OpenClaw (Lobster) is the open-source AI agent project with the fastest GitHub growth rate right now. It lets you set up a private, always-on AI assistant on your own computer, and chat with it through the communication apps you already use daily, such as Telegram, WhatsApp, Discord, and more.
This tutorial starts from scratch and walks you through installation, setup, and sending your first message—about 10 minutes in total.
What is OpenClaw? How is it different from ChatGPT?
ChatGPT is a Q&A tool. You ask it a question, it gives you an answer, and when the conversation ends, it’s over. OpenClaw is an AI agent: you give it a task, and it thinks for itself, calls tools, and performs actions—checks your Gmail, looks at your calendar, searches the web, writes code—then tells you the results.
The more critical difference is that OpenClaw runs on your own computer, with your data fully under your control. Nothing is uploaded to any third-party servers. It doesn’t include an AI model itself; you need to provide your own API Key (Anthropic Claude, OpenAI GPT, Google Gemini, etc. all work). The model costs are paid directly by you to the provider.
Pre-installation prep
System requirements
macOS, Linux, or Windows (recommended: use WSL2 for better stability)
Node.js 22.14 or higher (recommended: Node 24)
At least one set of AI model API Key(s)
Install Node.js
First, confirm whether it’s installed and the version is sufficient:
node -v
If your version is lower than v22.14 or it’s not installed yet, choose an installation method based on your system:
Get an AI model API Key
OpenClaw supports multiple model providers. Below are the three most commonly used options:
Anthropic Claude: Go to console.anthropic.com to create an account, then add a key on the API Keys page. Claude Sonnet 4.x is currently the most popular choice.
OpenAI GPT: Go to platform.openai.com and create an API Key.
Google Gemini: Go to aistudio.google.com to get an API Key for free.
Copy the API Key and keep it handy—you’ll need to paste it in during installation.
Step 1: Install OpenClaw
On macOS and Linux, you only need one command:
curl -fsSL | bash
Windows users run in PowerShell:
iwr -useb | iex
After installation, confirm the command works:
openclaw --version
Step 2: Run the Onboarding wizard
Run the following command to start the interactive setup wizard, and also have OpenClaw start with the system as a background service:
openclaw onboard --install-daemon
The wizard will ask a few questions in sequence:
Choose a model provider: select the one you just got an API Key from (Anthropic, OpenAI, Google, etc.)
Paste API Key: paste the key you copied earlier
Choose a model: it’s recommended to pick the newest main model (such as claude-sonnet-4-5)
Choose an installation mode: beginners can choose QuickStart; it will use default values to complete all settings
The whole process takes about 2 minutes. For most options, just press Enter to accept the defaults.
Step 3: Confirm the Gateway is running
openclaw gateway status
If you see that Gateway is listening on port 18789, the installation was successful.
Open the browser control interface to test:
openclaw dashboard
The browser will open the Control UI. Here, you can send messages directly to the AI assistant to confirm everything is working properly.
Step 4: Connect Telegram (the fastest way to get started)
Telegram is the most fully supported communication channel by OpenClaw, and the setup is the simplest—you only need a single Bot Token.
4-1 Apply for a Bot Token with BotFather
Open Telegram, search for and open @BotFather (confirm there’s an official blue checkmark next to it)
Send /newbot
Follow the prompts to enter the Bot name (display name) and Bot username (must end with bot, e.g., my_ai_assistant_bot)
BotFather will give you a Token in a format like 123456789:AAGxxxxxxxxxxxxxxxxx. Copy it.
4-2 Set up the Token
Edit the OpenClaw config file (default path
~/.openclaw/config.json5
), and add the Telegram settings:
{ channels: { telegram: { enabled: true, botToken: “Paste your Token here”, dmPolicy: “pairing”, }, }, }
4-3 Restart the Gateway and pair
openclaw gateway restart openclaw pairing list telegram
Then in Telegram, find the Bot you just created and send it any message. It will give you a pairing code. Run:
openclaw pairing approve telegram
After that, when you message your Bot in Telegram, you’ll receive a reply from the AI assistant.
Frequently Asked Questions (FAQ)
What if I get a Node version error?
Run
node -v
to confirm the version. If it’s lower than v22.14, reinstall Node.js according to the table above. If you’re on macOS and you have Homebrew installed, run brew upgrade node to upgrade.
Can’t start the Gateway?
Run
openclaw gateway status
to view the error message. The most common causes are port 18789 being in use, or an incorrect API Key format. You can run openclaw configure to reconfigure.
How do I change the API Key?
Run
openclaw configure
, and then choose the Model/Auth step to reset.
Will the Token burn continuously?
By default, OpenClaw calls the API only when it receives a message, so it won’t consume the Token without reason. But if you enable certain automated Skills or scheduled tasks, it’s recommended to set a monthly usage cap in the model provider’s backend to avoid unexpected overspending.
What you can do next
Connect Gmail and Google Calendar so the assistant can help you manage emails and schedules
Install Skills to expand the assistant’s capabilities (web search, code execution, memory management, etc.)
Cloud deployment: keep Lobster online 24/7 via Cloudflare Tunnel (advanced tutorial, coming soon)
This article “OpenClaw (Lobster AI) — Complete Installation Tutorial 2026: Set up a personal AI assistant from scratch and connect Telegram” first appeared on 鏈新聞 ABMedia.
