The Complete OpenClaw Guide: From First Install to Power User

What Is OpenClaw, and Why Should You Care?

OpenClaw is an open-source AI agent that lives on your own machine — your laptop, a home server, or a cloud VPS. Unlike browser-based AI tools that only answer questions, OpenClaw takes action: it reads your email, schedules meetings, organizes files, sends messages across WhatsApp, Telegram, and Discord, and automates repetitive workflows — all through natural language.

Since launching in January 2026, the project has crossed 244,000 GitHub stars, making it one of the fastest-growing open-source repositories ever. The reason is simple: people want an AI assistant that actually does things, not one that just talks about doing things.

Here's a quick way to understand the difference:

Traditional AI chatbot:

You: "Help me organize my desktop files." AI: "Sure! Here are some tips: 1. Create folders by type... 2. Move files accordingly..."

OpenClaw:

You: "Help me organize my desktop files." OpenClaw: renames, sorts, and moves 47 files into 5 folders — "Done! I organized everything by file type."

That's the core shift: from a chatbot to an agent.

This guide walks you through the entire journey — from your first install to writing custom Skills and managing multiple agents. Whether you have 30 minutes or an entire afternoon, work through each section at your own pace.

Getting Started: Installation & Setup

What You Need

• Node.js 22+ (LTS recommended) — check with node --version
• An AI model API key — Anthropic Claude (recommended), OpenAI, or a local model via Ollama
• macOS, Linux, or Windows with WSL2

Install OpenClaw

The fastest path is the one-line installer:

curl -fsSL https://openclaw.ai/install.sh | bash

This detects your system, installs dependencies if needed, and launches the setup wizard. If you prefer npm:

npm install -g openclaw@latest

For Docker users:

docker pull openclaw/openclaw:latest
docker run -d --name openclaw \
  -v ~/.openclaw:/root/.openclaw \
  openclaw/openclaw:latest

For a detailed walkthrough of each method, see our Quickstart guide.

Run the Onboarding Wizard

After installation, the wizard walks you through everything:

openclaw onboard

You'll configure three things:

1. AI provider & API key — Anthropic Claude works best. If you don't have a key yet, grab one at anthropic.com. Budget-conscious? See our model selection & cost guide.
2. Messaging channel — Pick at least one: Telegram is the easiest to start with (5-minute setup). We have dedicated guides for Telegram, WhatsApp, and Discord.
3. Security mode — Choose Sandbox for now. You can upgrade to full access later once you're comfortable.

Verify Everything Works

openclaw status

You should see:

Gateway: Running ✓
Model: Connected ✓
Channels: 1 active

Now open your chosen messaging app and send your first message. Try something simple:

• "What can you do?"
• "What time is it?"
• "Summarize the latest tech news."

If your bot responds, congratulations — you have a working AI agent.

Core Concepts: How OpenClaw Works

Before diving deeper, it helps to understand the four building blocks of OpenClaw. Think of them like this:

Gateway — The Front Door

The Gateway is how OpenClaw connects to the outside world. It handles messages coming from Telegram, Discord, WhatsApp, a web API, or even the command line. You can run multiple gateways simultaneously — for example, Telegram for personal use and Slack for work.

Skills — The Toolbox

Skills are what OpenClaw can do. Each Skill is a packaged capability: sending emails, searching the web, managing files, controlling your browser. Think of them like apps on a smartphone. OpenClaw ships with a few built-in Skills, but the real power comes from the community ecosystem on Clawhub and from writing your own.

Browse our Skills directory to see what's available.

Memory — The Brain

OpenClaw remembers things across conversations. Tell it your name, your work schedule, your preferences — and it retains that context. Unlike stateless chatbots, OpenClaw builds a persistent understanding of who you are and how you work, stored locally in ~/.openclaw/memory/.

Sandbox — The Safety Net

The Sandbox controls what OpenClaw is allowed to do on your system. In sandbox mode, it can't access arbitrary files or execute destructive commands. This is your safety layer — especially important when experimenting with new Skills or giving OpenClaw broader permissions.

For a detailed security setup, see our secure configuration template.

Installing and Using Skills

Skills are where OpenClaw goes from "interesting demo" to "daily driver." Here's how to get started.

