OpenClaw Setup Guide - Personal AI Assistant for All OS (2026)

OpenClaw (Clawdbot) is an open‑source, personal AI assistant that runs on your own devices using a local‑first Gateway. Instead of relying solely on cloud dashboards, you control the gateway that connects AI models to the apps you already use.

It acts like a Jarvis‑style AI butler that understands your context, remembers your preferences, and can call tools on your machines to automate work. The key difference: you own the infrastructure and control where your data lives.

You can talk to OpenClaw through multiple messaging surfaces—WhatsApp, Telegram, Discord, Slack, Signal, iMessage (on macOS), and a built‑in WebChat UI—using a single unified assistant.

Core Features

Local‑First Gateway: A single control plane for sessions, providers, tools, and events across all surfaces.
Multi‑Surface Messaging: WhatsApp, Telegram, Slack, Discord, Signal, iMessage, and WebChat in one assistant.
Persistent Memory: Remembers sessions and can maintain long‑term context tailored to you.
Browser Control: Web browsing, form filling, and data extraction via a dedicated browser tool.
System Access: Read/write files, run shell commands, and execute scripts—optionally in sandboxed containers.
Voice Capabilities: Voice and “talk mode” on macOS, iOS, and Android via companion apps and TTS providers like ElevenLabs.
Extensible Skills: Built‑in and community skills, plus the ability to write your own with TypeScript/JavaScript.
Live Canvas: Visual workspace/Canvas in the Gateway UI for rich, agent‑driven output and interaction.
Cross‑Platform: CLI and gateway for Windows (via Node/WSL), macOS, and Linux; optional macOS/iOS/Android companions.

Why Choose OpenClaw?

Privacy and Ownership: OpenClaw is designed as local‑first: the Gateway runs on hardware you control, and by default binds to loopback (localhost). This means your conversations, credentials, and tools stay under your control unless you explicitly expose the Gateway.
Deep Customization: You can adjust OpenClaw’s behavior at multiple levels: model choice, prompts and workspace files, tool permissions, provider policies, and even the Gateway’s source code if you build from the repo.
Seamless Integration: Through skills and providers, OpenClaw can connect to many services you already use—email, project trackers, infrastructure APIs, and smart home gateways—so you orchestrate everything via conversation instead of juggling dashboards.
Cost Efficiency: You reuse existing AI subscriptions or API keys (Anthropic Claude, OpenAI, etc.) instead of paying for yet another SaaS UI. The Gateway coordinates how often and when models are called so you stay in control of usage.
Automation‑First Design: OpenClaw emphasizes tool use and skills: instead of being just a chat box, it is built to call scripts, HTTP APIs, browser controllers, and system tools in a structured way. That makes it suitable for inbox triage, DevOps, content workflows, and more.
Developer‑Friendly
- TypeScript/Node.js stack
- CLI‑driven onboarding (openclaw onboard)
- Clear docs for CLI, Gateway, and providers
- Fits local dev environments, servers, and containers.

System Requirements

Requirement	Minimum	Recommended
Runtime	Node.js 22+	Latest Node.js 22 LTS
RAM	4 GB	8 GB+
Disk Space	~2 GB	5–10 GB (for caches, browser)
Network	Stable internet	Low latency for AI API calls
AI Provider	Any valid API key	Anthropic Claude / OpenAI APIs

Supported Operating Systems

Windows 10/11 (Node.js + CLI; WSL2 strongly recommended in official docs).
macOS 11+ (Intel & Apple Silicon; mac app available via docs).
Linux (Ubuntu/Debian and similar distros).
Docker / containers (for advanced/remote/self‑hosting scenarios).

Recommended Installation (CLI + Wizard)

The recommended way to set up OpenClaw is:

Install Node.js 22+
Install the OpenClaw CLI from npm
Run the onboarding wizard from the CLI.

1. Install Node.js and Git

See platform guides at nodejs.org and git-scm.com.

Example (macOS with Homebrew):

brew install node git
node --version
npm --version

2. Install OpenClaw CLI

Global install (simplest):

npm install -g openclaw
# or
pnpm add -g openclaw
# or
bun add -g openclaw

Check:

openclaw --help

3. Run the Onboarding Wizard

Run the wizard once to configure everything:

openclaw onboard

The CLI wizard (documented in Wizard - OpenClaw) will:

Configure local or remote Gateway (port, bind, auth, Tailscale, etc.)
Set up model/auth (Anthropic API key is recommended; OpenAI also supported).
Choose and bootstrap a workspace directory (default ~/clawd).
Enable providers (Telegram, WhatsApp, Discord, Signal, etc.).
Optionally install a daemon (LaunchAgent/systemd) so Gateway runs in the background.
Suggest recommended skills and run health checks.

You can later re‑run:

openclaw configure

to tweak config.

