How to Run OpenClaw on a Budget

If you've been running OpenClaw — or thinking about getting started — chances are you've already noticed how quickly the costs can pile up. Between the VPS bill and the API fees, a lot of people end up spending $30 to $50 a month, sometimes even more. And the frustrating part? You really don't have to.

In this guide, I'm going to walk you through exactly how to set up a fully functional OpenClaw instance for well under $30 a month. You'll learn which VPS provider gives you the most value for the specs, how to pick an affordable AI model that won't bleed your wallet dry, and — if you already have a ChatGPT paid subscription — how to plug that directly into your OpenClaw setup at no extra cost.

This is the exact setup I personally run with five agents (a researcher, a content writer, a web developer, a marketing specialist, and a social media manager) all coordinated through a custom mission control dashboard. Don't underestimate what you can build on a budget like this.

• ▸A VPS server (we'll go with Contabo — more on why below)
• ▸An AI model provider (ChatGPT Pro/Codex, Minimax, or OpenRouter)
• ▸A Telegram account (to communicate with your agents)
• ▸About 30–60 minutes of setup time

The first thing OpenClaw needs is a Virtual Private Server (VPS). A VPS is basically a remote computer that runs in the cloud 24/7, so your agents can keep working even when your laptop is off.

A lot of people in the OpenClaw community use Hostinger, and it's a perfectly solid provider. But when it comes to raw value for OpenClaw workloads — that is, the specs you get for the price you pay — Contabo consistently wins.

Contabo is a German company, and I've been using them for years without a single major downtime. Their entry-level VPS plan starts at close to $4 a month, and it comes with specs that will comfortably handle most OpenClaw setups. If you want a bit more headroom, there's a mid-tier option at around $6.36/month that's noticeably more powerful.

For the majority of users, the entry plan is more than enough to get started.

Note: If you use my referral link in the video description, it doesn't cost you anything extra — but it does help me keep making tutorials like this one.

1. Go to Contabo's website and select the plan that works for you
2. Click Get Started
3. For the server location, keep it as European Union — since Contabo is a German company, this is free. Choosing another region adds an extra cost you don't need
4. Set a strong password for your VPS — write this down somewhere safe. If you lose it, you won't be able to log into your server
5. Fill in your personal information, make the payment, and wait

After payment, Contabo will send you a confirmation email within a few minutes to one hour. This email will contain:

• ▸Your VPS IP address (unique to your server)
• ▸Your username (for most providers including Contabo, this is root)

Save both. You'll need them in the next step.

Once you have your credentials, it's time to actually log into your server. You don't need to install any special software — both Windows and Mac have everything built in.

• ▸Windows: Press the Windows key, type PowerShell, and open Windows PowerShell
• ▸Mac: Open the Terminal application (found in Applications > Utilities)

The commands are identical on both platforms.

ssh root@YOUR_VPS_IP_ADDRESS

Replace YOUR_VPS_IP_ADDRESS with the IP address from your Contabo email.

PowerShell tip: Right-clicking in PowerShell automatically pastes whatever you have in your clipboard. So you can copy your IP address from the email and right-click to paste it into the command.

After pressing Enter, the terminal will ask if you trust this connection (the first time only). Type yes and hit Enter. Then it will prompt you for your password.

⚠️ Important

When typing or pasting a password in terminal/PowerShell, you won't see any characters appear on screen. That's normal — it's a security feature. The password is still being entered. Just paste or type it and press Enter.

If you set up OpenClaw as part of the Contabo marketplace app, the OpenClaw installation wizard will load automatically once you're logged in.

Before we walk through the OpenClaw setup wizard, you need to decide which AI model you're going to use. This is honestly where most people get burned.

Most AI APIs charge on a pay-per-use basis — meaning the more your agents run, the more you pay. The first time I set up OpenClaw, I burned through close to $500 in a weekend just by choosing the wrong model for the job. Learn from my mistake.

You have two solid options here:

If you're already paying for a ChatGPT Plus or Pro subscription, you can connect OpenClaw to OpenAI's Codex model through a special OAuth flow — at no additional cost. This has been confirmed as an allowed use case by the OpenClaw founder, and I've been running this personally for months without any issues.

Codex is a genuinely powerful model and makes an excellent primary LLM for your agents.

If you don't have ChatGPT Pro, or you want a backup model (which I highly recommend — more on that in a moment), Minimax is the most cost-effective option I've found.

Unlike OpenAI's pay-per-use API pricing, Minimax offers a flat monthly subscription with an extremely generous token allowance. In practice, you'll barely hit the limits even with multiple agents running throughout the day.

I personally use the Plus plan at $20/month. For the vast majority of OpenClaw setups, this is more than sufficient.

My recommendation: Skip the yearly plan for now. The AI space moves fast, and a better option might come along. Start monthly and reassess.

