What You Will Build

By the end of this guide, you will have a working OpenClaw instance that:

• Responds to you on WhatsApp, Telegram, or your preferred chat app
• Uses the AI model of your choice (Claude, GPT, DeepSeek, or a free local model)
• Automates at least one real workflow — email summarization, file management, or web research
• Runs reliably in Docker with proper security hardening

This is not a feature overview. This is a build-it-yourself walkthrough.

---

OpenClaw: From Zero to a Working Personal AI Agent

OpenClaw is a self-hosted, open-source personal AI assistant that crossed 247,000 GitHub stars in early March 2026. That number matters less than what it represents: hundreds of thousands of developers decided that cloud chatbots are not enough — they want an AI that can actually do things on their behalf.

Created by Austrian developer Peter Steinberger and originally published as Clawdbot in November 2025, the project was renamed twice — first to Moltbot after trademark issues with Anthropic, then to OpenClaw three days later — before settling into the name that stuck. The lobster emoji (🦞) became its mascot, and the tagline "The lobster way" became a rallying cry for the self-hosted AI movement.

What makes OpenClaw different from ChatGPT, Claude, or any cloud chatbot is simple: it runs on your machine, connects to your chat apps, and takes action in the real world. It reads your email, controls your browser, manages files, runs shell commands, and automates multi-step workflows — all triggered by a text message from whatever platform you already use.

Let's build it.

---

Part 1: Installation and First Boot

System Requirements

OpenClaw is lightweight. You need:

• Node.js 24 (recommended) or Node.js 22.16+
• 2GB RAM minimum (8GB+ if running local models alongside)
• macOS, Linux, or Windows (WSL2 recommended on Windows)
• 20GB disk space for the base install and model caches

Source: OpenClaw Documentation

Install OpenClaw

The fastest path is a global npm install:

# Install OpenClaw globally
npm install -g openclaw@latest

# Or if you prefer pnpm
pnpm add -g openclaw@latest

Verify the installation:

openclaw --version
# Expected output: openclaw v0.x.x

Run the Onboarding Wizard

OpenClaw ships with an interactive onboarding command that walks you through connecting your first chat platform and AI provider:

openclaw onboard --install-daemon

The --install-daemon flag installs the Gateway as a background service that starts automatically on boot. This is the core process that stays running and routes messages between your chat apps and the AI agent.

Source: OpenClaw GitHub Repository

The wizard asks three questions:

1. Which chat platform? — Pick one to start (Telegram is recommended for first-timers)
2. Which AI provider? — Enter your API key for Claude, GPT, DeepSeek, or configure Ollama
3. What name for your assistant? — This becomes the display name in your chat app

After onboarding completes, the Gateway starts and your assistant is live.

---

Part 2: Connecting Chat Platforms

OpenClaw supports over 20 messaging platforms — more than any other AI assistant framework. Here is how to connect the most popular ones.

Telegram (Easiest Setup)

Telegram is the most mature OpenClaw integration and the one the community recommends for beginners.

1. Open Telegram and message @BotFather
2. Send /newbot and follow the prompts to create a bot
3. Copy the bot token BotFather gives you
4. Add the token to your OpenClaw config:

# ~/.openclaw/config.yaml
channels:
  telegram:
    enabled: true
    botToken: "YOUR_TELEGRAM_BOT_TOKEN"
    allowedUsers:
      - your_telegram_username

5. Restart the Gateway:

openclaw gateway restart

Message your bot on Telegram — it should respond immediately.

Source: OpenClaw Personal Assistant Setup

WhatsApp

WhatsApp integration uses the WhatsApp Web protocol. It is fully functional but requires a dedicated phone number — do not use your primary WhatsApp account.

openclaw onboard --channel whatsapp

The CLI displays a QR code directly in your terminal. On your phone:

1. Open WhatsApp → Settings → Linked Devices → Link a Device
2. Scan the QR code from the terminal

Your OpenClaw instance is now accessible through WhatsApp. Every message you send to that linked session reaches the agent.

Important: Use a dedicated number for the assistant. OpenClaw will be reading and responding to messages on this account — you want separation between your personal chats and your agent.

