What Is OpenClaw? Complete Guide to
the AI Agent That Broke GitHub

OpenClaw is a free, open-source personal AI assistant created by Peter Steinberger (@steipete), an Austrian developer best known for founding PSPDFKit, which was acquired by Insight Partners for approximately $100 million. OpenClaw runs locally on your own devices and connects to large language models — Claude, GPT, DeepSeek, and others — to execute real-world tasks autonomously.

Unlike cloud-based AI assistants such as ChatGPT or Google Gemini, OpenClaw operates as a local-first Gateway — a unified control plane for messaging, automation, and AI interactions. Its tagline is straightforward: "Your own personal AI assistant. Any OS. Any Platform."

What OpenClaw Can Do

The project has attracted extraordinary attention since its public launch in January 2026. Andrej Karpathy, the former head of AI at Tesla, described it as the "most incredible sci-fi takeoff-adjacent thing I have seen recently." At the same time, security researcher Simon Willison warned of "Challenger disaster" security risks. Both assessments have proven accurate.

The naming saga of OpenClaw is one of the most chaotic episodes in recent open-source history, involving trademark disputes, crypto scammers, handle snipers, and three name changes in under a week. Here is the full timeline.

OpenClaw is architected as a Gateway — a central WebSocket control plane that coordinates multiple runtime components. Understanding this architecture is essential for both using and securing the platform.

Messaging Platform Support

OpenClaw supports 12+ messaging channels, each implemented through platform-specific libraries:

Model Support

OpenClaw is model-agnostic but has a clear recommendation hierarchy:

Workspace & Skills

OpenClaw uses a file-based workspace system rooted at ~/.openclaw/workspace. Key prompt files are injected into every session:

Community extensions are distributed through ClawHub, the skill marketplace. Skills can add new capabilities ranging from simple utilities to complex integrations. However, as we will cover in the security section, ClawHub has become a significant attack vector.

# Install OpenClaw globally
npm install -g openclaw@latest

# Run the onboarding wizard (installs daemon)
openclaw onboard --install-daemon

{
  "agent": {
    "model": "anthropic/claude-opus-4-6"
  }
}

Chat Commands

OpenClaw's feature set goes well beyond a simple chatbot. It is a full platform with capabilities that rival commercial AI assistant products. Here are the most significant features for security teams, developers, and power users to understand.

OpenClaw's power is also its greatest security liability. The combination of capabilities that makes it useful — access to files, messaging, web browsing, command execution — creates an attack surface that has alarmed the security community. This section covers the major security concerns in detail.

The AI agent landscape in 2026 includes several major players. Understanding how OpenClaw compares to alternatives helps contextualize both its appeal and its risks.

One of the most fascinating and controversial byproducts of the OpenClaw phenomenon is Moltbook — an AI-only social network where AI agents, not humans, are the primary participants.

Simon Willison called Moltbook the "most interesting place on the internet right now." And in many ways it was. AI agents were posting opinions, debating each other, forming communities around shared interests, and even attempting to trade with one another. It felt like watching an alien civilization bootstrap itself in real-time.

The Controversies

If you decide to set up OpenClaw — with full awareness of the security considerations covered above — here is a practical guide to getting started.

# Step 1: Install Node.js 22+ (using nvm)
nvm install 22
nvm use 22

# Step 2: Install OpenClaw
npm install -g openclaw@latest

# Step 3: Run onboarding
openclaw onboard --install-daemon

# Step 6: Test commands (send via your paired messaging app)
# /status
# /new

Update Channels

OpenClaw offers three update channels: stable (recommended for most users), beta (early access to new features), and dev (bleeding edge, may be unstable).

# Switch to stable channel (recommended)
openclaw update --channel stable

# Switch to beta channel
openclaw update --channel beta

# Switch to dev channel (not recommended for production)
openclaw update --channel dev

If you choose to run OpenClaw, the following checklist represents the minimum security measures every deployment should implement. These recommendations are drawn from published advisories by CrowdStrike, Palo Alto Networks, Koi Security, and independent researchers.

Despite the security concerns, OpenClaw represents a genuine inflection point in how humans interact with AI. The project's trajectory will shape the broader AI agent ecosystem in the years ahead.