OpenClaw from Scratch 02 | Get It Running First: Install, Onboard, Dashboard, Gateway

"The first time you use OpenClaw, do not set the goal as 'finish every configuration.' Today you only need to do one thing: get it truly running, then hear its first reply in your browser."

Chapter Overview#

This chapter solves one thing only: get your first conversation working from zero.

You will complete these steps:

• Pick the lowest-friction environment to run it in
• Install the CLI
• Run openclaw onboard --install-daemon
• Check whether Gateway is running properly in the background
• Open the Dashboard and complete your first chat

Decide where it should live today#

If this is just your first experience, the easiest path is to get it working on your current computer first.

If you already want it online 24/7, then consider an Ubuntu server.

If you are on Windows, the official docs are very clear:

Use WSL2 first. Do not fight native Windows for this.

Step 1: Install the CLI#

The official recommended install method is:

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

If you are using Windows PowerShell:

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

Once the install finishes, do not start digging through config files yet. Go straight to the onboarding flow.

Step 2: Run the onboarding flow#

openclaw onboard --install-daemon

This step matters because it does more than ask a few questions. It helps you complete all of this in one pass:

• Choose a local or remote Gateway
• Configure models and auth
• Choose whether to connect channels
• Install the background service
• Generate the Gateway token
• Initialize the workspace and base files

What beginners should choose#

If your goal today is only to get the first conversation working, choose like this:

1. Pick local for Gateway
2. Pick Node for the runtime
3. Configure the model and auth first
4. Skip channels for now. We will connect Telegram in the next chapter
5. Choose to install the background service

The official docs call out one thing especially clearly:

Telegram and WhatsApp work better on Node later on. Bun is not the recommended path.

Step 3: Confirm that Gateway is actually running#

Once the wizard finishes, do not rely on a hunch to decide whether it worked.

Check with commands directly:

openclaw gateway status
openclaw status
openclaw health
openclaw security audit --deep

You can think of them as four layers of checks:

• openclaw gateway status: is the background service alive
• openclaw status: does the overall state show any obvious issues
• openclaw health: are critical dependencies such as model auth available
• openclaw security audit --deep: are there obvious security configuration problems

If openclaw health clearly tells you that auth is not configured correctly, that is not a small issue.

Because without model auth, the assistant may be online, but it still cannot answer you.

Step 4: Open the Dashboard#

The easiest way to get your first chat is not Telegram or WhatsApp.

The official docs are direct about this:

The fastest first conversation is through the Dashboard.

openclaw dashboard

If everything is working, it will open the Control UI in your browser.

That is where you say your first sentence to OpenClaw.

What success should look like#

You do not need to guess.

The signs are concrete:

1. The Dashboard opens normally
2. The page connects to your local Gateway
3. You send a message and the assistant replies normally

Start with three simple lines:

• "Who are you?"
• "Describe what you can do right now in three sentences."
• "What should you call me from now on?"

If it does not open, check these first#

1. The Dashboard shows unauthorized or WebSocket 1008#

This is most likely a Gateway token issue.

The simplest official fix is: run openclaw dashboard again, so it opens the entry point with the correct token, or fill gateway.auth.token back into the UI as documented.

2. The wizard finished, but gateway status never came up#

First check whether the background service failed to install, or whether you are running in an unsupported environment.

If you are on Windows, go back to WSL2 first.

3. You are using Bun#

Once you connect Telegram or WhatsApp later, you are likely to hit issues.

Do not try to save this step. Switch back to Node now.

4. You picked remote mode from the start#

Remember this: credentials, workspace files, and config all live on the remote host.

That means a lot of "but I already configured this locally" intuition does not apply in remote mode.

Why we are not connecting Telegram yet#

Because your goal today is not "connect every possible entry point."

Your goal today is this:

Move OpenClaw from an installed state to a state where it can already chat normally.

Once the Dashboard is working, troubleshooting Telegram later becomes much easier.

Because then you already know:

• Gateway is good
• Model auth is good
• The only truly new variable is the channel configuration

That is why we split the learning path apart.

Today's Task#

curl -fsSL https://openclaw.ai/install.sh | bash
openclaw onboard --install-daemon
openclaw gateway status
openclaw status
openclaw health
openclaw dashboard

Today's Achievement#

If you already received the first reply inside the Dashboard, then today you were not merely "reading a tutorial."

You completed your first real hands-on run.

Preview#

In the next chapter, we are not adding more features yet. We are going to make one especially confusing topic completely clear:

What Provider, Model, and Auth each actually mean.

In other words: who is doing the thinking, how you authenticate, and how to choose the default model.

Official References#

• OpenClaw Getting Started Guide
• OpenClaw Onboarding Guide
• OpenClaw Dashboard
• OpenClaw CLI dashboard