How to Set Up OpenClaw on a VPS: The No-Nonsense Guide

Most OpenClaw tutorials waste your first 10 minutes explaining what an AI agent is. You already know. You want your own agent running on a server, reachable from your phone, doing actual work. Let's get there.

This guide covers the full setup — from spinning up a VPS to chatting with your agent on Telegram. Total time: about 25 minutes if you type at a normal pace.

What You Need Before Starting

A VPS with at least 2 vCPUs, 4GB RAM, and 40GB storage (Ubuntu 22.04 or newer)
An API key from Anthropic, OpenAI, or Google — pick one
A Telegram account (or WhatsApp, Discord, Slack — Telegram is fastest to configure)
Basic comfort with a terminal. You don't need to be a sysadmin, but you should know what ssh means

Any VPS provider works. Hetzner, DigitalOcean, Linode, Vultr, Hostinger — they all run Ubuntu fine. The cheapest tier that meets the RAM requirement is enough for a single agent.

Step 1: SSH Into Your Server

Once your VPS is provisioned, grab the IP address from your provider's dashboard and connect:

ssh root@YOUR_SERVER_IP

First login on a fresh Ubuntu box? Update packages:

apt update && apt upgrade -y

Step 2: Install OpenClaw

One command. Seriously.

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

The installer handles Node.js, dependencies, and the OpenClaw binary. It takes 2-3 minutes depending on your server's internet speed.

After installation, run the onboarding wizard:

openclaw onboard --install-daemon

The --install-daemon flag sets OpenClaw to start automatically on boot. Skip it if you prefer manual control, but for a VPS that should run 24/7, you want it.

The wizard walks you through:

Choosing an AI provider (Anthropic, OpenAI, Google, or Ollama for local models)
Entering your API key
Setting a default model
Configuring your first chat channel

Step 3: Connect Telegram

Open Telegram and search for @BotFather. Start a chat and send:

/newbot

BotFather asks for a display name and a username (must end in bot). Give it whatever you want — "MyAgentBot", "WorkBuddyBot", doesn't matter.

Copy the API token BotFather gives you. Back in your terminal:

openclaw channels add --channel telegram --token YOUR_BOT_TOKEN

Now open a chat with your new bot in Telegram and send /start. The bot responds with a pairing code. Copy that code, go back to your terminal, and approve it:

openclaw pair approve PAIRING_CODE

Send a test message to your bot. If it replies, you're connected.

Step 4: Choose Your Model

During onboarding you picked a default, but you can change it anytime:

openclaw models set anthropic/claude-sonnet-4-20250514

Or if you want something cheaper for casual use:

openclaw models set anthropic/claude-haiku-4-5-20251001

The model choice affects cost and quality. Sonnet is the sweet spot for most people — smart enough for real work, affordable enough to leave running. Opus is for complex reasoning tasks. Haiku is for high-volume, low-stakes interactions.

Check your configuration:

openclaw doctor

This shows gateway status, connected channels, AI model, and any warnings.

Step 5: Give Your Agent an Identity

OpenClaw uses a file called SOUL.md in your workspace to define your agent's personality and behavior. You can edit it directly:

nano ~/.openclaw/workspace/SOUL.md

Or just tell your agent through Telegram: "Update your SOUL.md — you're a senior developer who speaks concisely and uses code examples. Your name is Atlas."

The agent will update its own personality file. This shapes every response going forward.

Step 6: Install Skills

Skills extend what your agent can do. Browse what's available:

openclaw skill search web

Install a few essentials:

openclaw skill install web-search
openclaw skill install weather

For a list of community-built skills, check ClawHub — it's the skill marketplace where you can find everything from Gmail integration to security auditing tools.

Step 7: Lock Down Security

Your agent has shell access to your server. That's powerful but dangerous if misconfigured. Run the security audit:

openclaw security audit --deep

At minimum:

Change SSH to key-based auth — disable password login
Set up a firewall — only allow ports 22 (SSH) and 443 (if using dashboard)
Review the agent's permissions — sandbox mode limits what commands it can run
Don't give your agent root credentials to other services unless you trust the setup

The official security docs cover hardening in detail. Read them. A compromised AI agent with shell access is worse than a compromised server alone.

Step 8: Set Up Cron Jobs (Optional but Useful)

Want your agent to check your email every morning? Send you a weather report? Summarize overnight messages?

openclaw cron add --schedule "0 8 * * *" --message "Check my inbox and summarize anything important"

Cron jobs run in isolated sessions, so they don't pollute your main conversation history.

Verify Everything Works

Run through this checklist:

openclaw gateway status — should say "running"
Send a message to your bot on Telegram — should get a response
Ask your agent to search the web — confirms skills are working
openclaw doctor — all checks should pass

If something's broken, openclaw doctor usually tells you what. The most common issues: wrong API key format, Telegram token not pasted correctly, or firewall blocking outbound connections.

Cost Expectations

The VPS itself: $5-15/month depending on provider and specs.

API costs depend entirely on usage. With Sonnet as your default model, casual use (20-30 messages/day) runs about $3-8/month. Heavy use with sub-agents and research tasks can hit $30-50/month. Monitor your spending through your API provider's dashboard.

What's Next

You've got a running agent. Now make it useful:

Set up your AGENTS.md to define autonomous behavior
Build a content pipeline for automated writing
Explore multi-agent architectures for complex workflows

Need help with setup? MayaWorks offers setup-as-a-service — we configure everything and hand you a working agent.