More people are choosing to host their own AI agents because when you use AI tools from other companies, you have to send your business data, questions, and files to their servers. 

OpenClaw solves this problem. It is an open-source tool that lets you run AI agents on your own setup. Meaning, you retain full control over your information and how your automations work.

When hosting OpenClaw on a VPS, you get your own dedicated resources, so your agents run smoothly without slowdowns from other users on the same machine. You also get full access to the system, which lets you install what you need, adjust settings, and connect other tools.

And, costs depend on what you plan to do. A basic plan with 1 vCPU and 2 GB RAM handles simple tasks well. 

If you want to run multiple agents simultaneously or build complex workflows, you will need more memory and processing power. 

Truehost has OpenClaw-specific plans starting at $11 per month when you pay for three years upfront.

What You Need Before You Self-Host

What you need	Why you need it
A VPS plan	OpenClaw requires dedicated resources and full root access. Shared hosting restricts background processes and lacks the CPU isolation needed for AI workloads.
An AI model API key	OpenClaw needs to connect to a language model provider such as OpenAI or Anthropic, or to a local model via Ollama. The API key authenticates your requests.
Messaging platform credentials	If you plan to use Telegram, WhatsApp, or Discord, you need bot tokens and properly configured channels.
SSH client	You will use this to connect to your VPS and run deployment commands.

7 Steps to Self-Host AI Agents with OpenClaw VPS

1) Choose and order your OpenClaw VPS plan from Truehost.

Truehost offers pre-configured OpenClaw plans built specifically for AI workloads. 

Our plans include Ubuntu environments, Playwright browser frameworks, Python runtimes, and the OpenClaw orchestration engine, all of which are already installed.

Why Truehost’s OpenClaw hosting stands out:

• High-performance AMD virtualization instances – custom-configured for headless web interaction and automation tasks, with deployment in under 45 seconds.
• Isolated VPS resources – secure dedicated vCPU time and guaranteed physical memory.
• Built for high-execution AI workloads – dedicated virtual core configurations prevent headless browser slowness and script interruptions.
• Secure infrastructure – customizable software firewalls, DDoS threshold filters, and daily SSD snapshots.

We have a variety of packages you can pick:

Package	Price (triennial)	Specs
OpenClaw Starter	$11/month	1 vCPU, 2 GB RAM, 50 GB NVMe, 4 TB bandwidth
OpenClaw Pro	$17/month	2 vCPU, 4 GB RAM, 100 GB NVMe, 6 TB bandwidth
OpenClaw Business	$23.08/month	4 vCPU, 8 GB RAM, 200 GB NVMe, 8 TB bandwidth

How do you deploy?

• Select your plan – choose the virtual AMD processor cores and memory for your workloads.
• Automated deployment – Truehost’s hypervisors install prebuilt node packages and script extensions in under 45 seconds.
• Connect your plan – access your console remotely via root SSH, paste deployment keys, and activate your worker loops.

2) Retrieve credentials from your welcome email (dashboard URL and gateway token)

After ordering, Truehost sends a welcome email containing your dashboard URL and gateway token. 

These credentials are your entry point. Save them in a secure location. 

You will need the dashboard URL to access the OpenClaw web interface and the gateway token to authenticate your connection.

3) Access the Dashboard and Configure the Interface

Open the dashboard URL from your welcome email in a browser. Log in using the provided credentials. 

The dashboard gives you a central view of your OpenClaw instance, gateway status, installed skills, and configuration options.

Check that the gateway status shows “running”. If it shows “unknown” or “stopped”, wait a few minutes and refresh. 

If it remains stuck, you can restart the gateway using the dashboard or via command line with openclaw gateway restart.

4) Connect an AI model by Entering Your API Key

Without a working API key, your agents cannot generate responses.

Navigate to the model configuration section in the dashboard. Enter your API key from your chosen provider (OpenAI, Anthropic, or others). Save the configuration.

