OpenClaw: The Complete Guide to Your Personal AI Assistant

OpenClaw is the fastest-growing open-source project in GitHub history, crossing 280,000 stars in just a few months. Unlike ChatGPT or other chatbots that live in a browser tab, OpenClaw runs locally on your machine and connects to the messaging apps you already use — WhatsApp, Telegram, Slack, Discord, and 20+ more.

It doesn't just chat. It has eyes and hands — it can browse the web, read and write files, run shell commands, and automate entire workflows autonomously. Think of it as your personal AI employee that never sleeps.

Let's break down everything you need to know.

What is OpenClaw?

OpenClaw is a self-hosted, open-source personal AI assistant created by Peter Steinberger (founder of PSPDFKit) in November 2025, originally under the name Clawdbot. It went viral in late January 2026 after the Moltbook project brought it mainstream attention.

Here's what makes it different from standard AI chatbots:

• Runs locally — your data stays on your machine
• Model-agnostic — works with Claude, GPT, Gemini, DeepSeek, or local models via Ollama
• Multi-channel — one AI, accessible from WhatsApp, Telegram, Slack, Discord, iMessage, Teams, and more
• Agentic — it doesn't just answer questions, it does things autonomously
• Extensible — thousands of community-built skills for every use case

The software itself is MIT licensed and completely free. You only pay for the LLM API calls (typically $10–70/month depending on usage).

Architecture: How OpenClaw Works

Understanding OpenClaw's architecture helps you use it effectively.

The Gateway

Everything flows through a single process called the Gateway. It's the central hub that handles:

• Session management — tracks conversations across channels
• Message routing — sends your messages to the right model and back
• Memory — maintains consistent context whether you message from WhatsApp or Discord
• Tool sandboxing — safely executes commands and file operations

The Plugin System

OpenClaw extends through four types of plugins:

1. Channel plugins — connect additional messaging platforms
2. Memory plugins — alternative storage backends (Redis, PostgreSQL, etc.)
3. Tool plugins — custom capabilities (web browsing, file management, APIs)
4. Provider plugins — custom LLM providers or self-hosted models

LLM Providers

OpenClaw is flexible about which AI model powers it:

• Anthropic (Claude) — recommended for complex reasoning
• OpenAI (GPT) — latest release supports GPT-5.4
• Google (Gemini) — via OpenAI-compatible API
• Local models — Ollama, vLLM, LM Studio for fully offline usage
• LiteLLM Proxy — access 100+ providers with cost tracking and automatic failover

Installation

OpenClaw supports Windows 10/11, macOS 11+, and Linux (Ubuntu 20.04+). You need Node.js 18+ and at least 4GB RAM (8GB recommended).

Method 1: One-Liner (Recommended)

The fastest way to get started. This installs Node.js (if needed) and OpenClaw in one command:

curl -fsSL https://get.openclaw.ai | bash

Method 2: npm/pnpm

If you already have Node.js installed:

npm install -g openclaw

Or with pnpm:

pnpm add -g openclaw

Method 3: Docker

Docker provides an isolated environment, ideal for servers or cloud VMs:

docker run -d \
  --name openclaw \
  -v openclaw-data:/data \
  -p 3000:3000 \
  ghcr.io/openclaw/openclaw:latest

Data lives in a mounted volume for persistence across restarts.

First Run

After installation, start the onboarding wizard:

openclaw onboard

This walks you through:

1. Choosing your AI provider (Anthropic, OpenAI, etc.)
2. Entering your API key
3. Connecting your first messaging channel
4. Configuring initial skills

Connecting Channels

Channels are pluggable connectors that let OpenClaw talk through your favorite messaging apps. The Gateway routes messages and shares a single memory system across all channels — so your AI has one consistent context everywhere.

Telegram (Easiest)

Telegram is the simplest channel to set up:

1. Message @BotFather on Telegram
2. Create a new bot and copy the token
3. Connect it:

openclaw channels add --channel telegram --token "YOUR_BOT_TOKEN"

That's it. Message your bot and OpenClaw responds.

WhatsApp

WhatsApp uses the Baileys library and connects via WhatsApp Web:

openclaw channels add --channel whatsapp

Scan the QR code from Settings > Linked Devices on your phone. OpenClaw runs as a linked device.

Note: WhatsApp may disconnect the session after ~14 days, requiring a re-scan.

Discord

openclaw channels add --channel discord --token "YOUR_BOT_TOKEN"

Create a Discord bot in the Discord Developer Portal, invite it to your server, and paste the token.

Slack

openclaw channels add --channel slack

Follow the OAuth flow to connect your Slack workspace. OpenClaw appears as a bot in your channels.

Running Multiple Channels

You can run WhatsApp, Telegram, Discord, and Slack simultaneously. Messages are automatically routed to the correct channel:

openclaw channels list

Understanding Skills

Skills are the secret sauce of OpenClaw. They're textbooks that teach your AI how to accomplish specific tasks by combining tools.

How Skills Work

When you ask OpenClaw to do something, it matches your request against skill descriptions. If one fits, the agent reads those instructions and follows them. You don't invoke skills by name — just ask naturally.

A skill is simply a folder with a SKILL.md file containing YAML frontmatter and markdown instructions:

my-skill/
  SKILL.md

Example SKILL.md:

---
name: daily-standup
description: Summarize git commits and open PRs for a daily standup
tools:
  - shell
  - github
---

## Instructions

1. Run `git log --oneline --since="yesterday"` to get recent commits
2. Use the GitHub tool to list open PRs
3. Format a concise standup summary with:
   - What was done (from commits)
   - What's in progress (from open PRs)
   - Any blockers mentioned in PR comments

Installing Skills from ClawHub

ClawHub is the public skills registry with 5,400+ community-built skills:

# Browse and install skills
clawhub install daily-standup
clawhub install web-scraper
clawhub install notion-sync

Skill Precedence

If a skill name conflicts, precedence is:

1. <workspace>/skills (highest)
2. ~/.openclaw/skills
3. Bundled skills (lowest)

This means you can always override a global skill with a workspace-specific version.

