OpenClaw: The Complete Guide to the Open-Source AI Personal Assistant

If you’ve spent any time on developer Twitter, Hacker News, or GitHub Trending in the past few weeks, you’ve almost certainly come across OpenClaw — the open-source AI personal assistant that has taken the developer world by storm. With over 165,000 GitHub stars in just two months, OpenClaw is the fastest-growing open-source AI project of 2025–2026 and a signal that the age of personal, self-hosted AI agents has truly arrived.

In this comprehensive guide, we’ll cover everything you need to know: what OpenClaw is, how it works, how to set it up with Docker, how to extend it with skills, and what security considerations you need to be aware of before deploying it.

What Is OpenClaw?

OpenClaw is an open-source autonomous AI personal assistant that runs locally on your own hardware — your laptop, a home server, or a VPS — and connects to the messaging platforms you already use every day. Unlike cloud-based AI assistants like ChatGPT or Siri, OpenClaw gives you full control over your data, your AI model provider, and the actions your assistant can take.

As the project’s tagline puts it: “Your assistant. Your machine. Your rules. 🦞”

At its core, OpenClaw operates through a Gateway process that sits between your messaging channels (WhatsApp, Telegram, Slack, Discord, iMessage, Signal, and more) and AI model providers (Anthropic Claude, OpenAI GPT, Google Gemini, and others via OpenRouter). When you send a message to your OpenClaw bot, the Gateway routes it to your configured AI model, which can then execute tasks using a range of tools and skills.

What makes OpenClaw different from simply chatting with ChatGPT is autonomy and action. OpenClaw doesn’t just respond — it can:

• Manage your calendar and send emails
• Run terminal commands on your machine
• Conduct web research and generate reports
• Automate workflows across connected services
• Send you proactive notifications and reminders
• Remember context across sessions with persistent memory
• Create files, write code, and deploy websites

As IBM’s research team noted, OpenClaw demonstrates that creating agents with true autonomy and real-world usefulness “is not limited to large enterprises” and “can also be community driven.”

The Story Behind the Name

OpenClaw has one of the most entertaining origin stories in open source. The project was created by Peter Steinberger, an Austrian software engineer known for building PSPDFKit. He released it in November 2025 under the name Clawdbot — a playful mashup of “Claude” (the Anthropic AI model it initially used) and “claw.”

That name didn’t last long. Anthropic’s legal team sent a trademark request, prompting a hasty rebrand. The community held a Discord brainstorming session at 5 AM and landed on Moltbot — inspired by the molting process of lobsters, symbolizing growth. As Steinberger later admitted, “it never quite rolled off the tongue.”

In January 2026, the project settled on its current and final name: OpenClaw. This time, the team did their homework — trademark research, domain purchases, and migration code were all completed before the announcement. The name captures the project’s dual identity: Open (open source, community-driven) and Claw (a nod to its lobster heritage).

The naming drama became part of the project’s charm. The lobster mascot, the crab memes, and the self-aware humor around the rebranding all contributed to OpenClaw’s viral appeal.

Key Features of OpenClaw

Here’s what makes OpenClaw stand out as a personal AI assistant:

Multi-Platform Messaging Integration

OpenClaw connects to virtually every messaging platform through a single Gateway process:

• WhatsApp — via QR code pairing
• Telegram — via Bot API token
• Discord — via bot token
• Slack — via app integration
• iMessage — via BlueBubbles
• Signal — via Signal CLI
• Microsoft Teams, Matrix, Google Chat, Zalo, and more

This means you can message your AI assistant from your phone, your laptop, or any device — and it responds in the same channel.

Model Provider Flexibility

OpenClaw is model-agnostic. You’re not locked into a single AI provider. Supported providers include:

• Anthropic Claude (Claude Opus 4.5, Sonnet 4.5, Haiku 4.5)
• OpenAI (GPT-4o, GPT-5, o1)
• Google Gemini
• OpenRouter (access 100+ models)
• MiniMax (M2.1)
• Kimi K2
• Local models via Ollama

You bring your own API keys, and OpenClaw routes to whichever provider you configure.

Persistent Memory

Unlike stateless chat interfaces, OpenClaw retains context across sessions. It learns your preferences, tracks ongoing projects, and remembers past interactions. This persistent memory is stored locally in a database, giving you full ownership of your conversation history and personal knowledge base.

Proactive Notifications

OpenClaw doesn’t just wait for you to ask. It can initiate communication — sending you morning briefings, calendar reminders, task summaries, and custom alerts. You can set up cron-like heartbeat jobs that trigger the assistant at regular intervals.

Extensible Skills System

Skills are modular capabilities that extend OpenClaw’s functionality. The community has built hundreds of skills covering everything from web deployment to habit tracking. We’ll dive deeper into skills below.