Source: OpenClaw WhatsApp Setup Guide

Discord

# ~/.openclaw/config.yaml
channels:
  discord:
    enabled: true
    botToken: "YOUR_DISCORD_BOT_TOKEN"
    allowedGuilds:
      - "your_server_id"

Create a Discord application at discord.com/developers, add a bot, and copy the token. OpenClaw responds in any channel where the bot is mentioned.

Slack

channels:
  slack:
    enabled: true
    appToken: "xapp-..."
    botToken: "xoxb-..."

Slack requires both an app-level token and a bot token. Create a Slack app at api.slack.com/apps with Socket Mode enabled.

Other Supported Platforms

OpenClaw also supports Google Chat, Signal, iMessage (macOS only), Microsoft Teams, Matrix, IRC, LINE, Mattermost, Nextcloud Talk, Feishu, Nostr, Synology Chat, WeChat, and more. Each follows a similar pattern: create a bot/token on the platform, add the credentials to config.yaml, restart the Gateway.

Source: OpenClaw Integrations

---

Part 3: Choosing and Configuring Your AI Model

This is where OpenClaw gets interesting. Unlike locked-in products, you choose which brain powers your assistant — and you can switch models on the fly or set up automatic fallback chains.

Option A: Claude (Anthropic)

Claude is the most popular choice among OpenClaw users for complex reasoning and long conversations.

# ~/.openclaw/config.yaml
providers:
  anthropic:
    apiKey: "${ANTHROPIC_API_KEY}"

agents:
  primary:
    model: "anthropic/claude-sonnet-4-6"
    contextTokens: 200000
    thinkingEnabled: true

Set your API key:

export ANTHROPIC_API_KEY="sk-ant-..."

Source: OpenClaw Model Configuration Guide

Option B: GPT (OpenAI)

providers:
  openai:
    apiKey: "${OPENAI_API_KEY}"
    baseUrl: "https://api.openai.com/v1"

agents:
  primary:
    model: "openai/gpt-4.1"

Option C: DeepSeek (Budget-Friendly Cloud)

DeepSeek offers strong performance at a fraction of Claude or GPT pricing — a popular choice for high-volume automations.

providers:
  openai-compatible:
    apiKey: "${DEEPSEEK_API_KEY}"
    baseUrl: "https://api.deepseek.com/v1"

agents:
  primary:
    model: "openai-compatible/deepseek-chat"

Source: OpenClaw LLM Setup Guide

Option D: Ollama (Free, Fully Local, Fully Private)

This is the zero-cost option. Ollama runs open-source models directly on your machine — no API keys, no internet connection, no data leaving your device.

First, install Ollama and pull a model:

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

# Pull a model (Llama 3 8B is a good starting point)
ollama pull llama3:8b

# Or for stronger reasoning
ollama pull deepseek-r1:14b

Then configure OpenClaw:

providers:
  ollama:
    baseUrl: "http://127.0.0.1:11434/v1"
    apiKey: "ollama-local"
    api: "ollama"

agents:
  primary:
    model: "ollama/llama3:8b"
    contextWindow: 8192
    maxTokens: 4096

Note: The baseUrl must include the /v1 suffix — this is the most common configuration mistake new users encounter.

Source: Using OpenClaw with Ollama — DataCamp

Multi-Model Fallback Chain

One of OpenClaw's most powerful features is the ability to define a fallback chain. The agent tries the primary model first, and if it fails (rate limit, timeout, outage), it automatically falls to the next model:

agents:
  primary:
    model: "openai-compatible/deepseek-chat"
    fallback:
      - model: "anthropic/claude-sonnet-4-6"
      - model: "ollama/llama3:8b"

This configuration uses DeepSeek for most interactions (cheapest), falls back to Claude for complex tasks, and uses the local Ollama model when both cloud providers are unavailable. You get cost optimization, reliability, and offline capability in a single config.

Source: OpenClaw API & Model Configuration

---

Part 4: What OpenClaw Can Actually Do

Knowing that OpenClaw is an "AI agent" is abstract. Here is what it concretely does, with examples you can try today.

Browser Control

OpenClaw can open a browser, navigate pages, fill forms, click buttons, extract data, and take screenshots — all from a chat message.

You (via Telegram): "Go to Amazon and find the best-rated mechanical keyboard under $100"

OpenClaw opens a headless browser, navigates to Amazon, searches, filters by rating and price, and returns a formatted list with links. It uses Playwright under the hood for reliable browser automation.

