GPT Plus - ChatGPT Plus/Pro Top-up Service

openclaw Step-by-Step Installation Tutorial: Locally Deploy an AI Agent with Root Privileges + ChatGPT OAuth Keyless Configuration Guide

1/28/2026 openclaw installationopenclaw tutoriallocal AI Agent deploymentChatGPT Business discountClaude OAuth loginNode.js environment setupautomated AI assistant

Recently, openclaw has completely blown up in the tech community! The hype is reminiscent of last year's Manus launch, exploding across the internet almost overnight.

It's worth noting that openclaw has now been officially renamed to openclaw, and the logo changed from a whole lobster to a "molted" lobster.

💡 About the naming: openclaw means "molting", taken from the molting behavior of lobsters. This not only retains the original lobster element but also cleverly avoids trademark conflict with Claude.

# What is openclaw? What can it do?

openclaw isn't just a chatbot — it's an AI Agent that can be deployed on a local computer or server.

In plain terms, you can think of it as a super assistant with system superuser (Root) privileges. If you're familiar with Claude Code, openclaw is an advanced version that has a rich set of Skills and can call various MCP services to complete complex tasks end to end.

Its core features include but are not limited to:

  • Automatically organizing emails and schedule management
  • Stock market analysis and operations
  • Automated tweet writing and social media management
  • Reading and summarizing PPT/PDF documents
  • Runs 24/7 in the background, always on standby

⚠️ High-risk warning (must read): Because openclaw has extremely high system privileges (Shell access), please operate with extreme caution! A wrong command could accidentally delete critical system files. In addition, openclaw's gateway is exposed to the public internet by default and lacks strict authentication, so security is a challenge. We strongly recommend deploying it in a virtual machine (VM) or on a spare Mac to prevent hacker intrusion.

# Step 1: Environment preparation and installation requirements

Before installing openclaw, make sure your computer meets the following requirements:

  • Node.js version: >= 22.0
  • Git: must be installed

System recommendations:

  • Windows users: we strongly recommend WSL2 (Ubuntu recommended) for a better experience.
  • Mac users: just use the terminal directly — works out of the box.

# Step 2: openclaw installation tutorial

The installation process is very simple — just copy the appropriate Shell command for your operating system and install it in one go.

# 1. macOS / Linux / Ubuntu / Debian install command (recommended)

Open the terminal and enter:

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

# 2. Windows (PowerShell) install command (recommended)

Open PowerShell and enter:

iwr -useb https://openclaw.ai/install.ps1 | iex

# 3. Install using the npm package manager (alternative)

Note: the npm version was just released by the official team and may be unstable; we recommend using the curl/iwr scripts above first.

npm i -g openclaw@latest

Tip: if you can't find the openclaw command after installing via npm, try using the openclaw command.

# Step 3: Hands-on deployment demo (using Windows as an example)

The following demo is based on Node.js 22.15.

After running the install command, the system will automatically clone the GitHub repository.

Downloading openclaw

npm i -g openclaw@latest

# 1. Risk confirmation and initialization

After installation, the system will warn you about openclaw's high risk.

  • Enter yes to continue.
  • Select QuickStart mode to load the default configuration.

openclaw assigns port 18789 by default.

# 2. Configure AI model authorization (OAuth)

One of openclaw's highlights is its support for ChatGPT and Claude OAuth authorization login. This means you don't need to buy expensive API Keys — you can drive the AI directly with your web membership privileges (such as Plus, Team, Pro).

🛑 Pitfall avoidance guide: Try to avoid using Claude authorization to log in, because Claude's official risk control is strict and may restrict calls that aren't from Claude Code. If you must use Claude but don't have an account, you can check here: Claude account channels (opens new window)

Recommended option: use OpenAI (Codex) authorization 3.png

As long as you're a ChatGPT Plus, Business, or Pro member, you can use it directly.

For convenience in this demo, we select the 2nd authorization method (browser authorization) in the configuration interface. 4.png

The terminal will generate a ChatGPT authorization link, your browser will open automatically, and you just log in with your member account. 5.png

Here I'm using a ChatGPT Business membership.

  • Reason: Business membership currently offers excellent value — for just a few dozen yuan you get the same Codex quota as a Plus member, making it perfect for running token-hungry Agents like this.

💰 Money-saving tip: For a cost-effective openclaw experience, we recommend using a ChatGPT Business account. Discounted sign-up address: https://gptplus.org.cn (opens new window)

6.png

Click "Continue", and once authorization succeeds, go back to the terminal to check the status. 7.png

PS: If the terminal doesn't automatically capture the Token, manually copy the full callback address starting with localhost from your browser's address bar and paste it into the terminal. 8.png

# 3. Choose a model and channel

The default is the GPT-5.2-Codex model. You can skip it or configure other channels. 9.png

# 4. Configure Skills

This is the soul of an AI Agent. 10.png

Choose a package manager (the default npm is fine). The system will recommend some basic Skills; you can pick what you need or skip them all. The coolest part: later, when you need a feature, just tell openclaw in the chat box "help me install such-and-such skill", and it will install it automatically — no manual work needed! 11.png

If you come across API Key settings (such as for image generation services) and don't have one, just press Enter on No to skip.

# 5. Enable Hooks (automation hooks)

Finally, the system will ask whether to enable the built-in hook scripts (Hooks). We recommend enabling them all for the complete experience: 12.png

  • boot-md: automatically executes the bootstrap instructions in BOOT.md after the gateway starts.
  • command-logger: audit log that records every command executed (essential for security auditing).
  • session-memory: conversation memory — automatically saves a snapshot of the current context when you enter /new.

# Step 4: Launch the Web UI console

For beginners, we strongly recommend the Web UI mode — a visual web interface is friendlier than a pure command line.

Once configured, run the following command to launch the console:

openclaw dashboard
  • Local access address: http://127.0.0.1:18789

📝 Important note: Although the tool has been renamed to openclaw, some old commands haven't been fully updated yet. If the openclaw command doesn't work, try replacing it with openclaw to run — this usually solves the problem.

Now, your dedicated AI super-employee is ready! If you don't have an OpenAI or Claude account, you can also bind a third-party API Key in the configuration and drive openclaw with a domestic large model.

Go explore the powerful capabilities of openclaw!