From Zero to OpenClaw: Workshop Walkthrough (Recording)

What You'll Build

By the end of this walkthrough, you'll have a working OpenClaw AI agent that you can talk to via Telegram and that can control a web browser on your behalf. The workshop uses a LinkedIn prospector as the example, but once the browser is connected your agent can do any browser-based workflow.

LinkedIn disclaimer: LinkedIn does not love bots accessing their platform. There's a small chance your account gets temporarily blocked. If LinkedIn is mission-critical for you, consider using a secondary account.

Prerequisites

• A Mac running virtualized Ubuntu from the pre-work, or your own Linux setup
• A Telegram account (phone, desktop, or both)
• Either an Anthropic (Claude) or OpenAI (ChatGPT) account with a paid subscription
• About 30-45 minutes of uninterrupted time

Step 1: Install Chromium

Open a terminal in your virtual machine and install Chromium. This is the browser your agent will use.

sudo apt install chromium-browser

Step 2: Create Your Telegram Bot

1. 1. Open Telegram and search for @BotFather (verified checkmark).
2. 2. Send /newbot to BotFather.
3. 3. Give your bot a display name.
4. 4. Give your bot a username that ends in bot.
5. 5. Save the API token BotFather gives you.

Step 3: Install Claude Code (Anthropic Users Only)

Skip this step if you're using OpenAI/ChatGPT.

brew install claude-code
claude setup-token

Complete the authorization flow and keep the resulting Claude API token for onboarding.

Step 4: Install OpenClaw

Install OpenClaw globally and pin a known stable version:

npm install -g openclaw@2026.3.1

Version 2026.3.2 had issues in the workshop. Pinning to 2026.3.1 keeps this setup aligned with the recording.

Step 5: Run OpenClaw Onboarding

openclaw onboard

If using Claude (Anthropic):

1. 1. Select Anthropic as provider.
2. 2. Select Anthropic Token auth.
3. 3. Paste the token from Step 3.
4. 4. Set a name (or leave blank).
5. 5. Choose Claude Sonnet as model.

If using ChatGPT (OpenAI):

1. 1. Select OpenAI.
2. 2. Select OpenAI Codex ChatGPT OAuth.
3. 3. Complete login and 2FA in the browser flow.
4. 4. Choose a non-Codex model (for example GPT 5.2) unless you explicitly want a coding-focused agent.

Step 6: Connect Telegram

1. 1. Select Telegram as communication channel.
2. 2. Paste your BotFather API token from Step 2.
3. 3. Save the generated bot link so you can message your agent.

Step 7: Configure Skills and Hooks

Choose ClawHub for Skills and use npm as install method.

For extra API keys, choose No to all prompts in this walkthrough.

Enable all of these hooks:

• Boot MD
• Bootstrap Extra Files
• Command Logger
• In-Session Memory

For interaction options, choose "Do this later".

Step 8: Pair Your Telegram Chat

1. 1. Message your bot in Telegram with Hi.
2. 2. It will return a pairing command like openclaw pairing approve telegram XXXXXXXX.
3. 3. Copy that full command and run it in your terminal.

This locks the agent so it only responds to your Telegram account.

Step 9: Wake Up Your Agent

Send your bot another message. It will ask for identity details (your name, its name, and optional personality guidance).

"I am Phil, your name is Lady Botlington The II. I'll let you pick your first name."

First replies can take up to 30 seconds. That delay is normal.

Step 10: Set Up the Browser Connection

On ARM Macs running Ubuntu, Snap Chromium can block normal OpenClaw launch behavior. Ask the agent for options:

"I need you to do some LinkedIn prospecting for me, but we have a challenge. You're installed on Ubuntu and the Chromium browser is installed via Snap, which has some restrictions that will cause problems for you accessing it. Can you give me some ideas on how to manage that?"

You'll usually get:

• Option A: Install Chrome (does not work on ARM Linux)
• Option B: Relaunch Chromium with modified settings

Tell the agent to proceed with Option B:

"Let's go with Option B - there's no Chrome version for ARM. Can you run the command yourself and remember it for the future?"

Step 11: Log In to LinkedIn

1. 1. Open LinkedIn in the VM browser.
2. 2. Sign in with your LinkedIn credentials.
3. 3. The agent can now browse LinkedIn on your behalf.

Tip: if risk is a concern, consider a separate LinkedIn account for agent work.

Step 12: Start Prospecting

Message your agent with an outcome-based request, for example:

"I'm running a course about AI for non-technical people. Can you use your own logic to decide who to target, and find me 10 potential leads?"

The agent will usually:

1. 1. Analyze your offer and target audience
2. 2. Search LinkedIn and inspect profiles
3. 3. Return a ranked list of leads with profile links

Ongoing Usage

If your agent stops responding after a reboot, run:

openclaw gateway restart

You do not need to manually open LinkedIn each time. The agent can launch and manage the browser. It may occasionally ask you to run terminal commands in the VM (for example, to restart services).