The more powerful variant is Live Browser Control, which connects to your existing Chrome session — with your logged-in accounts, cookies, and tabs intact. This means OpenClaw can interact with authenticated services like your email, banking dashboards, or internal tools without needing separate credentials.

Source: OpenClaw Live Browser Control — Goldie Agency

Email Management

One of the highest-impact automations. Connect OpenClaw to your Gmail or Outlook and it can:

• Summarize your unread inbox every morning and send the briefing to Telegram
• Draft replies based on conversation context
• Archive, label, or flag messages according to rules you define
• Identify calendar-related emails and automatically manage scheduling

You (via WhatsApp): "Summarize my inbox and flag anything urgent"

Source: OpenClaw Use Cases — TLDL

File and System Operations

OpenClaw can read, write, move, and delete files. It can run shell commands. It can execute code in a sandboxed environment.

You (via Slack): "Find all PDF invoices from this month in my Downloads folder, rename them with the vendor name and date, and move them to ~/Documents/Invoices/2026-03/"

The agent reads each PDF, extracts the vendor name and date, renames the files accordingly, and moves them. This kind of multi-step file operation is where OpenClaw saves real time.

Calendar and Scheduling

OpenClaw monitors your calendar, handles scheduling conflicts, and manages meeting logistics:

You (via Telegram): "When someone emails about rescheduling a meeting, check my availability, update the event, and send them a confirmation"

This is not a hypothetical — it is one of the most commonly deployed OpenClaw automations.

Content and Social Media

The most widely adopted use case category. OpenClaw users connect blog RSS feeds and have the agent automatically generate platform-specific posts for X, LinkedIn, and newsletters. One user reported saving 10+ hours per week on social media content alone.

Research and Competitive Intelligence

Set up weekly competitor monitoring that scrapes websites for product changes, pricing updates, and news, with OpenClaw formatting everything into structured reports delivered to your preferred channel.

Source: Advanced OpenClaw Workflows — LightNode

---

Part 5: Building Custom Skills

Skills are OpenClaw's extension mechanism — Markdown files that teach your agent new capabilities. The ClawHub registry contains thousands of community-contributed skills, and building your own takes minutes.

How Skills Work

Each skill is a directory containing a skill.md file with YAML frontmatter (declaring metadata, dependencies, and required tools) and natural language instructions that tell the agent what to do and how to do it.

Source: OpenClaw Skills Documentation

Installing Community Skills

# Browse available skills
openclaw skills search "email"

# Install a skill
openclaw skills install email-summarizer

# List installed skills
openclaw skills list

The awesome-openclaw-skills repository catalogs over 5,400 skills filtered and categorized from the official registry.

Creating a Custom Skill

Here is a minimal skill that monitors Hacker News for topics you care about:

mkdir -p ~/.openclaw/skills/hn-monitor

Create ~/.openclaw/skills/hn-monitor/skill.md:

---
name: hn-monitor
description: Monitors Hacker News for specified topics and sends daily digests
triggers:
  - schedule: "0 9 * * *"  # Every day at 9 AM
requires:
  tools:
    - browser
    - messaging
---

# Hacker News Monitor

## Instructions

1. Open https://news.ycombinator.com
2. Scan the first 30 stories for any of these topics: {{topics}}
3. For each matching story, extract: title, URL, points, and comment count
4. Format the results as a clean digest with the most relevant stories first
5. Send the digest to the user via their primary channel

## Output Format

**🦞 HN Daily Digest — {{date}}**

For each story:
- **Title** (points | comments)
  Link: url
  Why it matches: brief explanation

If no stories match, send: "Nothing matching your topics on HN today."

The skill automatically loads on the next Gateway restart and runs on the defined schedule.

Source: What Are OpenClaw Skills — DigitalOcean

The Plugin Architecture

Beyond skills, OpenClaw supports four types of plugins that extend the core system without modifying source code:

• Channel plugins — add new messaging platforms
• Memory plugins — swap in alternative storage backends
• Tool plugins — add custom capabilities (APIs, hardware control, specialized processing)
• Provider plugins — integrate custom or self-hosted LLM providers

The plugin loader scans for an openclaw.extensions field in package.json, validates against declared schemas, and hot-loads when configuration is present.

