OpenClaw VPS Hosting — The Right Way to Run Your AI Assistant

OpenClaw runs as a persistent Node.js process that maintains state between sessions, writes to the filesystem continuously, executes shell commands, and optionally runs browser automation through Playwright and Chromium. These requirements immediately disqualify shared hosting, serverless platforms, and ephemeral container environments.

A VPS (Virtual Private Server) gives you a dedicated slice of a physical machine: your own CPU cores, guaranteed RAM, persistent disk storage, and full root access to a Linux environment. OpenClaw can run as a systemd service, restart automatically on failure, write memory files to disk, and maintain open connections to the Telegram API — all of which require the persistent, process-capable environment that a VPS provides.

The practical implication: if you want to run OpenClaw and you're not renting a VPS (or using a managed service like PlugAndClaw that does it for you), you can't run OpenClaw properly. You might get it running temporarily in a terminal session, but the moment you close that session, your assistant disappears — losing any in-progress work and requiring a manual restart.

OpenClaw's heartbeat system is a good illustration. Heartbeats are scheduled checks that run every 30 minutes or whatever interval you configure — checking email, reviewing calendar events, monitoring websites. These only work if OpenClaw is continuously running. A VPS with systemd makes this trivial. Anything else makes it impossible.

Let's be specific about hardware requirements for different OpenClaw use cases.

Basic chat usage (no browser automation, light file management): 1 vCPU, 2GB RAM, 20GB SSD. This works on the cheapest Hetzner CX11 (€3.29/month) or equivalent. You'll hit memory pressure if you try to run Chromium, but basic AI conversations, web fetching, and file management work fine.

Standard usage with browser automation: 2 vCPU, 4GB RAM, 40GB SSD. This is the sweet spot. OpenClaw's core process takes ~300MB. Add the OS (~500MB), Caddy (~50MB), and a single Chromium browser instance (~300MB) and you're at about 1.1GB peak usage — well within 4GB headroom. The extra RAM handles multiple concurrent tasks without OOM risk.

Heavy usage (multiple browser contexts, extensive file processing, running additional services): 4 vCPU, 8GB RAM, 80GB SSD. This level is rarely needed for personal use but makes sense for teams sharing a single OpenClaw instance or running intensive automation workflows.

CPU matters less than RAM for most OpenClaw workloads. The limiting factor is almost always waiting for LLM API responses, not local compute. Exception: browser automation with many concurrent tabs can be CPU-intensive.

Network: any modern VPS has sufficient network capacity for OpenClaw. You're making API calls to Claude/GPT/Gemini (small JSON payloads) and receiving Telegram webhook events. Even a 100Mbps connection is vastly more than needed.

Not all VPS providers work equally well for OpenClaw. Here's a practical comparison of the major options.

Hetzner (recommended): Best price-to-performance in Europe. CX22 (2 vCPU, 4GB RAM, 40GB SSD) costs €4.35/month. Data centers in Germany and Finland (GDPR-covered). Excellent network, reliable performance, LUKS2 encryption supported. The clear choice for European users and the provider PlugAndClaw uses exclusively.

DigitalOcean: 4GB RAM Droplet costs $24/month — 5x more expensive than Hetzner for similar specs. Excellent developer experience, good documentation, US and EU data centers. Worth considering if you're already in the DigitalOcean ecosystem or need their Managed Kubernetes. Expensive for single-server OpenClaw.

Vultr: Competitive pricing (4GB RAM instance at ~$24/month in the US, cheaper in some regions). Similar quality to DigitalOcean. Global data center network. Good choice if you need specific locations not covered by Hetzner.

Contabo: Very cheap (4GB RAM VPS at €5.99/month) but with a catch: Contabo uses heavily oversubscribed infrastructure. CPU performance can be unpredictable. Not recommended for latency-sensitive workloads like real-time AI assistant interactions.

AWS/GCP/Azure: Technically viable but overkill and expensive for a single OpenClaw instance. Use if you need specific compliance certifications, have existing cloud credits, or need to integrate with cloud services. An AWS t3.medium (2 vCPU, 4GB RAM) costs about $30/month before traffic — more than PlugAndClaw's full managed service.

If you want to self-host, here's the realistic setup process. This is what PlugAndClaw automates, documented here so you understand the complexity.

Step 1: Provision the VPS. Choose Ubuntu 22.04 or Debian 12. Enable SSH key authentication during provisioning. Note the server IP.

Step 2: Initial hardening. SSH in, disable root login, disable password authentication, configure UFW (ufw default deny incoming, ufw allow ssh, ufw allow http, ufw allow https, ufw enable), install fail2ban.

Step 3: Install Caddy as the reverse proxy. Add the Caddy GPG key and apt repository, install caddy, write a Caddyfile pointing to your domain, configure Caddy as a systemd service.

Step 4: Install Node.js (OpenClaw requires Node 18+). Use nvm or the NodeSource APT repository.

Step 5: Clone OpenClaw, install dependencies (npm install), configure environment variables (Telegram bot token, API keys, etc.).

Step 6: Create a Telegram bot via @BotFather, configure the webhook URL to point to your Caddy-proxied OpenClaw endpoint.

Step 7: Write a systemd service unit for OpenClaw, enable and start it. Test that it survives a reboot.

Step 8 (optional but recommended): Configure LUKS2 encryption. This requires encrypting the disk partition and setting up automatic unlock with a key file — a multi-step process that varies by provider and OS version.

Total time: 3-6 hours for experienced Linux admins, 8-12 hours for those less familiar. PlugAndClaw does all of this in under 60 seconds.

The self-hosting calculation looks simple on paper: €4/month VPS vs $39.50/month PlugAndClaw. But the math changes when you account for the actual costs.

Setup time: 4-6 hours at even $20/hour = $80-120. That's 2-3 months of PlugAndClaw. And that's if everything goes right the first time — which it usually doesn't. A misconfigured Caddy, a botched LUKS2 setup, or a subtle UFW rule that blocks Telegram webhooks adds hours of debugging.

Ongoing maintenance: OpenClaw updates every few weeks. Ubuntu security patches need to be applied. Certbot (or Caddy's cert renewal) occasionally fails and needs investigation. If your systemd service crashes at 3am, you won't know until you check in the morning. These tasks add up to 30-60 minutes per month for a well-configured system.

Cognitive overhead: Knowing you have a server to maintain is a background tax on your attention. With managed hosting, when something breaks, PlugAndClaw fixes it. You get notified, not paged.

PlugAndClaw at $39.50/month includes $20 AI credits (Claude, GPT, Gemini access without separate API contracts), managed updates, monitoring, and support. Competitors SimpleClaw and ClawDrift charge $49/month for similar service without the AI credit inclusion. PlugAndClaw is the best value in the managed OpenClaw market.