Browser-Based Control UI

OpenClaw includes a built-in web dashboard (default port 18789) for chat, configuration, session management, and debugging. You can interact with your assistant through the browser UI or through any connected messaging platform.

OpenClaw Architecture: How It Works

Understanding OpenClaw’s architecture helps you deploy and troubleshoot it effectively.

┌─────────────────────────────────────────────────┐
│                   YOUR DEVICES                  │
│  WhatsApp │ Telegram │ Discord │ Slack │ Web UI │
└─────────────┬───────────────────────────────────┘
              │
              ▼
┌─────────────────────────────────────────────────┐
│              OPENCLAW GATEWAY                   │
│  ┌──────────┐  ┌──────────┐  ┌───────────────┐ │
│  │ Channels │  │ Sessions │  │ Memory DB     │ │
│  │ Router   │  │ Manager  │  │ (Persistent)  │ │
│  └──────────┘  └──────────┘  └───────────────┘ │
│  ┌──────────┐  ┌──────────┐  ┌───────────────┐ │
│  │ Skills   │  │ Security │  │ Config        │ │
│  │ Engine   │  │ Layer    │  │ Management    │ │
│  └──────────┘  └──────────┘  └───────────────┘ │
└─────────────┬───────────────────────────────────┘
              │
              ▼
┌─────────────────────────────────────────────────┐
│            AI MODEL PROVIDERS                   │
│  Anthropic │ OpenAI │ Google │ OpenRouter │ ... │
└─────────────────────────────────────────────────┘

The Gateway is the central process — it’s the single source of truth for sessions, routing, and channel connections. It manages isolated sessions per agent, workspace, or sender, and handles media (images, audio, documents) across platforms.

Agent tools run in sandboxed Docker containers by default, providing isolation between the agent’s code execution and your host system.

How to Install OpenClaw with Docker

Docker is the recommended way to run OpenClaw for most users. It provides isolation, reproducibility, and a cleaner security boundary between OpenClaw’s agent execution and your host system.

Prerequisites

• Docker Desktop (or Docker Engine) + Docker Compose v2
• Node.js ≥ 22 (if running natively)
• An API key from your preferred AI model provider (Anthropic, OpenAI, etc.)
• A messaging platform account (Telegram is the easiest to start with)

Step 1: Clone the Repository

git clone https://github.com/openclaw/openclaw.git
cd openclaw

Step 2: Run the Docker Setup Script

./docker-setup.sh

This single script handles everything:

1. Builds the OpenClaw Docker image locally from the included Dockerfile
2. Runs the onboarding wizard inside a Docker container
3. Generates a gateway token for accessing the Control UI
4. Creates configuration directories (~/.openclaw/ and ~/openclaw/workspace/)
5. Starts the Gateway via Docker Compose

During the onboarding wizard, you’ll be asked to configure:

• Gateway bind: Choose lan for container use
• Gateway auth: Choose token
• AI model provider: Select Anthropic, OpenAI, Google, OpenRouter, etc. and enter your API key
• Messaging channels: Configure WhatsApp, Telegram, Discord, etc.

Step 3: Access the Control UI

Open your browser and navigate to:

http://127.0.0.1:18789/

You’ll need the gateway token generated during setup. If you lose it, regenerate it with:

docker compose run --rm openclaw-cli dashboard --no-open

Step 4: Verify the Installation

docker compose exec openclaw-gateway node dist/index.js health --token "YOUR_TOKEN"

Alternative: Pre-Built Docker Image

If you want to skip the local build, you can use the community-maintained alpine/openclaw image:

export OPENCLAW_IMAGE="alpine/openclaw"
./docker-setup.sh

Cloud Deployment Options

Several cloud providers now offer one-click OpenClaw deployments:

• DigitalOcean — 1-Click Deploy with hardened security defaults
• Hostinger — VPS template with Docker Manager
• Vultr — Deploy guide with Serverless Inference integration

Connecting OpenClaw to Messaging Platforms

Telegram (Easiest to Start)

Telegram is the lowest-friction way to connect to OpenClaw:

1. Chat with @BotFather on Telegram to create a new bot
2. Copy the bot token
3. Add the channel:

docker compose run --rm openclaw-cli channels add --channel telegram --token <YOUR_BOT_TOKEN>

4. Message your bot on Telegram. OpenClaw will send you a pairing code
5. Approve the pairing:

docker compose run --rm openclaw-cli pairing approve telegram <CODE>

You’re now messaging your personal AI assistant from your phone!

WhatsApp

docker compose run --rm openclaw-cli channels login

This generates a QR code that you scan with your WhatsApp app, similar to WhatsApp Web.