Source: Deep Dive into OpenClaw Architecture — Medium

---

Part 6: Production Deployment with Docker

Running OpenClaw on your laptop is fine for testing. For a reliable, always-on assistant, deploy it in Docker on a VPS.

Why Docker?

Docker isolates OpenClaw from your host system, provides consistent behavior across environments, and makes updates trivial. It is the recommended production deployment method.

Minimum VPS Requirements

• 1 vCPU, 2GB RAM, 20GB SSD — enough for cloud models
• 2 vCPU, 8GB RAM, 40GB SSD — required if running Ollama alongside
• Providers: Hetzner, Contabo, and DigitalOcean all offer suitable plans starting at $5/month

Source: How to Deploy OpenClaw with Docker — CyberNews

Docker Compose Setup

Create a docker-compose.yml:

version: "3.8"

services:
  openclaw:
    image: openclaw/openclaw:latest
    container_name: openclaw-agent
    restart: unless-stopped
    ports:
      - "127.0.0.1:3000:3000"  # Bind to localhost only
    volumes:
      - ./config:/app/config
      - ./data:/app/data
      - ./skills:/app/skills
    environment:
      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}
      - OPENAI_API_KEY=${OPENAI_API_KEY}
    env_file:
      - .env
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:3000/health"]
      interval: 30s
      timeout: 10s
      retries: 3

Create a .env file with your API keys:

ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
DEEPSEEK_API_KEY=sk-...

Launch:

docker compose up -d

Check logs:

docker compose logs -f openclaw

Adding a Reverse Proxy with TLS

Never expose port 3000 directly to the internet. Use Caddy or nginx as a reverse proxy:

# Caddyfile
openclaw.yourdomain.com {
    reverse_proxy localhost:3000
}

Caddy automatically provisions and renews TLS certificates. Your OpenClaw webhooks are now encrypted in transit.

Source: How to Install and Securely Run OpenClaw with Docker — IONOS

---

Part 7: Security Hardening

OpenClaw is powerful precisely because it can execute real actions on your behalf. That power requires careful security configuration.

The Threat Model

Security researchers from Bitsight found over 40,000 internet-exposed OpenClaw instances, with 35.4% flagged as vulnerable to remote code execution. Microsoft's security team published a detailed analysis of the identity, isolation, and runtime risks of exposed AI agents.

The core issue: OpenClaw can execute shell commands, download and run skills from external sources, and perform actions using stored credentials. If the Gateway is accessible from the internet without proper controls, an attacker can instruct it to do anything you can do.

Essential Hardening Checklist

1. Bind to localhost

# config.yaml
gateway:
  host: "127.0.0.1"  # Never use 0.0.0.0
  port: 3000

2. Use Docker for isolation

Run OpenClaw in a container so it only accesses files you explicitly mount:

volumes:
  - ./config:/app/config:ro  # Read-only config
  - ./data:/app/data         # Only the data directory is writable

3. Restrict allowed users

Every channel configuration should include an allowlist:

channels:
  telegram:
    allowedUsers:
      - your_username_only
  whatsapp:
    allowedNumbers:
      - "+1234567890"

4. Rotate and protect secrets

OAuth tokens and API keys are stored under ~/.openclaw/. Ensure this directory has restricted permissions:

chmod 700 ~/.openclaw
chmod 600 ~/.openclaw/config.yaml

5. Keep OpenClaw updated

npm update -g openclaw@latest
# Or with Docker
docker compose pull && docker compose up -d

6. Monitor Gateway logs

# Watch for unexpected tool invocations
docker compose logs -f openclaw | grep -E "tool_call|exec|shell"

Source: Running OpenClaw Safely — Microsoft Security Blog

---

Part 8: Five Automations to Build This Weekend

Theory is useful. Running automations are better. Here are five practical workflows you can deploy today, ranked from easiest to most advanced.

1. Morning Inbox Briefing (15 minutes to set up)

What it does: Every morning at 7 AM, OpenClaw reads your unread emails, categorizes them by urgency, and sends a prioritized summary to Telegram.

