OpenClaw Setup Guide 2026: From Zero to AI Agent in 15 Minutes

OpenClaw is the most popular open-source autonomous AI agent, with 68K+ GitHub stars and a skills ecosystem of 13,000+ extensions. This guide takes you from a fresh machine to a fully configured, channel-connected AI agent in 15 minutes.

No prior experience with OpenClaw required. If you can use a terminal, you can follow this guide.

Table of Contents

1. Prerequisites
2. Step 1: Install OpenClaw
3. Step 2: Run the Onboarding Wizard
4. Step 3: Configure Your LLM Provider
5. Step 4: Connect Your First Channel
6. Step 5: Install Your First Skills
7. Step 6: Test Your Agent
8. Common Issues and Fixes
9. Next Steps
10. FAQ

Prerequisites {#prerequisites}

Before starting, make sure you have the following:

Required

• Node.js 22 or later: OpenClaw requires Node.js 22+ for its runtime.

  # Check your version
  node --version
  # Should output v22.x.x or higher
  
  # Install via nvm (recommended)
  nvm install 22
  nvm use 22
  
  # Or install via Homebrew (macOS)
  brew install node@22
  

• An LLM API key: You need an API key from at least one provider:

  • Anthropic (Claude) — recommended
  • OpenAI (GPT-4o)
  • Google AI Studio (Gemini)

Optional (but recommended)

• A Telegram bot token: For connecting your first channel (easiest to set up)
• An Inbounter API key: For email and SMS capabilities — sign up at inbounter.com

System Requirements

Step 1: Install OpenClaw {#step-1}

Open your terminal and install OpenClaw globally via npm:

npm install -g openclaw@latest

This installs the openclaw command globally. Verify the installation:

openclaw --version

You should see output like:

openclaw v2.4.1

Alternative Installation Methods

Using yarn:

yarn global add openclaw@latest

Using pnpm:

pnpm add -g openclaw@latest

From source (for contributors):

git clone https://github.com/openclaw/openclaw.git
cd openclaw
npm install
npm link

Step 2: Run the Onboarding Wizard {#step-2}

OpenClaw includes an interactive onboarding wizard that walks you through initial configuration:

openclaw onboard --install-daemon

The --install-daemon flag installs the background daemon that lets OpenClaw run persistently and respond to messages across channels even when you close your terminal.

What the Wizard Does

1. Creates the configuration directory: ~/.openclaw/ with default config files
2. Prompts for your LLM provider: Select Anthropic, OpenAI, Google, or local model
3. Stores your API key: Encrypted in ~/.openclaw/credentials.json
4. Installs the daemon (if flag used): Registers a system service that starts on boot
5. Runs a quick test: Sends a simple prompt to verify your API key works

The wizard output looks something like this:

Welcome to OpenClaw! Let's get you set up.

? Select your LLM provider: Anthropic (Claude)
? Enter your Anthropic API key: sk-ant-••••••••••••
? Install background daemon? Yes

✔ Configuration saved to ~/.openclaw/config.yaml
✔ API key verified — Claude is responding
✔ Daemon installed and started

OpenClaw is ready! Try: openclaw run "What can you do?"

Step 3: Configure Your LLM Provider {#step-3}

If you want to change or fine-tune your LLM settings after onboarding, use the config command:

# View current config
openclaw config show

# Change model
openclaw config set model claude-sonnet-4-20250514

# Switch provider
openclaw config set provider openai
openclaw config set api_key sk-your-openai-key
openclaw config set model gpt-4o

Recommended Models by Use Case

Using Local Models

For local models, install Ollama first:

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull a model
ollama pull llama3.3

# Configure OpenClaw
openclaw config set provider ollama
openclaw config set model llama3.3

Note: Local models are significantly less capable than cloud models for complex agentic tasks. Use them for simple automation and privacy-sensitive scenarios.

Step 4: Connect Your First Channel {#step-4}

Channels let OpenClaw communicate through external platforms. We recommend starting with Telegram because setup is the simplest.

Option A: Telegram (Recommended First Channel)

Create a Telegram Bot:

1. Open Telegram and search for @BotFather
2. Send /newbot
3. Follow the prompts to name your bot
4. BotFather gives you a token like 7123456789:AAF1x2y3z4...

Connect to OpenClaw:

openclaw channels connect telegram --token 7123456789:AAF1x2y3z4...

Test it:

1. Open your new bot in Telegram
2. Send it a message: "Hello, what can you do?"
3. OpenClaw should respond within a few seconds

Option B: Slack

# Requires a Slack Bot Token (starts with xoxb-)
# Create one at https://api.slack.com/apps
openclaw channels connect slack --token xoxb-your-slack-bot-token

Option C: Discord

# Requires a Discord Bot Token
# Create one at https://discord.com/developers/applications
openclaw channels connect discord --token your-discord-bot-token

Verify Connections

openclaw channels list

Output:

Connected channels:
  telegram  @YourBotName        ✔ active
  slack     YourWorkspace       ✔ active

Step 5: Install Your First Skills {#step-5}

Skills extend OpenClaw's capabilities. Out of the box, OpenClaw can chat and reason. Skills let it take action: browse the web, send emails, generate images, and thousands more.