Discord & Slack

Create a bot application on the respective platform, get the token, and add it:

# Discord
docker compose run --rm openclaw-cli channels add --channel discord --token <TOKEN>

# Slack
docker compose run --rm openclaw-cli channels add --channel slack --token <TOKEN>

For full channel setup documentation, see: docs.openclaw.ai/channels

OpenClaw Skills: Extending Your AI Assistant

Skills are what make OpenClaw truly powerful. They are modular capability packages — folders containing instructions, scripts, and resources that extend what your assistant can do.

What Can Skills Do?

The awesome-openclaw-skills repository lists hundreds of community-built skills across categories:

Productivity & Knowledge Management:

• obsidian-daily — Manage Obsidian Daily Notes
• notebook — Local-first personal knowledge base
• people-memories — Capture notes about people you mention
• project-context-sync — Keep a living project state document updated after each commit
• work-report — Generate daily/weekly work reports from git commits

Development & DevOps:

• pr-commit-workflow — Standardize commits and pull requests
• web-deploy-github — Deploy static websites to GitHub Pages
• flyio-cli — Manage Fly.io deployments
• gcloud — Manage Google Cloud Platform resources
• dokploy — Manage Dokploy deployments and domains

Apple/Swift Development:

• apple-docs — Query Apple Developer Documentation and WWDC videos
• swiftui-liquid-glass — Implement iOS 26+ Liquid Glass API features
• swift-concurrency-expert — Swift Concurrency review for Swift 6.2+
• macos-spm-app-packaging — Package SwiftPM-based macOS apps

AI & Research:

• penfield — Persistent agent memory with hybrid search
• agent-relay-digest — Create curated digests of agent conversations
• ontology — Typed knowledge graph for structured agent memory

Security:

• skill-vetter — Security-first skill vetting for AI agents

Installing Skills

You can install skills from the community marketplace or create your own:

# Browse and install from the skills marketplace
openclaw skills install <skill-name>

Or manually add skills by placing them in your ~/.openclaw/skills/ directory.

Creating Your Own Skills

Each skill is a folder containing a SKILL.md file:

---
name: morning-briefing
description: Generate a morning briefing with weather, calendar, and news
---

When the user asks for a morning briefing:
1. Check the weather for the user's location
2. Summarize today's calendar events
3. Pull top 3 news headlines relevant to the user's interests
4. Format as a concise, scannable message

For a deeper understanding of how skills work across AI coding tools, check out our blog on Claude Code and Skills.

OpenClaw Security: What You Need to Know

This is the most important section of this guide. OpenClaw is powerful, but that power comes with significant security implications that every user must understand.

The Security Landscape

OpenClaw has faced serious scrutiny from cybersecurity researchers. Here’s what you need to know:

Recent Security Advisories:

• A one-click remote code execution vulnerability was discovered that allowed attackers to steal authentication tokens via malicious web pages and execute arbitrary code on the host machine
• Two command injection vulnerabilities were patched in rapid succession
• Koi Security identified 341 malicious skills submitted to ClawHub, the community skill repository

What Security Researchers Are Saying:

As The Register reported, prominent developer Laurie Voss (founding CTO of npm) described OpenClaw as “a security dumpster fire.” The project’s extensible architecture introduces supply chain risks — compromised or poorly audited skill modules could enable privilege escalation or arbitrary code execution.

Security Best Practices

If you decide to run OpenClaw, follow these guidelines:

1. Run in Docker containers — Never run OpenClaw directly on a production machine. Docker provides meaningful isolation between the agent and your host system.
2. Use DM pairing — Enable pairing mode (dmPolicy="pairing") so that unknown senders must be explicitly approved before interacting with your bot.
3. Audit skills before installing — Don’t blindly install community skills. Read the SKILL.md and any bundled scripts before adding them. Use the skill-vetter skill to help.
4. Keep OpenClaw updated — The project is actively patching vulnerabilities. Run openclaw update regularly.
5. Use a dedicated VPS — Don’t run OpenClaw on the same machine where you store sensitive credentials, production databases, or business-critical systems.
6. Set network restrictions — Configure Docker with docker.network: "none" for sandboxed agent sessions to prevent unauthorized egress.
7. Monitor API costs — Set spending limits on your AI provider accounts to prevent runaway token consumption.
8. Treat inbound DMs as untrusted input — OpenClaw connects to real messaging surfaces. Every message from an external user is a potential prompt injection vector.

Who Should (and Shouldn’t) Use OpenClaw

Use OpenClaw if you:

• Are a developer comfortable with Docker and terminal operations
• Understand the security implications of running autonomous agents
• Want full control over your AI assistant and data
• Are willing to actively monitor and maintain your deployment

Avoid OpenClaw if you:

• Are not comfortable reviewing code and security advisories
• Need a production-grade, enterprise-ready solution today
• Don’t want to manage infrastructure
• Are looking for a consumer-friendly AI assistant (use Claude.ai or ChatGPT instead)

OpenClaw Cost Considerations

One aspect that catches many new users off guard is the API cost of running an always-on AI assistant.