Optional: Install from Source (Developers)

If you want to read or modify the source:

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

Install dependencies:

npm install
# or
pnpm install
# or
bun install

Build the core and UI:

npm run build
npm run ui:build

Then either:

Use npx openclaw onboard, or
Install the npm package globally and use openclaw ... as above.

Starting the Gateway

Once onboarding is complete, you can start the Gateway with:

openclaw gateway --port 18789

By default it binds to 127.0.0.1, exposing:

Gateway WebSocket on ws://127.0.0.1:18789
Control UI / Canvas on http://127.0.0.1:18789/ui
WebChat UI on http://127.0.0.1:18789/chat.

You can also install a daemon (via wizard or docs) to keep it running in the background.

Connecting Messaging Providers (Multi‑Surface)

Official docs provide a WhatsApp provider using a web session (Baileys).

Typical flow:

Ensure WhatsApp provider is enabled in config (wizard can do this).
Log in:

openclaw providers login

Scan the QR code with your WhatsApp app (Linked Devices).
Optionally, restrict who can DM the bot in ~/.openclaw/openclaw.json:

{
  "whatsapp": {
    "dmPolicy": "allowlist",
    "allowFrom": ["+1234567890"]
  }
}

Create a bot via @BotFather and copy the token.
Add token in config or environment, e.g.:

{
  "telegram": {
    "enabled": true,
    "botToken": "123456:ABCDEF..."
  }
}

Start the Gateway:

openclaw gateway --port 18789

Message your bot on Telegram.

Discord / Slack / Others

Providers such as Discord and Slack are configured in the same style:

Set tokens via env or config (DISCORD_BOT_TOKEN, SLACK_BOT_TOKEN, etc.).
Ensure provider sections exist in openclaw.json per docs.
Invite/install the bot into your workspace/server.

WebChat

Once the Gateway is running, open:

http://127.0.0.1:18789/chat

for a browser‑based chat interface with file uploads and Canvas rendering.

Connecting to AI Models

Anthropic (Claude) – Recommended

Get an API key or subscription at Anthropic’s console.
Set:

export ANTHROPIC_API_KEY="sk-ant-..."

Choose a model such as claude-3.5-sonnet or a newer Opus variant in the wizard or config.

Example config snippet:

{
  "agent": {
    "model": "anthropic/claude-3.5-sonnet"
  },
  "providers": {
    "anthropic": {
      "apiKey": "sk-ant-..."
    }
  }
}

OpenAI (ChatGPT)

Get API key from OpenAI.
Set:

export OPENAI_API_KEY="sk-..."

Pick a model like openai/gpt-4o in wizard or config.

Local Models (Advanced)

OpenClaw’s core docs focus on cloud providers, but you can route requests to local LLM runtimes such as Ollama using custom skills or tools. This usually involves:

Running a local LLM server (e.g., Ollama on http://127.0.0.1:11434).
Writing a skill/tool that calls that endpoint.

Because this is an advanced/custom setup, refer to skill and tools docs instead of assuming a built‑in local/llama2 model string in config.

Example Uses

Once configured, you can:

Chat from WhatsApp: “Summarize my last 10 important emails and create a to‑do list.”
Use Telegram to manage infra: “Check the health of my servers and restart the staging API if it’s down.”
From WebChat: “Research competitive pricing and draft a proposal.”
From Discord: “Generate a weekly status report from these GitHub issues and send it to this channel.”

These kinds of workflows are typically implemented via skills that call Gmail, GitHub, infrastructure APIs, or the browser tool, based on your configuration.

Essential CLI Commands (Typical Patterns)

Actual command names/flags evolve, but common patterns from the current CLI docs include:

openclaw onboard – Run the onboarding wizard.
openclaw configure – Reconfigure workspace, providers, or gateway.
openclaw gateway --port 18789 – Start the Gateway on a given port.
openclaw status – Show Gateway/daemon/agent/session status (recent releases).
openclaw doctor – Run diagnostics for common config/environment issues.

Use openclaw --help and docs at docs.clawd.bot/cli for updated subcommands and options.

Troubleshooting Highlights

Typical issues and checks:

Port already in use:
Change port via openclaw gateway --port 18790 or stop conflicting service.
Providers not responding:
Verify tokens and enabled flags in openclaw.json or environment.
Check provider‑specific docs (e.g., WhatsApp provider doc for QR login).
High latency / timeouts:
Check network connectivity and API rate limits.
Try a faster model (e.g., Claude Sonnet vs. Opus) for interactive tasks.
Skills not loading:
Ensure skill folder layout matches examples in docs and that dependencies are installed.

Official Resources

For the latest and most accurate information, always cross‑check with:

Website: https://openclaw.ai
Docs index: https://docs.openclaw.ai
GitHub repo: https://github.com/openclaw/openclaw