Common points to check:

• Ensure the key is copied correctly with no extra spaces or missing characters.
• Verify that the key has not expired and that you have sufficient credits.
• Restart the gateway after saving to apply the change.

You can test if the connection works by sending a simple prompt through the dashboard. If the model responds, you are good to go.

5) Connect Messaging Channels

OpenClaw supports Telegram, Discord, and WhatsApp.

To get started with Telegram:

• Create a bot using BotFather on Telegram.
• Copy the bot token.
• In the OpenClaw dashboard, go to channel settings and paste the token.
• Save and restart the gateway.

And for WhatsApp:

WhatsApp integration requires Meta’s Business API. The QR code method is common; scan the code in the dashboard to link your WhatsApp account. If the QR code expires, you can regenerate it.

Discord:

• Create a bot in Discord’s developer portal
• Get the token
• Add it to OpenClaw’s channel settings

After connecting a channel, send a test message to verify that the bot responds.

6) Add Skills as Needed

Skills extend OpenClaw’s capabilities. They allow your agents to perform specific tasks, such as searching the web, sending emails, or interacting with external tools.

You can add skills in a few ways:

• Via dashboard – look for the skill configuration section and install from available options.
• Via command line – use openclaw skills list to see available skills and clawhub install [skill-name] to install them.
• Via chat – in some configurations, you can ask the bot what skills are available.

If a skill installation fails, it may be due to network timeouts or ClawHub community rate limiting. 

Try installing skills during off-peak times or use a ClawHub token to bypass rate limits.

7) Test one simple workflow

Set up a basic workflow like:

• The bot listens for a command like /summarize [text].
• It sends the text to the connected AI model.
• The model returns a summary.
• The bot posts the summary back to the chat.

This confirms everything is working – the messaging channel receives input, the AI model processes it, and the response is delivered. 

Once this works, you can build more complex multi-step workflows that chain API calls and interact with multiple systems.

Common Issues to Troubleshoot

Issue	Fix
Gateway won’t start or shows “Runtime: stopped”	Check the XDG_RUNTIME_DIR is set: export XDG_RUNTIME_DIR=/run/user/$(id -u). Verify systemd user instance is running: systemctl --user status. Check port conflicts with ss -tlnp | grep 3000
API key errors (invalid, expired, or insufficient credits)	Copy the key again, ensuring no spaces. Check your provider dashboard for expiry and credits. Restart the gateway after saving the key. Test the key independently with a direct API call.
Telegram bot not responding	Verify the webhook URL is accessible from the internet. Check if the bot token is correct. If you regenerated it on BotFather, you need to update it in OpenClaw. Ensure the SSL certificate is valid
WhatsApp QR code expired or “not linked”	Regenerate the QR code from the dashboard. If it keeps expiring, check if the server time is correctly synced.
High memory usage or slow performance	Check memory with free -h. If usage is above 80%, consider upgrading your VPS plan or uninstalling unused skills. Reduce conversation history length and simplify prompts for better performance
Cron jobs not firing	Ensure the service is enabled for startup: systemctl enable openclaw. Verify your cron expression format is correct. Check logs for cron execution records

How to Self-Host AI Agents with OpenClaw VPS FAQs

Ready to Self-Host OpenClaw?

Truehost makes OpenClaw deployment simple. Our plans come pre-configured with the orchestration engine, Python stack, and Playwright frameworks already installed. 

You get stable Ubuntu environments built for high-execution AI workloads, so you can focus on building agents, not wrestling with server setup.

We offer flexible billing in USD, reliable support when you need help, and a fully scalable VPS environment that grows with your needs. AMD virtualization instances are provisioned in under 45 seconds, and your resources are isolated.

Choose from OpenClaw Starter ($11/mo), Pro ($17/mo), or Business ($23.08/mo) – all backed by advanced security, DDoS protection, and daily SSD snapshots.