How: Install the email-summarizer skill and configure your Gmail credentials:

openclaw skills install email-summarizer

skills:
  email-summarizer:
    schedule: "0 7 * * *"
    emailProvider: gmail
    outputChannel: telegram
    categories:
      - urgent
      - needs-reply
      - informational
      - newsletter

Users report this single automation justifies their entire OpenClaw setup.

2. Meeting Notes to Action Items (20 minutes)

What it does: After a meeting, send OpenClaw the transcript (or an audio file). It extracts action items, assigns them to participants, and emails each person their tasks.

You (via Slack): [uploads meeting_recording.m4a]
"Extract action items from this meeting and email each participant their tasks"

3. Dependency Security Scanner (30 minutes)

What it does: Weekly, OpenClaw checks your project dependencies for security vulnerabilities and available updates, then sends a prioritized report.

Create a custom skill at ~/.openclaw/skills/dep-scanner/skill.md:

---
name: dep-scanner
description: Weekly dependency security audit
triggers:
  - schedule: "0 10 * * 1"  # Every Monday at 10 AM
requires:
  tools:
    - exec
    - messaging
---

# Dependency Security Scanner

1. Navigate to each project directory listed in {{projects}}
2. Run the appropriate audit command (npm audit, pip audit, cargo audit)
3. Categorize findings: critical, high, medium, low
4. Format a report with upgrade commands for each vulnerability
5. Send the report via the user's primary channel

4. Competitor Price Monitor (45 minutes)

What it does: Daily, OpenClaw visits competitor pricing pages, extracts current prices, compares with yesterday's data, and alerts you to any changes.

This workflow uses OpenClaw's browser tool to navigate pricing pages, the file system to store historical data in JSON, and the messaging channel to deliver alerts.

Source: OpenClaw Business Use Cases — Codebridge

5. Full Content Pipeline (1 hour)

What it does: When you publish a blog post, OpenClaw automatically generates platform-specific social media posts for X, LinkedIn, and a newsletter draft — each with appropriate tone, length, and formatting.

Connect your blog RSS feed, configure output templates for each platform, and let OpenClaw handle the distribution. The community reports saving 10+ hours per week with this workflow.

---

Part 9: When You Want to Build the App, Not Just Automate It

OpenClaw excels at personal automation — connecting existing services and executing tasks on your behalf. But there is a gap between "automating a workflow with a chat message" and "building a real application that other people can use."

If you have validated a workflow with OpenClaw and want to turn it into a standalone product — a SaaS tool, an internal dashboard, a customer-facing app — you need an application builder, not an automation framework.

ZBuild is an AI app builder designed for exactly this transition. You describe what you want to build in plain language, and ZBuild generates a full-stack application with a proper UI, database, authentication, and deployment pipeline. Where OpenClaw automates your workflows, ZBuild helps you ship products that others can use.

The workflow looks like this:

1. Prototype with OpenClaw — validate that your automation idea works
2. Build with ZBuild — turn the validated concept into a real application at studio.zbuild.io
3. Deploy — ship it to users

Many of the best SaaS ideas start as personal automations. If you have built something useful with OpenClaw and find yourself thinking "other people would pay for this," that is the signal to move from automation to application.

---

Part 10: Troubleshooting Common Issues

Gateway won't start

# Check if port 3000 is already in use
lsof -i :3000

# Check Node.js version (need 22.16+ or 24+)
node --version

# View detailed Gateway logs
openclaw gateway logs --level debug

WhatsApp disconnects frequently

WhatsApp Web sessions expire periodically. To minimize disconnections:

• Keep the Gateway running continuously (use Docker or systemd)
• Do not open WhatsApp Web in a browser simultaneously
• Use the --install-daemon flag during onboarding

Model timeouts

If your agent times out on complex tasks:

agents:
  primary:
    model: "anthropic/claude-sonnet-4-6"
    timeout: 120000  # Increase timeout to 120 seconds
    maxRetries: 3

Skills not loading

# Verify skill structure
openclaw skills validate ~/.openclaw/skills/your-skill/

# Check skill logs
openclaw gateway logs --filter skills

High API costs

Set up the multi-model fallback chain described in Part 3. Route simple queries to DeepSeek or a local model, and reserve Claude or GPT for tasks that require stronger reasoning.

