specialized agents

Need help? Join our Discord community for support and discussions.