Let's Get This Lobster Running

Before we dive in — you're going to want Telegram for this. While OpenClaw supports a bunch of messaging platforms, Telegram is hands-down the best one to use with it. It's free, lightning fast, works on every device, and the OpenClaw integration is buttery smooth. Make sure you have Telegram installed — we'll walk through connecting it in the next chapter.

Alright, you know what OpenClaw is. Now let's actually install it. Don't worry — this is one of those things that looks way scarier than it actually is.

But first, let's talk about where to install it.

Where Should You Run OpenClaw?

OpenClaw is an AI agent that can execute commands, browse the web, manage files, and interact with your messaging apps. That's powerful — and it means you should think about where you run it.

The Safest Way: A Dedicated Machine or VM

Here's the deal — running an AI agent on your daily-driver laptop means it has access to your files, your browser sessions, your everything. For learning, that's fine. But the safest approach is to give OpenClaw its own playground:

Virtual Machine (VM) — The easiest "safe sandbox." Use VirtualBox (free) or UTM (free, great for Mac). Spin up an Ubuntu VM, install OpenClaw there. If anything goes sideways, nuke the VM and start fresh. Your real machine stays untouched.
Spare laptop or Raspberry Pi — Got an old laptop collecting dust? Perfect OpenClaw host. A Raspberry Pi 4/5 works great too and costs under $100.
Cloud VPS — A $5-10/month server from DigitalOcean, Hetzner, or Linode. Bonus: it runs 24/7 without keeping your laptop open.

Tip

For this tutorial, installing directly on your machine is totally fine — we're just learning. But when you start giving OpenClaw access to real messaging accounts and automations in Molt 2+, consider moving it to an isolated environment. We'll cover full security hardening in Molt 3.

Running on Your Main Machine

If you just want to get started quickly and explore, your own computer works great. OpenClaw supports macOS, Linux, and Windows (via WSL2). Just be mindful of what permissions you grant it.

Pick Your AI Model

OpenClaw is the agent — the brain that decides what to do. But it needs an AI model underneath to actually think. You'll need an API key from one of these providers.

Premium Models (Best Quality)

These give the best results but cost more:

Claude (Anthropic) — Best at following complex instructions and staying safe. Our recommendation for beginners who want reliability.
GPT (OpenAI) — Solid all-around. The most well-known option.

Budget-Friendly Models (Best Value)

If you want to keep costs rock-bottom, these models punch well above their price:

Kimi K2.5 (Moonshot AI) — Excellent reasoning at a fraction of the cost of premium models. One of the best price-to-performance ratios out there.
MiniMax 2.5 — Strong general-purpose model with very competitive pricing. Great for daily automation tasks.
GLM-5 (Zhipu AI) — Powerful and affordable. Particularly good at structured tasks and following instructions.
DeepSeek — The OG budget pick. Reliable and cheap.

Tip

Our honest recommendation: Start with a budget model like Kimi K2.5 or MiniMax 2.5 for learning. They're great for everyday tasks and you'll spend pennies while figuring things out. You can always switch to Claude or GPT later for specific tasks that need premium quality — OpenClaw makes it easy to swap models.

You'll need to create an account with your chosen provider and grab an API key:

Warning

Your API key is like a password — never share it publicly, never commit it to GitHub, never paste it in a Discord chat. If someone gets your key, they can run up your bill. We'll talk more about key security in Molt 3, but for now: keep it secret, keep it safe.

Installation

The OpenClaw team built a one-command installer that handles everything. Open your terminal and run:

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

The wizard will:

Check your system requirements
Install OpenClaw and its dependencies
Ask you which AI model you want to use
Walk you through entering your API key
Ask you to pick a messaging channel — choose Telegram
Ask if you want to install any skills (like 1Password, Google, etc.) — feel free to skip this for now
Start OpenClaw for the first time

That's it. Seriously. When you see the little lobster emoji in your terminal, you're in business.

Verify It's Working

Let's make sure everything is running. Open your terminal and run:

openclaw gateway status

You should see something like this (output shown for macOS — yours may look slightly different on Linux or WSL2):

Service: LaunchAgent (loaded)
File logs: /tmp/openclaw/openclaw-2026-02-19.log
Command: /opt/homebrew/bin/node /opt/homebrew/lib/node_modules/openclaw/dist/index.js gateway --port 18789
Service file: ~/Library/LaunchAgents/ai.openclaw.gateway.plist
Service env: OPENCLAW_GATEWAY_PORT=18789

Config (cli): ~/.openclaw/openclaw.json
Config (service): ~/.openclaw/openclaw.json

Gateway: bind=loopback (127.0.0.1), port=18789 (service args)
Probe target: ws://127.0.0.1:18789
Dashboard: http://127.0.0.1:18789/
Probe note: Loopback-only gateway; only local clients can connect.

Runtime: running (pid 36821, state active)
RPC probe: ok

Listening: 127.0.0.1:18789

If you see Runtime: running and RPC probe: ok — congratulations! You've got a running AI agent on your machine.

What's Next?

Your lobster is alive! In the next chapter, we'll connect OpenClaw to Telegram so you can chat with your agent from your phone. That's when things start to feel real.

Already have OpenClaw installed? No worries — the next chapter will also cover how to set up Telegram on an existing installation.

Challenge

Install OpenClaw using the install wizard above. Once it's running, send it this message: "What's your name and what can you do?" Paste its response somewhere you can find it later — that's your proof of completion for this chapter.