OpenClaw

OpenClaw is an open-source AI assistant framework that runs on infrastructure you control. Instead of depending on a hosted AI agent service, you can deploy it on a small VPS and keep your models, tools, and data flow closer to your own environment.

In this post, I will focus on a practical question: why run OpenClaw on a cloud node, and what does a minimal secure setup look like on a low-cost VPS such as Hetzner or AWS Lightsail?

Why run it on a cloud node?

Running OpenClaw on a VPS is useful if you want:

  • Always-on availability: Your assistant stays online even when your laptop is closed.
  • Stable network reachability: Easier remote access, webhooks, and gateway access than on a home network.
  • Separation from your daily machine: Agent tooling, logs, and automations live on a dedicated host.
  • Predictable cost: A small instance is often enough for a personal agent, especially if inference happens through external model APIs.

What is OpenClaw?

OpenClaw is a personal AI agent platform built for developers and power users. Key traits:

  • Self-hosted: Runs on your hardware (laptop, server, Raspberry Pi, or cloud node).
  • Messaging integrations: Can connect to services like WhatsApp, Telegram, Discord, Slack, and others through its gateway and integrations.
  • Agentic: Supports subagents, tools (file ops, shell exec, web search), skills (plugins for GitHub, calendars, etc.), and persistent memory.
  • Model-agnostic: Works with hosted and local models, depending on your provider setup.
  • Open-source: GitHub repo with a large and active community.

It’s not just a chatbot; it’s an extensible system for automation, with heartbeats for proactivity, cron jobs, and community skills.

What Can It Do?

OpenClaw shines in real-world automation:

  • Task automation: Clear inboxes, manage calendars (gcalcli), send emails, check flights.
  • Development workflows: GitHub issues/PRs (skills like gh-issues), code review, CI monitoring.
  • Research & content: Web search, image gen/analysis, note-taking (Obsidian publish).
  • System ops: Shell access (safe exec), file management, tmux control.
  • Proactive: Heartbeats check emails/calendar/weather; subagents for parallel tasks.
  • Voice/media: TTS, image handling, voice notes transcription.

Examples: “Ship checklist” can become an automated deployment checklist, or you can spawn subagents for parallel work across multiple repos.

Security Considerations

Self-hosting improves control, but it also means you own the operational risk:

  • Minimal exposure: Prefer loopback-only binding plus Tailscale or another private access layer instead of opening the gateway publicly.
  • Secrets handling: Store API keys carefully and keep them out of shell history, screenshots, and public dotfiles.
  • Exec safety: Treat shell and file tools as privileged capabilities. Keep approval requirements in place for destructive or elevated commands.
  • Firewalling: Allow only what you need, ideally SSH and your private networking layer.
  • Updates: Stay current with OpenClaw and its runtime requirements.
  • Isolation: Run as a non-root user and keep the workspace scoped to what the agent actually needs.
  • Plugin hygiene: Review skills and plugins before enabling them on a machine with real data.

The main risks are straightforward: prompt injection, accidental over-permissioning, and exposing an interface to the public internet without enough access control.

Practical Setup on Hetzner or AWS Lightsail

  1. Provision node:
    • Hetzner: a small shared vCPU instance with Ubuntu 24.04 is enough to start.
    • Lightsail: a small Ubuntu instance is also fine for a personal deployment.
    • Create a non-root user and use SSH keys.
  2. Secure basics: 
    sudo apt update && sudo apt upgrade -y
sudo apt install ufw curl -y
sudo ufw default deny incoming
sudo ufw allow ssh
sudo ufw --force enable
# Disable root SSH, password auth in /etc/ssh/sshd_config
sudo systemctl restart ssh
  3. Install Node & OpenClaw: 
    curl -fsSL https://openclaw.ai/install.sh | bash
openclaw onboard --install-daemon

    OpenClaw currently recommends Node 24, with Node 22.16+ also supported if you are installing through npm instead of the installer script.

  4. Configure & pair:
    • Pair the channels you actually use, such as WhatsApp or Telegram.
    • Set your model provider and workspace during onboarding.
    • Test: openclaw agent --message "Hello from cloud!".
  5. Remote access (optional):
    • Tailscale is a good fit if you want private remote access without exposing the gateway directly to the internet.
    • Keep the gateway bound to loopback where possible and let Tailscale handle access.
  6. Monitor: openclaw doctor, journalctl -u openclaw (systemd).

At that point you have an always-on OpenClaw node with a cleaner separation between your assistant runtime and your personal machine.