OpenClaw: The Complete Guide to Your Personal AI Assistant

In the rapidly evolving world of AI assistants, most solutions lock you into a single platform, a single interface, and a single vendor's ecosystem. OpenClaw takes a radically different approach. It's a fully open-source, self-hosted personal AI assistant that meets you where you already are — whether that's WhatsApp, Telegram, Slack, Discord, or any of its 22+ supported messaging channels.

With over 260,000 GitHub stars, 50,000 forks, and 1,058 contributors, OpenClaw has become one of the most popular open-source AI projects in the world. Built in TypeScript and released under the MIT License, it represents a new paradigm: an AI assistant that runs on your own devices, respects your privacy, and gives you full control over your data.

🦞 "EXFOLIATE! EXFOLIATE!" — The OpenClaw mascot, a lobster, embodies the project's philosophy of shedding old constraints and embracing a new, more personal way of interacting with AI.

Repository: OpenClaw on GitHub
Website: openclaw.ai
Documentation: docs.openclaw.ai

---

What Problem Does OpenClaw Solve?

Traditional AI assistants like ChatGPT, Gemini, or Copilot are powerful but fundamentally limited. They live in a browser tab or a dedicated app. They can't control your computer, read your local files, or integrate with the messaging apps your team already uses. They're stateless — each conversation starts from scratch unless you explicitly provide context.

OpenClaw bridges this gap by providing:

• A single AI that works across all your messaging platforms — no need to switch between apps
• Direct system access — it can run shell commands, browse the web, manage files, and control your machine
• Persistent memory and personality — through workspace files like SOUL.md and AGENTS.md
• Proactive automation — cron jobs, webhooks, and Gmail hooks let it act autonomously
• Complete privacy — everything runs locally on your hardware

As one Reddit user put it: "It's not just a chatbot. It's like having a smart developer sitting at your desk with eyes and hands."

---

Key Features & Capabilities

Multi-Channel Messaging Integration

OpenClaw's most distinctive feature is its support for 22+ messaging channels:

Category	Channels
Consumer	WhatsApp, Telegram, Signal, iMessage (BlueBubbles), Zalo
Enterprise	Slack, Discord, Microsoft Teams, Google Chat, Mattermost
Open/Federated	Matrix, IRC, Nostr, Nextcloud Talk
Other	Feishu, LINE, Synology Chat, Tlon, Twitch, WebChat

Each channel is a first-class citizen. You can have separate agent configurations per channel, route specific contacts to isolated agents, and even have the AI respond differently based on which platform the message comes from.

# Send a message via CLI
openclaw message send --to +1234567890 --message "Hello from OpenClaw"

# Talk to the assistant with high-level reasoning
openclaw agent --message "Ship checklist" --thinking high

Local-First Gateway Architecture

The heart of OpenClaw is its Gateway — a WebSocket-based control plane that runs on your machine:

WhatsApp / Telegram / Slack / Discord / Signal / ...
                    │
                    ▼
    ┌───────────────────────────────┐
    │         Gateway               │
    │      (control plane)          │
    │   ws://127.0.0.1:18789        │
    └──────────────┬────────────────┘
                   │
        ├─ Pi agent (RPC)
        ├─ CLI (openclaw …)
        ├─ WebChat UI
        ├─ macOS app
        └─ iOS / Android nodes

The Gateway handles:

• Session management — each conversation gets its own session with isolated context
• Channel routing — messages flow from messaging platforms through the Gateway to the AI agent
• Tool orchestration — browser control, file access, cron jobs, and more
• Presence and typing indicators — making the AI feel responsive and alive
• Model failover — automatic switching between AI providers if one fails

Voice & Visual Capabilities

OpenClaw goes beyond text with:

• Voice Wake — wake words on macOS and iOS (like "Hey Claw")
• Talk Mode — continuous voice conversation on Android with ElevenLabs TTS or system TTS fallback
• Live Canvas — an agent-driven visual workspace where the AI can render interactive content using A2UI (Agent-to-UI push)
• Camera and Screen Recording — the AI can take snapshots, capture screen recordings, and process visual data from your devices

Browser Control

OpenClaw manages its own dedicated Chrome/Chromium instance:

# The AI can browse the web autonomously
# - Navigate to URLs
# - Take page snapshots
# - Fill forms and click buttons
# - Upload files
# - Manage browser profiles

This enables use cases like automated web research, form filling, monitoring websites for changes, and web scraping — all orchestrated conversationally through your favorite messaging app.