Recommended Starter Skills

1. Web Browsing

openclaw skills install web-browser

Gives OpenClaw the ability to visit URLs, read web pages, click links, fill forms, and extract content. Essential for research and data gathering tasks.

2. Gog (Search)

openclaw skills install gog

Adds web search capability. OpenClaw can search for current information, find documentation, look up error messages, and more.

3. Inbounter (Email + SMS)

openclaw skills install inbounter-email

Connects OpenClaw to Inbounter's Email and SMS API. After installation, configure with your Inbounter API key:

openclaw skills configure inbounter-email --api-key your-inbounter-api-key

This gives OpenClaw the ability to:

• Create email inboxes programmatically
• Send and receive emails
• Manage threaded conversations
• Send and receive SMS messages
• Search messages using natural language (semantic search)
• Hold messages for human review (quarantine)

Why Inbounter for email? Traditional email APIs like Gmail and SendGrid were not designed for AI agents. They lack programmatic inbox creation, have human-centric rate limits, and require complex setup for webhooks and threading. Inbounter is purpose-built for agent workflows. See our article on why Gmail and SendGrid don't work for AI agents for a detailed breakdown.

Alternative: Connect Inbounter via MCP

If you prefer MCP integration over the ClawHub skill:

openclaw config set mcp.servers.inbounter '{"type": "http", "url": "https://mcp.inbounter.com/v1", "headers": {"Authorization": "Bearer your-api-key"}}'

Verify Installed Skills

openclaw skills list

Output:

Installed skills:
  web-browser       v3.1.2    ✔ active
  gog               v2.0.4    ✔ active
  inbounter-email   v1.5.0    ✔ active

Step 6: Test Your Agent {#step-6}

Now that OpenClaw is installed, configured, connected to a channel, and equipped with skills, let us test it end to end.

Test 1: Basic Chat

openclaw run "What's the weather like in San Francisco today?"

If you installed the web-browser and gog skills, OpenClaw should search the web and return current weather information.

Test 2: Web Research

openclaw run "Find the top 3 trending GitHub repositories this week and summarize what each one does"

OpenClaw should use its search and web browsing skills to research and compile the answer.

Test 3: Email (Requires Inbounter)

openclaw run "Create a new inbox called 'test-agent' and send a test email to myself at your@email.com"

OpenClaw should use the Inbounter skill to create an inbox and send the email.

Test 4: Interactive Chat

openclaw chat

This opens an interactive chat session where you can have a back-and-forth conversation:

You: Can you help me draft a professional email to a client about a project delay?

OpenClaw: I'd be happy to help draft that email. Let me ask a few questions:
1. What's the client's name?
2. What project is delayed?
3. How long is the delay?
4. What's the reason for the delay?

You: Client is Acme Corp, the website redesign project, delayed by 2 weeks due to design feedback taking longer than expected.

OpenClaw: Here's a draft:

Subject: Update on Website Redesign Timeline

Dear Acme Corp team,

I wanted to provide you with an update on the website redesign project...
[draft continues]

Shall I send this via email, or would you like to modify it first?

Test 5: Channel Test

Send a message to your Telegram bot (or whichever channel you connected):

Can you search for the latest news about AI agents and give me a 3-bullet summary?

OpenClaw should respond directly in Telegram within a few seconds.

Common Issues and Fixes {#common-issues}

"command not found: openclaw"

Cause: npm global bin directory is not in your PATH.

Fix:

# Find where npm installs global packages
npm config get prefix

# Add to your PATH (add to ~/.zshrc or ~/.bashrc)
export PATH="$(npm config get prefix)/bin:$PATH"

# Reload shell
source ~/.zshrc

"Error: Could not connect to LLM provider"

Cause: Invalid API key or network issue.

Fix:

# Verify your API key
openclaw config show | grep api_key

# Test the key directly
openclaw run "Say hello" --verbose

If the key is correct, check your network connection and firewall settings. Some corporate networks block API endpoints.

"Daemon failed to start"

Cause: Port conflict or permissions issue.

Fix:

# Check if another process is using the daemon port
lsof -i :3458

# Restart the daemon
openclaw daemon stop
openclaw daemon start --verbose

# If permissions issue on Linux
sudo openclaw daemon install

"Skill installation failed"

Cause: Network issue or incompatible Node.js version.

Fix:

# Verify Node.js version
node --version  # Must be 22+

# Clear skill cache and retry
openclaw skills cache clear
openclaw skills install web-browser --verbose

"Telegram bot not responding"

Cause: Daemon not running, or bot token is invalid.