1. Go to minimax.io and sign up
2. Navigate to API in the menu, then click Token Plan
3. Select the monthly plan you want and complete payment
4. After subscribing, go to Account → Token Plan
5. Click to generate your API key — store it securely, you'll need it shortly

Now back to your VPS terminal. Once the OpenClaw installer loads, here's what to expect:

1. Accept the terms and conditions — Use the left arrow key to toggle the option to "Yes" and press Enter
2. Select Quick Start — Press Enter (it should already be highlighted in green)
3. Choose your model provider:
   • ▸ If you have ChatGPT Pro → Select OpenAI
   • ▸ If you're going with Minimax → Scroll down and select Minimax

After selecting OpenAI, the wizard will ask how you want to authenticate. Do not choose the API key option — that would charge you per-use. Instead:

1. Select the OpenAI Codex / ChatGPT option using the arrow keys
2. The terminal will generate a URL — click it (or use Ctrl+Click) to open it in your browser
3. Log into your OpenAI account in the browser
4. Confirm that you want to sign into Codex using your ChatGPT account
5. After confirming, you'll land on a somewhat blank-looking page — that's fine
6. Copy the full URL from the address bar of that page
7. Go back to your terminal/PowerShell and paste that URL, then press Enter

This is the OAuth authentication flow. Once it completes, keep the suggested version (usually the latest Codex version) and press Enter.

OpenClaw lets you communicate with your agents through several messaging apps — Telegram, WhatsApp, Discord. From experience, Telegram is the most reliable and responsive option, so that's what I'll show here.

1. Open Telegram and search for BotFather
2. Make sure it has the blue verification checkmark — there are fakes out there
3. Type /newbot and press Enter
4. Give your bot a display name (e.g., "My OpenClaw Agent")
5. Give it a username — this must end in bot (e.g., openclaw_mysetup_bot)
6. BotFather will reply with a token like this:

1234567890:ABCDefGHIJKlMNOPQRST-uvwxyz1234567

Copy this token — it's what connects your Telegram bot to OpenClaw.

Back in the installation wizard:

1. Select Telegram as your communication channel
2. Choose Enter Telegram Bot Token
3. Paste the token you copied from BotFather and press Enter

For the search provider (next screen), you can skip this for now and set it up later. Same for skills and webhooks — skip them for the initial setup.

When prompted, select Hatch in TUI (Terminal User Interface). This is a quick sanity check that lets you see if everything is working before you leave the terminal. You'll see your agent wake up and respond — something like "Hey, I'm awake. Fresh workspace, blank memory."

That means it worked.

Once confirmed, press Ctrl+C a few times to exit the TUI and return to the main command line (you'll know you're back when the prompt shows root@...).

Now go back to Telegram. In your BotFather conversation, you'll see a link to your new bot. Click it, then click Start. The bot will give you a pairing command — copy it.

Back in your terminal, paste that command and press Enter. You should see:

Approved Telegram sender: [your Telegram ID]

Now test it. Open the bot in Telegram and send a message like "Hey, are you there?" — if your agent responds, you're fully set up.

Having only one LLM connected to OpenClaw is risky. If that model goes down or gets rate-limited, your agents simply get stuck. Adding a backup model takes about two minutes and can save you a lot of headaches.

To add Minimax as a secondary model, run this command in your terminal:

openclaw configure

This opens the OpenClaw configuration menu. From there:

1. Select Local (already green) and press Enter
2. Navigate to Model using the arrow keys and press Enter
3. Scroll down and select Minimax, then press Enter
4. Select Minimax API Key (Global) and press Enter
5. Paste your Minimax API key and press Enter

OpenClaw will default to Minimax as the primary model. To fix this — if you want Codex as your primary and Minimax as fallback — restart the gateway first:

openclaw gateway restart

Then open the TUI again:

openclaw tui

And tell your agent directly in the chat:

"Set [OpenAI Codex model name] as my primary model and [Minimax model name] as fallback #1."

The agent will handle the configuration for you. Just wait for it to finish before sending any follow-up messages — interrupting a configuration task can cause issues.

You can also switch models on the fly directly from your Telegram chat:

/models

This command shows all your configured providers. Click through to select the exact model version you want to activate. It's that simple.

That's a full multi-agent OpenClaw setup, running 24/7, for under $30/month.

Here's an example of what I'm personally running on this exact setup:

• ▸5 specialized agents — researcher, content writer, web developer, marketing specialist, social media manager
• ▸A mission control dashboard where every agent logs their activity in real time
• ▸A personal to-do tracker integrated into the same system

All of this is powered by the affordable stack described above. If you want to build something similar, I have a full tutorial on setting up the mission control dashboard.

Getting started with OpenClaw doesn't have to be expensive. With Contabo for your VPS and Minimax (or your existing ChatGPT subscription) for the AI model, you can build a genuinely powerful multi-agent system for less than the cost of a streaming subscription.

If you run into any trouble at any step, feel free to reach out — I'm happy to help you work through the setup.

And if this guide was useful, check out the full video walkthrough on the Komputer Mechanic YouTube channel where you can see every step done live. More tutorials at komputermechanic.com/tutorials.