Tools & Automation

The platform includes a rich set of built-in tools:

• Browser control — CDP-based Chrome automation with snapshots and actions
• Canvas — visual workspace with A2UI push/reset, JavaScript eval, and snapshot
• Nodes — camera snap/clip, screen record, location query, system notifications
• Cron & Wakeups — scheduled task execution
• Webhooks — HTTP-driven automation triggers
• Gmail Pub/Sub — email-driven event hooks
• Skills Platform — bundled, managed, and workspace skills with install gating

---

Getting Started

Prerequisites

• Node.js 22 or higher (required)
• macOS, Linux, or Windows via WSL2 (WSL2 strongly recommended on Windows)
• npm, pnpm, or bun (your choice of package manager)

Installation

The recommended approach is the onboarding wizard:

# Install globally
npm install -g openclaw@latest

# Run the interactive wizard
openclaw onboard --install-daemon

The wizard guides you through:

1. Setting up the Gateway daemon (auto-starts on boot via launchd/systemd)
2. Configuring your AI model provider
3. Pairing messaging channels
4. Installing initial skills

Quick Start (TL;DR)

# Install + start daemon
npm install -g openclaw@latest
openclaw onboard --install-daemon

# Start the gateway
openclaw gateway --port 18789 --verbose

# Send a message
openclaw message send --to +1234567890 --message "Hello from OpenClaw"

# Direct agent interaction
openclaw agent --message "What's on my calendar today?" --thinking high

From Source (Development)

git clone https://github.com/openclaw/openclaw.git
cd openclaw
pnpm install
pnpm ui:build          # builds the WebChat UI
pnpm build             # compiles TypeScript to dist/
pnpm openclaw onboard --install-daemon

# Development mode with auto-reload
pnpm gateway:watch

Health Check

After setup, always run the doctor to verify everything is configured correctly:

openclaw doctor

This surfaces misconfigured DM policies, missing dependencies, and other common issues.

---

Deep Dive: The Agent Workspace & Skills

Workspace Structure

Every OpenClaw agent has a workspace at ~/.openclaw/workspace/ containing:

• AGENTS.md — Defines the agent's capabilities and behavior
• SOUL.md — The agent's personality, preferences, and persistent memory
• TOOLS.md — Custom tool definitions and configurations

These files are injected into every prompt, giving the AI persistent context about who it is and how it should behave.

Skills System

Skills extend the agent's capabilities. They live in ~/.openclaw/workspace/skills/<skill>/SKILL.md and are managed through:

• Bundled skills — ship with OpenClaw out of the box
• Managed skills — installed from ClawHub (the skills registry at clawhub.com)
• Workspace skills — custom skills you write yourself

# The agent can even search and install skills automatically
# when ClawHub is enabled

Customizing Personality (SOUL.md)

One of OpenClaw's most powerful features is the ability to deeply customize its personality. Through SOUL.md, you can define:

• How the assistant communicates (formal vs. casual)
• Domain expertise (e.g., "You are a senior TypeScript developer")
• User preferences it should remember
• Specific behaviors and rules to follow

This creates a truly personal assistant that learns and adapts to your workflow over time.

---

Deep Dive: Multi-Agent Routing & Sessions

Session Model

OpenClaw uses a sophisticated session model:

• Main session — your direct 1:1 conversation with the AI
• Group sessions — isolated per-group contexts with activation modes (mention-only or always-on)
• Channel sessions — separate contexts per messaging platform

Agent-to-Agent Communication

The sessions_* tools enable coordination between sessions:

# Discover active sessions
sessions_list

# Fetch transcript logs from another session
sessions_history

# Send a message to another session (with optional reply-back)
sessions_send

This is particularly powerful for multi-agent workflows — for example, having a research agent pass findings to a writing agent, all coordinated through your messaging channels.

