How to Run a Personal AI on Your Own VPS

Running your own AI assistant on a VPS sounds complicated. It's not. If you can SSH into a server and copy-paste commands, you can have a personal AI running in under an hour — no cloud subscriptions, no data leaving your machine. Here's the full process, from picking a server to having an AI that checks your email and writes your blog posts.

Step 1: Pick Your VPS

You don't need a beefy machine. For running OpenClaw with cloud models (Claude, GPT), a $5/month VPS is plenty — the heavy compute happens at the API provider. If you want to run local models via Ollama, you'll need more RAM:

• Cloud models only (Claude/GPT): 1 vCPU, 1GB RAM — $5/month on Hetzner, DigitalOcean, or Vultr
• Small local model (3B-7B): 2 vCPU, 4GB RAM — $10-15/month
• Larger local model (13B+): 4+ vCPU, 16GB+ RAM — $30+/month, or a dedicated server

For UK-based users, Hetzner offers the best price-to-performance. For North America, DigitalOcean or Vultr. Avoid AWS/GCP for personal projects — the pricing is opaque and overkill.

Step 2: Server Setup

Once you have your VPS, SSH in and run the basics:

• Update packages: apt update && apt upgrade -y
• Create a non-root user: adduser spark
• Set up SSH key auth (disable password login)
• Install a firewall: ufw allow 22 && ufw enable

Security matters here — your AI will have shell access. Don't skip the firewall and SSH hardening. See our VPS setup guide for the full security checklist.

Step 3: Install Node.js and OpenClaw

OpenClaw runs on Node.js. Install the LTS version:

• curl -fsSL https://deb.nodesource.com/setup_22.x | bash -
• apt install -y nodejs
• npm install -g openclaw

Then run the initial setup: openclaw configure. This walks you through connecting your AI provider (Anthropic, OpenAI, or local Ollama), setting up your messaging channel, and configuring memory.

For a full walkthrough with Telegram integration, follow our step-by-step VPS install guide or the Telegram setup tutorial.

Step 4: Add a Local Model (Optional)

If you want a free, fully local fallback model — install Ollama:

• curl -fsSL https://ollama.ai/install.sh | sh
• ollama pull llama3.2:3b (starts with 3B, upgrade later)
• Ollama runs on localhost:11434 by default

Configure OpenClaw to use Ollama as a fallback: set agents.defaults.fallbackModel to ollama/llama3.2:3b in your config. Now your AI uses Claude for complex tasks and the free local model for simple checks. Full details in our OpenClaw + Ollama guide.

Step 5: Connect Your Channels

The whole point of a personal AI is that you can reach it from anywhere. OpenClaw supports:

• Telegram — most popular, full-featured, see setup guide
• WhatsApp — via WhatsApp Business API
• Discord — bot integration
• Signal — via signal-cli

Telegram is the easiest to set up and the most reliable. Start there.

Step 6: Set Up Proactive Features

This is where a self-hosted AI gets interesting. Through OpenClaw's heartbeat and cron systems, your AI can:

• Check your email and summarize urgent messages
• Publish blog posts on a schedule
• Monitor websites for changes
• Run maintenance tasks (backups, log rotation)
• Send you daily digests

Configure these through HEARTBEAT.md and cron jobs. Your AI reads the instructions and executes them autonomously. If you use n8n for automation, OpenClaw can trigger n8n workflows via webhooks — combining the best of both.

What It Costs

• VPS: $5-15/month (Hetzner, DigitalOcean)
• AI API: $5-20/month depending on usage (Claude Sonnet is ~$3/million input tokens)
• Local model (Ollama): $0 after hardware cost
• OpenClaw: Free (open source)

Total: $5-35/month for a fully autonomous personal AI assistant. Compare that to hiring a VA at $500+/month.