Lobster peeking in from the top right
Openclaw setup | Build Sesh with Felix

Getting started with OpenClaw

Use this to get from zero to a first working OpenClaw chat — optimized for free infra, free model access, and the highest odds of getting to a clean first reply fast.

Start with the recommended path Jump to the presenter runbook
🦞 build fast, don’t overcomplicate it
Free infra = your laptop Best no-card path = Qwen OAuth Dashboard-first, then expand however you want Win condition = “say hello”
Primary path
Qwen free-tier OAuth
Surface
Local Dashboard / Control UI
Workshop goal
Boot + first successful reply
First goal
Boot + first successful reply

Why this path

The cleanest route is: run OpenClaw locally on your laptop, use the Dashboard / Control UI in the browser, and authenticate a free model provider.

That avoids bot creation, pairing friction, webhook debugging, and a lot of setup pain.

Interpretation of “free infra” here: your own laptop is the infra. No VPS, no database, no deploy required just to get to a first chat.

Recommended model path

Primary

Qwen OAuth free tier

OpenClaw docs explicitly support qwen-portal via OAuth.

It also notes a documented free tier with 2,000 requests/day, subject to rate limits.

Fallback #1

Gemini API key

Good if you already have access to Google AI Studio without billing friction.

Fallback #2

Hugging Face Inference

Useful as a backup path, but pre-test a known-good model before the workshop.

Recommendation: start with the local dashboard because it is the fastest path to a first working setup.
Lobster rule: get to a working hello first, then add extra claws later — Telegram, channels, skills, and other polish can come after the core setup works.

Track A - recommended event flow

0

What you need

  • macOS, Linux, or Windows with reasonable terminal comfort
  • a browser
  • a Google account you can use to sign into Qwen during OAuth
  • working internet
1

Install OpenClaw

macOS / Linux
curl -fsSL https://openclaw.ai/install.sh | bash
Windows (PowerShell)
iwr -useb https://openclaw.ai/install.ps1 | iex
Security note: OpenClaw is powerful software with access to tools, files, and external services. Treat onboarding seriously, read prompts carefully, and only grant the level of access you are comfortable with.
Fallback if onboarding does not auto-start
openclaw onboard
2

Choose Quick Start

After the install command, OpenClaw onboarding should usually start automatically. When you see the setup flow, choose Quick Start instead of Manual.

If onboarding did not start automatically, run the fallback command above and then choose Quick Start.

3

Choose model/auth provider

At the model/auth provider step, choose Qwen.

Then use your Google account to log in. This is the most straightforward path for getting a free working setup quickly.

4

Finish the wizard with the default choices

Once you choose Qwen and complete login, keep the default model Qwen gives you.

After that, continue through the wizard and keep it simple:

  • skip channel setup/status for now
  • skip web search
  • skip skills
  • skip extra search/provider add-ons unless you specifically need them

The goal is to get through onboarding cleanly and land in a working setup fast.

Fallback: if auth gets interrupted or you need to redo it manually later, use the provider login flow from the CLI — but for the workshop, the intended path is still Quick Start + Qwen + Google login + default model + skip the optional extras.
5

Verify the Gateway and open the Dashboard

CLI
openclaw gateway status
openclaw dashboard

If the browser does not auto-open, try http://127.0.0.1:18789/.

6

First prompt: the “hello” test

Prompt: Hello! Please introduce yourself in one sentence.
Alternative: Say hello and confirm you are working. 🦞

If it answers, the workshop objective is already met.

7

Minimal sanity checks

CLI
openclaw gateway status
openclaw models list
openclaw status

Fallback tracks

Fallback A - Gemini API key

Use this when you already have Gemini access or Qwen OAuth is flaky in your region/account.

Non-interactive example
openclaw onboard --non-interactive \
  --mode local \
  --auth-choice gemini-api-key \
  --gemini-api-key "$GEMINI_API_KEY" \
  --gateway-port 18789 \
  --gateway-bind loopback

Fallback B - Hugging Face Inference

Useful as a second fallback. If you're the one organizing the session, pre-test one model before the event instead of letting the room freestyle their choice.

CLI
openclaw onboard --auth-choice huggingface-api-key

What not to do during the event

Don't overoptimize while starting

The local dashboard is the quickest way to get a working OpenClaw setup. Once that is stable, you can add channels like Telegram right after.

Don’t start with local LLMs

Too many laptop-dependent failure modes, too much downloading, too much debugging.

Don’t optimize for perfection

The win condition is not a fully personalized assistant. The win condition is: it boots and says hello.

Presenter runbook

Slide 1 - Install

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

The installer should usually kick off onboarding automatically.

Slide 2 - Quick Start + Qwen

Choose: Quick Start → Qwen → sign in with Google → keep the default model
Then: skip channel setup, skip web search, skip skills, and finish onboarding
Fallback if onboarding did not start
openclaw onboard

Slide 3 - Open UI

CLI
openclaw gateway status
openclaw dashboard

Slide 4 - Test

Prompt: Hello! Please introduce yourself in one sentence.

Troubleshooting cheatsheet

openclaw: command not found

CLI
node -v
npm prefix -g

Gateway not running

CLI
openclaw gateway status
openclaw gateway start

Qwen model not available

CLI
openclaw models list
openclaw models set qwen-portal/coder-model

Then retry login and restart the Gateway.

Dashboard doesn’t open

CLI
openclaw dashboard

Or manually open http://127.0.0.1:18789/.