March 9, 2026

How to Install OpenClaw: Complete Guide for Mac & Windows (2026)

Step-by-step instructions to install OpenClaw on Mac or Windows. From zero to running AI agent in under 5 minutes. Includes Telegram, WhatsApp, and WeChat setup.

This guide walks you through installing OpenClaw from scratch on macOS or Windows. By the end, you’ll have a running AI agent connected to a messaging channel and your first skills installed.

Using WeChat or QQ as your primary channel? The raw OpenClaw installer does not include native WeChat or QQ support. If that’s your setup, use QClaw instead — Tencent’s one-click launcher handles the WeChat binding automatically.

Prerequisites

Before you start, confirm the following:

macOS 12 (Monterey) or later, or Windows 10/11 64-bit
An AI model API key from one of the supported providers: Anthropic (Claude), OpenAI, DeepSeek, Moonshot (Kimi), or any OpenAI-compatible endpoint. OpenClaw itself is free; you pay only for the model calls you make.
A Telegram account (easiest channel to start with) or a WhatsApp Business account (more setup required)
Internet connection during installation. After setup, the agent runs locally and only calls out for AI inference and tool-specific requests.

You do not need Node.js, Python, or any other runtime pre-installed. The OpenClaw installer bundles everything it needs.

Step 1: Run the Installer

Open a terminal (Terminal.app on Mac, Git Bash or PowerShell on Windows) and run:

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

What this does:

Downloads the latest stable OpenClaw release for your platform and architecture (Apple Silicon, Intel Mac, or Windows x64)
Installs the openclaw CLI and clawhub CLI to /usr/local/bin (Mac) or %USERPROFILE%\.openclaw\bin (Windows)
Creates your agent config directory at ~/.openclaw/
Registers OpenClaw as a background service that starts on login

The installer typically completes in 30–90 seconds depending on your connection speed.

After installation, verify it worked:

openclaw --version

You should see output like OpenClaw 1.4.2 (stable).

Step 2: Configure Your AI Model

OpenClaw’s configuration lives at ~/.openclaw/config.yaml. Open it in any text editor. It will look like this by default:

# OpenClaw Configuration
# Full reference: https://openclaw.ai/docs/config

agent:
  name: "My Agent"
  soul_file: "~/.openclaw/SOUL.md"

model:
  provider: ""       # Set this: anthropic | openai | deepseek | kimi | custom
  api_key: ""        # Your API key for the chosen provider
  model_id: ""       # The specific model to use

channels: []         # Configured in Step 3

skills_dir: "~/.openclaw/skills/"
log_level: "info"
allow_shell: false   # Set to true only if you need shell command execution

Fill in the model section. Examples for each supported provider:

Anthropic (Claude Sonnet 4.6 — recommended):

model:
  provider: "anthropic"
  api_key: "sk-ant-..."
  model_id: "claude-sonnet-4-6"

OpenAI (GPT-4o):

model:
  provider: "openai"
  api_key: "sk-..."
  model_id: "gpt-4o"

DeepSeek:

model:
  provider: "deepseek"
  api_key: "sk-..."
  model_id: "deepseek-chat"

Kimi (Moonshot AI):

model:
  provider: "kimi"
  api_key: "sk-..."
  model_id: "moonshot-v1-8k"

Custom OpenAI-compatible endpoint:

model:
  provider: "custom"
  base_url: "https://your-endpoint.example.com/v1"
  api_key: "your-key"
  model_id: "your-model-name"

Save the file after editing.

Step 3: Connect a Messaging Channel

You need at least one channel configured before your agent can receive messages.

Option A: Telegram (Recommended for First-Time Setup)

Telegram is the fastest channel to set up because bot creation is instant and free.

Open Telegram and search for @BotFather
Send /newbot
Follow the prompts — give your bot a name and a username (must end in bot, e.g., myagent_bot)
BotFather will reply with a bot token that looks like 7123456789:AAH...
Copy that token and add the following to your config.yaml:

channels:
  - type: "telegram"
    bot_token: "7123456789:AAH..."

Restart the OpenClaw service:

openclaw restart

Open Telegram, find your new bot by its username, and send it a message to verify.

Option B: WhatsApp

WhatsApp setup requires a Meta Business account and a verified phone number. This process typically takes 10–30 minutes.

Go to developers.facebook.com and create a Meta app with the WhatsApp product
Set up a WhatsApp Business phone number (you can use a free test number for development)
Note your Phone Number ID and Access Token
Add to config.yaml:

channels:
  - type: "whatsapp"
    phone_number_id: "1234567890"
    access_token: "EAAx..."
    verify_token: "choose-any-string"
    webhook_url: "https://your-public-url/openclaw/webhook"

WhatsApp requires a publicly accessible webhook URL, which means you’ll need to either (a) deploy OpenClaw behind a reverse proxy with a public domain, or (b) use a tunneling tool like ngrok during development. For most personal-use setups, Telegram is significantly simpler.

Step 4: Test Your Agent

With a channel connected, send your agent a message:

Hello, what can you do?

A healthy agent will reply with a summary of its current capabilities based on your installed skills and SOUL.md configuration. The first response may take 3–5 seconds as the model warms up.

If you get no response within 30 seconds, check the logs:

openclaw logs --tail 50

Step 5: Install Your First Skills

A fresh OpenClaw install has no skills — it can answer general questions but can’t take actions yet. Install your first skills with the clawhub CLI:

clawhub install web-search
clawhub install gmail
clawhub install calendar

After installing, send your agent a command that exercises the new skill:

Search the web for the latest OpenClaw release notes

Check my Gmail inbox and summarize the 5 most recent unread emails

What’s on my calendar for this week?

Skills install instantly and take effect without restarting the agent.

To see all installed skills:

clawhub list

To browse the full ClawHub registry:

clawhub search [keyword]

Or visit clawhub.io in a browser for a searchable web interface.

Troubleshooting

Agent not responding to messages

Check that the service is running: openclaw status
If stopped, start it: openclaw start
Verify your channel config is correct in config.yaml — bot tokens and API keys are the most common source of errors
Check logs for error messages: openclaw logs --tail 100

API errors in logs

Confirm your API key is correct and has available credits
Check that the model_id matches exactly what your provider expects (these are case-sensitive)
If you see rate limit errors, consider switching to a model with higher rate limits

Can’t find config.yaml

The config file is always at ~/.openclaw/config.yaml. On Windows, ~ resolves to C:\Users\YourUsername\. If the file doesn’t exist, run openclaw init to regenerate it.

Skill not triggering

Make sure the skill installed successfully: clawhub list
Skills are triggered by the AI model deciding the skill is relevant, not by exact keyword match. If a skill isn’t being used, check your SOUL.md to ensure it doesn’t restrict the agent’s scope
Try asking explicitly: “Use the gmail skill to check my inbox”