OpenClaw VPS Setup: Beginner's Guide to Clawdbot on Hostinger

Most OpenClaw tutorials are 30 minutes of terminal commands that make your eyes glaze over. This one isn't. I'll walk you through the fastest, safest way to get Clawdbot running on a VPS, connect it to WhatsApp, and get your API costs under control from day one.

Why a VPS, Not Your Computer#

The most common mistake I see with OpenClaw is installing it directly on your personal machine. That gives the agent access to your passwords, emails, API keys, and calendar. If something goes wrong, or the agent does something unexpected, there's no barrier.

A VPS keeps your AI agent fully isolated. It runs in its own environment, has no connection to your personal files, and you can shut it down or wipe it entirely without touching anything on your computer. For $7-$10 a month, that peace of mind is worth it.

Some people are buying $600 Mac minis for this. You don't need to.

Setting Up OpenClaw on Hostinger#

Hostinger has an auto-deploy option for OpenClaw that removes the need for any terminal work. Here's the exact flow:

1. Go to the OpenClaw setup page on Hostinger (link in the description below) and click deploy
2. Select the KVM2 plan, which has enough RAM and storage to run OpenClaw comfortably
3. Use code MOE-LUEKER for 10% off, then choose your server location and continue
4. After checkout, you'll see an access token. Copy it somewhere safe. You'll need it to log into the dashboard.
5. Paste in your API key. Start with OpenAI for simplicity, then add Anthropic and Gemini keys later for multi-model routing.
6. Add your WhatsApp number and click deploy.

Once deployed, click the link to open the dashboard. Paste in your access token and log in. That's it. You're looking at a live OpenClaw instance.

First Things to Configure#

The moment you're in, ask it a few questions. Let it gather context about you. OpenClaw stores everything it learns in an identity.md file inside the agent's workspace, and that file gets smarter the more you interact with it.

A few settings worth enabling early:

Memory flush. This tells OpenClaw to periodically compact everything it remembers into a tighter format. Without it, each session starts with a bloated context window that burns tokens fast.

Reverse prompt engineering. Instead of you explaining yourself, prompt the agent to ask you questions. It builds out its own memory files based on your answers. Way more efficient than trying to write a perfect system prompt yourself.

API budget limits. Go into your OpenAI billing settings and set a hard limit of $25-$50. This prevents any runaway cost from an unexpected task loop. Set it before you do anything else.

Connecting to WhatsApp#

OpenClaw becomes genuinely useful when you can message it from your phone without opening a browser.

In the dashboard, go to Channels, select WhatsApp, and click Show QR Code. On your phone, open WhatsApp, go to Link a Device, and scan the code. Once linked, you can send messages to your own number and OpenClaw will respond.

I tested it by sending "are you there?" from my own WhatsApp. It replied in seconds. The chat also shows up in the dashboard so you can see the full conversation history.

If WhatsApp isn't your thing, Telegram works as an alternative.

What You Can Do With It#

The morning briefing is the best first automation to set up. Prompt it to run a cron job that messages you every morning with weather, top headlines, a market snapshot, and one key thing you should focus on. It takes about 30 seconds to configure and teaches you how scheduled tasks work.

After that, the use cases start stacking up. A few I've been using:

• Deep research reports. Tell it to research Anthropic, a competitor, or the top 20 S&P 500 companies. It runs parallel searches and saves a formatted report to your files.
• Document creation. Ask it to generate Excel or PDF files from research it compiles.
• Purchase research. Give it a product category and a budget. It finds options, compares specs, and drafts a recommendation.
• Draft negotiation messages or interview prep. Feed it context and let it do the writing.

The more feedback you give it on outputs, the sharper its memory files become.

Cutting API Costs by 80%#

Here's where most people leave money on the table. By default, OpenClaw sends every request to whatever model you set as the default. That's fine for complex tasks, but overkill for simple ones.

The fix is multi-model routing. You tell the agent to route simple questions and quick lookups to cheap models like GPT-4o mini, Gemini Flash, or Claude Haiku, and reserve the heavy models for research and writing. The routing logic lives in a prompt you paste once.

The other big one is context compaction. Before this, I was burning around 2-3 million tokens per session start, which came to about 40 cents each time. After running a context compaction prompt, that dropped to under 5 cents. Sometimes even lower when routing through cheap models.

I put both prompts in a free sheet you can grab from the description. Paste them in once, and the savings are automatic from there.

For a deeper breakdown of cost optimization strategies, including OpenRouter routing and heartbeat tuning, check out OpenClaw Cost Optimization: Run 19 Agents for $2/Month.

And once you're past the basics, the 15 First Prompts to Configure OpenClaw the Right Way covers the full security, personalization, and cost setup in one copy-paste session.

Watch the full video on YouTube: https://youtu.be/pSvsLwGMy4A

Some links below may be affiliate links. I only recommend tools I actually use, and it may give you a discount if you use my links.