Browse & Install

# Search for Skills by keyword
openclaw skills search email

# View details about a specific Skill
openclaw skills info @openclaw/email-manager

# Install it
openclaw skills install @openclaw/email-manager

Five Skills Worth Installing First

| Skill | What It Does | Install Command | |-------|-------------|-----------------| | Email Manager | Read, draft, and send emails | openclaw skills install @openclaw/email-manager | | Calendar | Create events, check schedule, send invites | openclaw skills install @openclaw/calendar | | File Organizer | Sort, rename, and clean up files | openclaw skills install @openclaw/file-organizer | | Web Search | Search the web and summarize results | openclaw skills install @openclaw/tavily-search | | Browser Control | Automate browser tasks | openclaw plugins install @openclaw/browser-control |

After installing, just talk to your agent naturally:

• "Check my unread emails and summarize the top 3."
• "What's on my calendar tomorrow?"
• "Search for the latest OpenClaw release notes."

Managing Your Skills

# List all installed Skills
openclaw skills list

# Update all Skills
openclaw skills update

# Remove a Skill you no longer need
openclaw skills remove @openclaw/file-organizer

Before installing third-party Skills, review our Skills security checklist to stay safe.

Setting Up Automations (Cron Jobs)

One of OpenClaw's most practical features is scheduled automations. Instead of asking your agent to do something every morning, you can set it up once and let it run.

Create a Daily Briefing

Just tell your agent in plain language:

"Every morning at 8 AM, send me a briefing with today's weather, my calendar events, and a count of unread emails."

OpenClaw will create a cron job automatically. You can also manage them manually:

# List all scheduled tasks
openclaw cron list

# View details of a specific task
openclaw cron show <task-id>

# Pause a task
openclaw cron disable <task-id>

# Resume it
openclaw cron enable <task-id>

# Delete it permanently
openclaw cron delete <task-id>

Automation Ideas

• Daily briefing — Weather + calendar + unread emails every morning
• Weekly summary — Friday afternoon report of your week's activity
• Price tracker — Monitor a product page and alert you when the price drops
• End-of-day reminder — A friendly nudge at 6 PM to wrap up work
• Inbox cleanup — Auto-archive low-priority emails every evening

Configuring Persistent Memory

OpenClaw's memory system is what makes it feel personal. The more context you give it, the better it serves you.

Teach Your Agent About You

Send a message like this:

"Remember this about me: My name is Alex. I'm a product manager. I work Monday through Friday, 9 AM to 6 PM, timezone GMT+8. I prefer concise responses. I communicate in English but can read Chinese."

OpenClaw stores this in its long-term memory and applies it to every future interaction.

View and Edit Memory

# See what your agent remembers
cat ~/.openclaw/memory/long-term.json

# Or ask directly in chat
"What do you know about me?"

You can also correct or remove memories:

"Forget my old work schedule. I now work 10 AM to 7 PM."

How Memory Works

OpenClaw uses three layers of memory:

• Short-term — Current conversation context (cleared after each session)
• Long-term — Persistent facts about you (preferences, habits, biographical info)
• Episodic — Logs of past interactions (used for pattern recognition over time)

This layered approach means your agent gets more useful the longer you use it, without needing constant reminders.

Writing Your First Custom Skill

When the built-in and community Skills don't cover your specific need, you can write your own. OpenClaw Skills use a simple YAML format.

Skill Structure

Create a new file at ~/.openclaw/skills/daily-news.yaml:

name: "Daily Tech News"
description: "Fetches and summarizes today's top tech headlines"
version: "1.0.0"

triggers:
  - "tech news"
  - "today's news"
  - "news summary"

steps:
  - action: web_search
    query: "top tech news today"
    max_results: 5

  - action: summarize
    content: "{{search_results}}"
    style: "bullet_points"

  - action: respond
    message: |
      Here's your tech news summary:

      {{summary}}

Test It

# Reload Skills to pick up the new file
openclaw skills reload

# Test it
openclaw skills test "Daily Tech News"

Then try it in conversation: "Give me today's tech news."

Building More Complex Skills

Skills can chain multiple actions together:

• web_search — Search the web
• summarize — Condense text into bullet points or paragraphs
• respond — Send a formatted reply
• file_read / file_write — Read or create files
• shell_exec — Run shell commands (requires appropriate permissions)
• http_request — Call external APIs

You can combine these to build workflows like: search for flight prices → compare results → write a summary to a file → notify you on Telegram.

Multi-Agent Management

As your usage grows, you may want separate agents for different parts of your life — one for work, one for personal tasks, one for experiments.

