TL;DR: 358,890 GitHub stars. 72,986 forks. One TypeScript app that lets AI agents run your terminal, send WhatsApp messages, and execute shell commands autonomously. But 42,900 instances are exposed online with no default authentication.

What Is OpenClaw?

OpenClaw is a self-hosted, open-source personal AI assistant framework that connects to your messaging apps (WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Teams, Matrix) and gives AI agents direct shell command execution on your machine.

Created by Peter Steinberger (PSPDFKit founder) in January 2026, it became one of the fastest-growing open-source projects ever, reaching 358K+ stars in just months.

Technical Architecture

OpenClaw runs a 5-component architecture:

Component Function
Channel Adapters Parse messages from WhatsApp/Slack/etc.
Control Interfaces Web UI, CLI, macOS app
Gateway Control Plane WebSocket hub on port 18789
Agent Runtime Session resolution, context assembly
Canvas/A2UI Rich content rendering

Tech Stack: TypeScript, WebSocket gateway, AgentSkills standard.

Supported Models: Claude Opus 4.6, GPT, DeepSeek, MiniMax, Gemini, Qwen, and local models via Ollama/LMStudio.

Installation

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

Requirements: Node.js v22+ (breaks silently on older versions), API key for LLM provider.

Key Features

Feature Description
Multi-Channel Persistence Same agent across WhatsApp, Slack, Telegram, Discord
Shell Command Execution Direct terminal access for AI agents
Persistent Memory Context survives across conversations
700+ ClawHub Skills Tesla, Gmail, GitHub automation pre-built
Cron Jobs & Heartbeats 24/7 autonomous background tasks
Voice Integration Whisper transcription support
Multi-Agent Routing Isolated workspaces for different tasks
MIT License Free and open source

Competitor Comparison

Aspect OpenClaw Open WebUI AutoGPT
Purpose AI agent framework Chat interface Autonomous agent
Stars 358K+ 60K+ 160K+
Auth None by default Built-in N/A
Execution Direct shell Plugin-based Simulated
Setup Single command Docker Python pip

Key Difference: OpenClaw executes commands directly (efficient, risky). AutoGPT simulates execution (safer, slower).

Community Sentiment

Positive

  • Andrej Karpathy: "The most incredible sci-fi takeoff-adjacent thing I've seen"
  • "First time I've felt like I'm living in the future since ChatGPT launch"
  • 700+ attendees at ClawCon SF

Critical

  • "OpenClaw is a security nightmare" - Ars Technica investigation found 42,900 exposed instances
  • "OVERHYPED" - Some users find it unnecessary after testing
  • Corporate bans emerging due to security concerns

Model Ratings from Community:

  • Qwen3.6: 9/10 (best balance)
  • GLM5.1: 8/10 (strong reasoning)
  • MiniMax2.7: 8/10 (fast, capable)
  • Claude Opus: 7/10 (expensive but good)

Security Vulnerabilities

Critical CVEs:

  • CVE-2026-33579 (CVSS 8.1-9.8): Privilege escalation
  • CVE-2026-25253 (CVSS 8.8): Remote code execution
  • CVE-2026-26326: Information disclosure

Risks:

  • No default authentication
  • 42,900 exposed instances online
  • Supply chain risk from 700+ community skills
  • Prompt injection potential
  • Credential theft possible
  • No audit logging

Performance Metrics

Metric Value
Max Concurrent Users (P95 <5s) ~40 on 2C/4G
Max Throughput ~12 req/s
Memory Usage ~1.2GB / 4GB
Enterprise Goal: Task Completion >=85%
Enterprise Goal: Tool Error Rate <=5%

The Bottom Line

OpenClaw is revolutionary but high-risk. It offers unprecedented AI agent capabilities with multi-channel persistence and self-improvement, but security vulnerabilities make it dangerous for enterprise use without careful configuration.

For personal use: Exceptional productivity boost if you configure authentication and run behind a firewall.

For enterprise: Needs security audit, VPN, and authentication before deployment.

GitHub: https://github.com/openclaw/openclaw Website: https://openclaw.ai