A widely-shared cautionary tale: developer Benjamin De Kraker discovered that OpenClaw burned through $20 of Anthropic API tokens overnight — simply because a heartbeat cron job checking the time every 30 minutes was sending ~120,000 tokens of context per request. At $0.75 per time check, the projected monthly cost just for reminders was around $750.

Cost Optimization Tips

• Choose cost-effective models — Use smaller models (Claude Haiku, GPT-4o-mini) for routine tasks and reserve larger models for complex work
• Limit context size — Configure your agent to send minimal context for simple operations
• Set API spending limits — All major providers offer spending caps
• Monitor token usage — Track usage through your provider’s dashboard
• Use local models — For privacy-insensitive tasks, run models locally via Ollama to eliminate API costs entirely
• Optimize heartbeat/cron intervals — If using proactive notifications, ensure they don’t send excessive context

OpenClaw vs Other AI Agents

How does OpenClaw compare to other AI agent solutions?

OpenClaw’s unique value proposition is the combination of self-hosting + messaging platform integration + model flexibility + open source. No other tool offers all four today.

Getting Involved: Community and Contributing

OpenClaw has one of the most active open-source communities in the AI space right now:

• GitHub: github.com/openclaw/openclaw — 149K+ stars
• Documentation: docs.openclaw.ai
• Official Website: openclaw.ai
• Discord: Join the “Claw Crew” community on the OpenClaw Discord server
• Skills Repository: VoltAgent/awesome-openclaw-skills
• Docker Image: alpine/openclaw on Docker Hub

The project is actively seeking contributors and sponsors. The roadmap prioritizes security as the top concern, followed by gateway reliability and support for additional models and providers.

Frequently Asked Questions

What is OpenClaw?

OpenClaw (formerly Moltbot/Clawdbot) is an open-source, self-hosted AI personal assistant that connects to messaging platforms like WhatsApp, Telegram, and Discord. It runs locally on your own hardware and integrates with AI model providers like Anthropic Claude and OpenAI GPT to execute tasks autonomously.

Is OpenClaw free?

OpenClaw itself is free and open source under the MIT License. However, you’ll need to pay for API usage with your chosen AI model provider (Anthropic, OpenAI, etc.) unless you run local models via Ollama.

Is OpenClaw safe to use?

OpenClaw has had several significant security vulnerabilities, and cybersecurity researchers have raised concerns about its architecture. It is currently best suited for advanced users who understand the security implications of running autonomous agents. Always run it in Docker, audit skills before installing, and keep it updated.

Can I run OpenClaw with Docker?

Yes! Docker is the recommended deployment method. Use the included docker-setup.sh script or the alpine/openclaw pre-built image. See our installation guide above for step-by-step instructions.

What AI models does OpenClaw support?

OpenClaw supports Anthropic Claude, OpenAI GPT, Google Gemini, MiniMax, Kimi K2, and any model available through OpenRouter. You can also use local models via Ollama.

What happened to Moltbot and Clawdbot?

Clawdbot was the original name (November 2025). It was renamed to Moltbot after Anthropic raised trademark concerns, then finally renamed to OpenClaw in January 2026. They are all the same project.

How much does it cost to run OpenClaw?

Costs depend entirely on your AI model usage. Light usage might cost $5–10/month, but poorly configured setups (especially with proactive notifications and large models) can cost significantly more. Set API spending limits to avoid surprises.

Can OpenClaw connect to WhatsApp?

Yes. OpenClaw supports WhatsApp via QR code pairing. Run openclaw channels login and scan the QR code with your WhatsApp app.

Final Thoughts

OpenClaw represents a fascinating inflection point in the AI agent landscape. It’s the first open-source project to successfully combine self-hosting, multi-platform messaging, model flexibility, and extensible skills into a single, community-driven package. The fact that it hit 149,000 GitHub stars in two months tells you something about the hunger for personal AI agents that users actually control.

But OpenClaw is also a reminder that with great power comes great responsibility. The security concerns are real, the costs can spiral, and the project is still maturing rapidly. It’s not yet a tool for everyone — but for developers and power users willing to get their hands dirty, it offers a genuine glimpse of what personal AI assistants will look like in the near future.

As IBM’s research scientist Kaoutar El Maghraoui put it, OpenClaw shows that autonomous AI with real-world usefulness “can also be community driven.” That’s a powerful idea — and one worth paying attention to.

Found this guide helpful? Share it with your community! Join the Collabnix Slack to discuss OpenClaw, Docker, and AI agents with 17,000+ developers. Follow us at collabnix.com for more hands-on guides.

Related Reads on Collabnix:

• Claude Code and Skills: Turning AI from a Chatbot into Your Personal Developer Team
• Docker Model Runner: Running AI Models Locally
• MCP Gateway: Building Multi-Agent Systems with Docker