---

OpenClaw vs. the Alternatives

Feature	OpenClaw	Apple Intelligence	Google Gemini	Microsoft Copilot
Open Source	Yes (MIT)	No	No	No
Self-Hosted	Yes	No	No	No
Chat Platforms	20+	iMessage only	Google Chat	Teams
Choose Your Model	Any LLM	Apple models	Gemini only	GPT only
Browser Control	Full automation	None	Limited	Limited
Shell Commands	Yes	No	No	No
Custom Skills	5,400+ community	None	Gems (limited)	Copilot Studio
Privacy	Fully local option	On-device processing	Cloud only	Cloud only
Cost	Free + model costs	Included with devices	Free tier + paid	$30/month (365)

OpenClaw wins on flexibility, privacy, and extensibility. The tradeoff is setup complexity — the alternatives work out of the box but give you far less control.

Source: What Is OpenClaw — DigitalOcean

---

The Community and Ecosystem

OpenClaw's growth has spawned a significant ecosystem:

• ClawHub — the official skills registry with thousands of community contributions
• awesome-openclaw-skills — a curated list of 5,400+ skills
• nanobot — an ultra-lightweight OpenClaw variant for resource-constrained environments
• IronClaw — a Rust-based reimplementation focused on privacy and security
• OpenClaw Showcase — real examples of what people are building

One-click deployment templates are available on Zeabur, Hostinger, DigitalOcean, and other hosting platforms, making it possible to go from zero to running in under five minutes.

Source: OpenClaw Deploy Guide — Zeabur

---

What's Next for OpenClaw

The project shows no signs of slowing down. With 247K+ stars and 47,700 forks, it has become the de facto standard for self-hosted AI agents. The plugin ecosystem is expanding rapidly, with new channel integrations, tool plugins, and skills being published daily.

The bigger picture: OpenClaw represents a shift in how people interact with AI. Instead of visiting a website to chat with a bot, you send a text message and your AI assistant handles the rest — on your machine, under your control, with the model of your choice.

If you have been waiting for AI to move beyond chatbots and into real agency, OpenClaw is the place to start.

---

Quick Reference

Task	Command
Install	npm install -g openclaw@latest
Onboard	openclaw onboard --install-daemon
Start Gateway	openclaw gateway start
Stop Gateway	openclaw gateway stop
Restart Gateway	openclaw gateway restart
View logs	openclaw gateway logs
Install a skill	openclaw skills install <name>
Search skills	openclaw skills search "<query>"
List skills	openclaw skills list
Update	npm update -g openclaw@latest
Docker start	docker compose up -d
Docker logs	docker compose logs -f openclaw

---

Sources

• OpenClaw GitHub Repository
• OpenClaw Official Website
• OpenClaw Documentation
• OpenClaw — Wikipedia
• What is OpenClaw? — DigitalOcean
• OpenClaw Integrations
• OpenClaw Personal Assistant Setup Guide
• OpenClaw Skills Documentation
• ClawHub — Official Skills Registry
• awesome-openclaw-skills — 5,400+ Curated Skills
• Using OpenClaw with Ollama — DataCamp
• OpenClaw LLM Setup Guide — LaoZhang AI
• OpenClaw API & Model Configuration — Ofox
• How to Change Your AI Model in OpenClaw
• Best Models for OpenClaw — Haimaker
• OpenClaw Live Browser Control — Goldie Agency
• OpenClaw Use Cases 2026 — TLDL
• Top 10 OpenClaw Use Cases — Simplified
• OpenClaw Use Cases — Hostinger
• Advanced OpenClaw Workflows — LightNode
• OpenClaw Business Use Cases — Codebridge
• Deep Dive into OpenClaw Architecture — Medium
• What Are OpenClaw Skills — DigitalOcean
• How to Deploy OpenClaw with Docker — CyberNews
• OpenClaw Docker Documentation
• How to Install and Securely Run OpenClaw — IONOS
• OpenClaw Deploy Guide — Zeabur
• OpenClaw Security Risks — Bitsight
• Running OpenClaw Safely — Microsoft Security Blog
• Mastering OpenClaw — Medium (Vignaraj Ravi)
• OpenClaw Tutorial — AIML API
• nanobot — Ultra-Lightweight OpenClaw
• IronClaw — Rust-based OpenClaw