Fix:

# Check daemon status
openclaw daemon status

# Check channel connection
openclaw channels list

# Reconnect with verbose output
openclaw channels disconnect telegram
openclaw channels connect telegram --token YOUR_TOKEN --verbose

"Rate limit exceeded" from LLM Provider

Cause: Too many requests to the LLM API.

Fix: Configure rate limiting in OpenClaw:

openclaw config set rate_limit.requests_per_minute 30
openclaw config set rate_limit.retry_delay_seconds 5

For Anthropic, check your usage dashboard to see your current tier limits.

Next Steps {#next-steps}

Congratulations. You have a working AI agent. Here is where to go next:

1. Explore ClawHub

Browse the ClawHub marketplace for skills that match your use cases. With 13,729+ skills available, there is almost certainly something relevant to your workflow.

openclaw skills search "your use case here"

2. Connect More Channels

Add Slack, Discord, or WhatsApp to give your agent multi-channel reach. See the channels documentation for setup guides for each platform.

3. Set Up Scheduled Tasks

Automate recurring tasks:

# Check email every 5 minutes and summarize new messages
openclaw schedule create --cron "*/5 * * * *" --task "Check my Inbounter inbox and summarize any new emails"

# Daily standup summary at 9 AM
openclaw schedule create --cron "0 9 * * 1-5" --task "Compile yesterday's Slack messages from #engineering and create a standup summary"

4. Set Up Email and SMS with Inbounter

If you have not already, sign up for Inbounter and install the skill. Email and SMS are among the most impactful capabilities for an AI agent, enabling customer communication, notifications, and automated follow-ups.

5. Build a Custom Skill

If ClawHub does not have what you need, build your own. The skill development guide walks through the process step by step.

6. Join the Community

• Discord: discord.gg/openclaw — 25K+ members
• GitHub: github.com/openclaw/openclaw — star the repo, report issues, contribute
• Forum: community.openclaw.ai — longer-form discussions and tutorials

Configuration Reference

Here is a quick reference for the most commonly used configuration options:

# ~/.openclaw/config.yaml

# LLM Settings
provider: anthropic          # anthropic, openai, google, ollama
model: claude-sonnet-4-20250514  # Model identifier
api_key: sk-ant-...          # Stored encrypted

# Daemon Settings
daemon:
  enabled: true
  port: 3458
  auto_start: true

# Rate Limiting
rate_limit:
  requests_per_minute: 60
  retry_delay_seconds: 2

# Permissions
permissions:
  mode: ask_sensitive        # always_ask, ask_sensitive, autonomous
  sensitive_actions:
    - send_email
    - delete_file
    - execute_command

# Logging
logging:
  level: info                # debug, info, warn, error
  file: ~/.openclaw/logs/openclaw.log

# MCP Servers
mcp:
  servers:
    inbounter:
      type: http
      url: https://mcp.inbounter.com/v1

Frequently Asked Questions {#faq}

How long does installation take?

About 2-3 minutes for the npm install and onboarding wizard. Adding channels and skills takes another 5-10 minutes. Total: under 15 minutes.

Can I install OpenClaw on a server?

Yes. OpenClaw works on any machine with Node.js 22+. For server installations, use the daemon mode and skip channel connections that require GUI interaction (like WhatsApp QR code scanning). Telegram and Slack work headlessly.

Do I need a paid LLM API plan?

Most providers offer free tiers or credits for new accounts. Anthropic provides $5 in free credits. OpenAI offers free credits for new accounts. Google's Gemini API has a generous free tier. You can get started without spending money.

Can I use multiple LLM providers?

Yes. You can configure a primary provider and fallback providers:

openclaw config set provider anthropic
openclaw config set fallback_providers openai,google

If the primary provider is down or rate-limited, OpenClaw automatically falls back.

Is my data secure?

OpenClaw runs locally. Your API keys are stored encrypted on your machine. Conversations are processed by your chosen LLM provider (subject to their privacy policy) but are not stored by OpenClaw's infrastructure. For maximum privacy, use a local model via Ollama.

How do I update OpenClaw?

npm update -g openclaw

Or for a clean install:

npm install -g openclaw@latest

Can I run multiple OpenClaw instances?

Yes, but each instance needs its own configuration directory and daemon port. Use the --config-dir flag:

openclaw --config-dir ~/.openclaw-work onboard
openclaw --config-dir ~/.openclaw-personal onboard

Where can I find more documentation?

• Official docs: docs.openclaw.ai
• freeCodeCamp guide: Search "OpenClaw tutorial freeCodeCamp" for community-written tutorials
• DigitalOcean tutorial: Covers server-based deployment on DigitalOcean droplets

Related Articles

• What is OpenClaw? The Complete Beginner's Guide
• Why Gmail and SendGrid Don't Work for AI Agents
• Email Webhooks for AI Agents: Complete Developer Guide
• Getting Started with Inbounter