Multi-Agent Routing

You can route different channels, accounts, or contacts to completely separate agents, each with their own workspace, skills, and model configuration. This enables scenarios like:

• A work agent connected to Slack with access to your company tools
• A personal agent on WhatsApp managing your home automation
• A development agent on Discord with full system access

---

Deep Dive: Security Model

Default Behavior

By default, OpenClaw tools run on the host for the main session, meaning the AI has full access to your system. This is by design for single-user setups where you trust the AI.

DM Pairing (Inbound Security)

For messaging channels, OpenClaw uses a pairing system to prevent unauthorized access:

{
  "channels": {
    "telegram": {
      "dmPolicy": "pairing"
    }
  }
}

When an unknown sender messages the bot, they receive a pairing code. You approve them with:

openclaw pairing approve telegram ABC123

Sandbox Mode for Groups

For group chats and channel sessions, you can enable Docker sandboxing:

{
  "agents": {
    "defaults": {
      "sandbox": {
        "mode": "non-main"
      }
    }
  }
}

This runs non-main sessions inside per-session Docker containers, restricting bash execution to within the container. The sandbox defaults allow safe tools (read, write, edit, sessions) while denying dangerous ones (browser, canvas, cron).

macOS Permissions

The macOS app integrates with Apple's TCC permission system:

• system.run — runs local commands (requires explicit screen-recording permission for certain operations)
• system.notify — posts user notifications (fails gracefully if denied)
• Camera, Canvas, Screen Recording — all routed through node.invoke and respect TCC permission status

---

Advanced Configuration

Minimal Configuration

OpenClaw's configuration lives in ~/.openclaw/openclaw.json:

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

Remote Gateway Setup

You can run the Gateway on a remote Linux server and connect clients over Tailscale or SSH:

{
  "gateway": {
    "tailscale": {
      "mode": "serve"
    }
  }
}

Options:

• off — no Tailscale automation (default)
• serve — tailnet-only HTTPS via Tailscale Serve
• funnel — public HTTPS via Tailscale Funnel (requires password auth)

Chat Commands

Control the assistant directly from any connected chat:

Command	Description
/status	Session status (model, tokens, cost)
/new or /reset	Reset the session
/compact	Compact session context
/think <level>	Set thinking level (off/minimal/low/medium/high/xhigh)
/verbose on∣off	Toggle verbose mode
/usage off∣tokens∣full	Per-response usage footer
/restart	Restart the gateway (owner-only)
/activation mention∣always	Group activation mode

---

Companion Apps

macOS App

A native menu bar app providing:

• Gateway control and monitoring
• Voice Wake and Push-to-Talk
• Talk Mode overlay for hands-free conversation
• WebChat access and debug tools
• Remote gateway control

iOS Node

• Live Canvas rendering
• Voice Wake and Talk Mode
• Camera access for visual tasks
• Screen recording
• Bonjour-based device pairing

Android Node

• Connect via setup code or manual configuration
• Chat sessions with full agent access
• Voice tab for spoken interaction
• Canvas rendering
• Camera, screen recording, and device commands (notifications, location, SMS, photos, contacts, calendar, motion data)

---

Real-World Use Cases

Based on community feedback from Reddit, blogs, and forums, here are the most popular use cases:

1. Personal Productivity Butler

Users configure OpenClaw as an "always-on assistant" that manages their inbox, schedules meetings, and generates daily summaries — all through WhatsApp or Telegram messages.

2. Development Assistant

Developers use it as a coding companion connected to Discord, with skills for code review, PR analysis, and automated testing. The browser tool enables automated web scraping for research.

3. Smart Home Controller

With cron jobs and webhooks, users automate home routines — turning on lights, adjusting thermostats, and sending notifications based on schedules or events.

4. Content Creation Pipeline

Writers and marketers use OpenClaw to research topics, draft content, and manage editorial calendars, all orchestrated through conversational commands.