Create Agents

# Create a work agent with Claude Sonnet
openclaw create-agent work
openclaw config --agent work set ai.model "claude-sonnet-4-6"

# Create a personal agent
openclaw create-agent personal
openclaw config --agent personal set ai.model "claude-haiku-4-5"

# List all agents
openclaw list-agents

# Switch between them
openclaw switch-agent work

Why Use Multiple Agents?

| Agent | Model | Connected To | Purpose | |-------|-------|-------------|---------| | Work | Claude Sonnet | Slack, Gmail, Linear | Professional tasks, email triage | | Personal | Claude Haiku | Telegram, Google Calendar | Scheduling, reminders, personal errands | | Lab | Ollama (local) | CLI only | Testing new Skills, experiments |

Each agent has its own memory, Skills, and channel connections — completely isolated from each other.

Browser Automation

OpenClaw can control a browser to perform tasks that go beyond simple API calls: filling forms, extracting data from web pages, comparing prices across sites, or taking screenshots.

Enable Browser Control

# Install the browser plugin
openclaw plugins install @openclaw/browser-control

# Enable it
openclaw config set browser.enabled true

# Show the browser window (set to true for debugging)
openclaw config set browser.headless false

Use Cases

• "Open GitHub and find the top trending repositories today."
• "Go to Amazon, search for mechanical keyboards, and list the top 3 by rating with prices."
• "Fill out the contact form on example.com with my information."
• "Take a screenshot of my portfolio website."

Browser automation is powerful but resource-intensive. Use it for tasks that genuinely need visual interaction with web pages — for simple data fetching, a web search Skill is faster and cheaper.

Integrations Beyond Chat

OpenClaw connects to more than just messaging apps. Here are some popular integrations:

Google Workspace

openclaw config set integrations.google.enabled true
openclaw config set integrations.google.credentialsPath "/path/to/credentials.json"
openclaw integrations google authorize

Once connected, you can manage Gmail, Google Calendar, and Google Drive through natural language.

Project Management (Linear, Jira)

openclaw skills install @openclaw/linear
openclaw config set integrations.linear.apiKey "YOUR_API_KEY"

"Create a high-priority task in Linear: Fix the login page redirect bug."

Note-Taking (Obsidian, Notion)

openclaw skills install @openclaw/obsidian
openclaw config set integrations.obsidian.vaultPath "/path/to/vault"

"Create a new note in Obsidian titled 'Meeting Notes — March 2' with today's discussion points."

For a full list of supported platforms, check out our channel integrations directory.

Security & Cost Control

Docker Sandbox

For maximum isolation, run OpenClaw inside a Docker sandbox:

openclaw config set sandbox.mode "docker"
openclaw config set sandbox.docker.image "openclaw/sandbox:latest"
openclaw sandbox test

This restricts file system access, network access, and shell execution to predefined boundaries.

Run a Security Audit

openclaw security audit --deep

This checks for misconfigured permissions, exposed tokens, and other common issues.

For a complete security hardening guide, read Port Exposure & Remote Access.

Control Your Spending

API costs can add up. Here's how to set guardrails:

# Set a daily request limit
openclaw config set ai.dailyLimit 1000

# Set a monthly budget cap (in USD)
openclaw config set ai.monthlyBudget 50

# Check your usage
openclaw stats usage
openclaw stats cost

Typical monthly costs:

| Usage Level | Description | Estimated Cost | |------------|-------------|----------------| | Light | 10-20 conversations/day | $5–10/month | | Moderate | 50-100 conversations + cron jobs | $20–30/month | | Heavy | Automations + browser control | $50–100/month |

Want to go deeper on cost optimization? See our model selection and cost guide.

Troubleshooting Quick Reference

| Symptom | Likely Cause | Fix | |---------|-------------|-----| | EADDRINUSE on startup | Port 18789 already in use | openclaw gateway start --port 18790 | | 401 Unauthorized | Invalid or expired API key | openclaw config set api-key | | Bot doesn't respond | Channel not connected | Check openclaw status and channel config | | Slow responses | Model overloaded or network issues | Try a lighter model or check connection | | Skills not loading | Need to reload after install | openclaw skills reload |

For more detailed troubleshooting, see our common errors guide.

Useful Diagnostic Commands

# Full health check
openclaw doctor

# View logs
openclaw logs

# Debug mode
openclaw config set logging.level "debug"

What's Next?

You've covered a lot of ground. Here's where to go depending on your interests:

• Want more channels? Explore our 50+ channel integration guides — from Slack and Teams to WeChat and Line.
• Interested in security? Read the Skills security checklist and secure configuration guide.
• Want to self-host? Check our guide on remote access and port exposure.
• Ready to contribute? The OpenClaw GitHub repository welcomes pull requests, Skill contributions, and bug reports.

OpenClaw represents a real shift in how we interact with AI — from asking for advice to delegating real tasks. The project is moving fast, the community is growing, and the best time to start building your own AI agent setup is now.