Popular Skills

Some of the most useful skills from the registry:

• web-scraper — extract data from websites
• github-manager — manage PRs, issues, and releases
• notion-sync — two-way sync with Notion databases
• email-assistant — draft, send, and organize emails
• calendar-manager — manage Google Calendar events
• code-reviewer — review PRs with detailed feedback
• daily-digest — morning summary of news, calendar, and tasks

Building Custom Skills

Creating your own skill takes minutes:

mkdir -p ~/.openclaw/skills/my-custom-skill

Write your SKILL.md with clear instructions, and OpenClaw picks it up immediately. The key is writing good instructions — think of it as writing a playbook for a smart intern.

Practical Use Cases

Developer Workflows

OpenClaw shines for developers. Message it from Telegram while commuting:

"What PRs are open on the frontend repo? Summarize any that need my review."

It runs the GitHub API, reads the diffs, and gives you a summary — all from your phone.

Other developer use cases:

• Automated debugging and log analysis
• DevOps monitoring and alerts
• Codebase search and documentation
• CI/CD pipeline management
• Dependency updates and security patches

Personal Productivity

Connect OpenClaw to your productivity tools:

• Apple Notes/Reminders — "Add 'buy groceries' to my reminders for 5pm"
• Notion — "Create a new page in my journal database with today's notes"
• Obsidian — "Search my vault for notes about project planning"
• Trello — "Move the 'API redesign' card to In Progress"
• Things 3 — "What's due this week?"

Web Automation

OpenClaw can browse the web on your behalf:

• Fill out forms
• Scrape and structure data
• Monitor websites for changes
• Compare prices across stores
• Research topics and compile reports

Security Best Practices

OpenClaw is powerful, but power requires responsibility. Here are the critical security considerations.

The Risks

Prompt injection — attackers can craft messages that manipulate the model into unsafe actions. This is an unsolved problem in AI, and OpenClaw is not immune.

Exposed credentials — OpenClaw stores API keys and channel tokens in plaintext. If your instance is accessible to others, these become targets.

Malicious skills — research found that nearly 20% of skills in the registry were suspicious or malicious, including cloned skills, typosquats, and social engineering attempts.

Network exposure — OpenClaw's Canvas Host defaults to binding 0.0.0.0, meaning any device on your local network can reach it (tracked as GitHub Issue #5263).

Security Checklist

Follow these best practices to run OpenClaw safely:

1. Isolate in Docker — run OpenClaw in a container, not directly on your host machine
2. Bind to localhost — set the Gateway's bind to 127.0.0.1 in openclaw.json
3. Vet all skills — read every SKILL.md before installing. Treat it like a code review
4. Use least-privilege API tokens — give OpenClaw only the permissions it needs
5. Never expose to the internet — keep your instance behind a firewall or VPN
6. Keep secrets separate — don't store sensitive files in the agent's reachable filesystem
7. Don't connect primary accounts — use dedicated bot accounts for messaging channels
8. Update regularly — security patches are frequent; stay on the latest version

As Microsoft's security team recommends: use OpenClaw only in isolated environments that do not have access to any non-dedicated credentials or data which must not be leaked.

Using Local Models (Fully Offline)

If you want complete privacy with zero data leaving your machine, run OpenClaw with a local model:

With Ollama

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

# Pull a model
ollama pull llama3.1:70b

# Configure OpenClaw to use it
openclaw config set provider ollama
openclaw config set model llama3.1:70b

With LM Studio

1. Download and install LM Studio
2. Load your preferred model
3. Start the local server (OpenAI-compatible API)
4. Point OpenClaw to it:

openclaw config set provider custom
openclaw config set api-url http://localhost:1234/v1

Trade-offs

Local models give you full privacy but come with trade-offs:

• Quality — local models are less capable than Claude or GPT for complex tasks
• Speed — depends on your hardware; a good GPU helps significantly
• RAM — 70B models need 40GB+ RAM; smaller models (7B–13B) work on 16GB

For most users, a hybrid approach works best: use Claude/GPT for complex tasks and a local model for sensitive or private operations.

Cost Breakdown

OpenClaw itself is free. Here's what you'll spend on API calls:

Tips to reduce costs:

• Use a cheaper model for simple tasks (Haiku for quick lookups)
• Route through LiteLLM for automatic cost tracking and spend limits
• Cache frequent queries with the memory system
• Use local models for non-critical tasks

What's Next for OpenClaw?

OpenClaw is evolving fast:

• Foundation governance — Steinberger joined OpenAI in February 2026 and the project is moving to an open-source foundation for long-term sustainability
• Memory hot swapping — the latest release (v2026.3.8) supports swapping memory backends without restarting
• GPT-5.4 support — native integration with OpenAI's latest model
• Growing ecosystem — 5,400+ skills and counting, with an active community building more daily

The project's trajectory suggests it will become a fundamental piece of personal computing infrastructure — your always-on AI layer that connects everything.

Final Thoughts

OpenClaw represents a shift in how we interact with AI. Instead of visiting a website to chat, you have an AI assistant embedded in the tools you already use, capable of taking real action on your behalf.

Start small: install OpenClaw, connect Telegram, and ask it to summarize your GitHub notifications. Once you see it work, you'll start finding uses everywhere.

Just remember: security first. Run it in Docker, vet your skills, and never expose it to the public internet. With those basics covered, OpenClaw is genuinely one of the most useful tools to come out of the open-source AI movement.