5. 24/7 Autonomous Agent

Power users run OpenClaw on dedicated hardware (often a Mac Mini or a Linux VPS) to monitor emails overnight, process incoming requests, and even build things based on recurring context.

---

OpenClaw vs Alternatives

Feature	OpenClaw	ChatGPT	Nanobot	Jan.ai	SuperAGI
Self-hosted	✅	❌	✅	✅	✅
Multi-channel	22+	1 (web)	8+	1 (app)	API-based
System access	Full	❌	Limited	❌	Sandboxed
Voice support	✅	✅	❌	❌	❌
Model agnostic	✅	GPT only	✅	✅	✅
Skills ecosystem	✅ ClawHub	GPTs	❌	❌	✅
Stars	260K	N/A	~5K	~25K	~38K
Language	TypeScript	N/A	Python	TypeScript	Python

When to choose OpenClaw:

• You want a single AI assistant across ALL your messaging platforms
• Privacy and data ownership are priorities
• You need deep system integration (shell, files, browser)
• You want proactive automation (cron, webhooks, email hooks)

When to choose alternatives:

• You want zero-setup simplicity → ChatGPT
• You need a lightweight, auditable solution → Nanobot (4K lines of Python)
• You want 100% offline local AI → Jan.ai
• You're building enterprise autonomous agents → SuperAGI

---

Frequently Asked Questions

1. Is OpenClaw free to use?

OpenClaw itself is free and open-source (MIT License). However, you'll need API keys for AI model providers (OpenAI, Anthropic, etc.), which have their own costs. You can also use free/local models via Ollama.

2. Does OpenClaw work on Windows?

Yes, via WSL2 (Windows Subsystem for Linux). Native Windows support is not available, but WSL2 provides a seamless experience.

3. How secure is it to give an AI full system access?

OpenClaw includes multiple safety layers: DM pairing for unknown contacts, Docker sandboxing for group sessions, and configurable tool allowlists/denylists. For personal single-user setups, the default full-access model is convenient. For shared environments, enable sandbox mode.

4. Can I use multiple AI models?

Yes. OpenClaw supports model failover — you can configure primary and fallback models. For example, use Claude for complex reasoning and a cheaper model for simple queries.

5. How do I keep OpenClaw running 24/7?

The onboarding wizard installs a daemon (launchd on macOS, systemd on Linux) that keeps the Gateway running. Alternatively, run it on a remote Linux server with Tailscale access.

6. What's ClawHub?

ClawHub is OpenClaw's skill registry. When enabled, the agent can automatically search for and install new skills as needed, extending its capabilities without manual configuration.

7. Can multiple people use the same OpenClaw instance?

Yes, through multi-agent routing. Different users or channels can be routed to isolated agents with separate workspaces and configurations.

8. How does it compare to running ChatGPT in a browser?

OpenClaw is fundamentally different — it's an agentic assistant that can execute real tasks on your system, while ChatGPT is a conversational interface. OpenClaw can run shell commands, control a browser, manage files, send messages on your behalf, and automate recurring tasks.

---

Conclusion

OpenClaw represents a paradigm shift in how we interact with AI assistants. By meeting you on the channels you already use, running locally on your hardware, and providing deep system integration, it transforms AI from a tool you visit in a browser to a companion that's always by your side.

With 260K+ stars and a thriving community of over 1,000 contributors, OpenClaw is not just a project — it's a movement toward personal, private, and powerful AI. Whether you're a developer looking for a coding companion, a power user wanting to automate your digital life, or someone who simply wants a smarter way to interact with AI, OpenClaw delivers.

The lobster way. 🦞

---

Resources & Links

• GitHub Repository: openclaw/openclaw
• Official Website: openclaw.ai
• Documentation: docs.openclaw.ai
• Skills Registry: ClawHub
• Discord Community: discord.gg/clawd
• DeepWiki: deepwiki.com/openclaw/openclaw
• Getting Started Guide: docs.openclaw.ai/start/getting-started