OpenClaw 如何在 VPS 上自託管運行：AI 助手逐步教學

How to Run OpenClaw on a VPS (Self-Hosted AI Assistant, Step by Step)

OpenClaw is the open-source, self-hosted AI assistant that hit 380k+ GitHub stars in 2026 — it turns your machine (or a server) into a personal assistant you can message from WhatsApp, Telegram, Slack, Discord, and 20+ other apps, powered by whatever model you want, with all data staying on your hardware.

The one question everyone asks: where do I actually run it so it's always on? Not your laptop — that goes to sleep. The answer is a small, cheap VPS. This is the exact setup I use, written so you can replicate it in about an hour.

Why a VPS (and not your laptop)

A personal assistant is only useful if it's there 24/7. Your laptop sleeps, reboots, and takes your Wi-Fi with it. A VPS:

• Stays on 24/7 regardless of your devices
• Gives OpenClaw its own persistent home (config, memory, sessions)
• Can sit in a datacenter close to your AI model provider

All you need is roughly 2 GB RAM and a couple of vCPUs — OpenClaw itself is light; your AI provider does the heavy lifting. That's a low-cost VPS tier, which is the whole point.

Step 1 — Get the hardware

Pick a small VPS. For a personal OpenClaw setup, a single 2 GB / 1-2 vCPU instance is plenty:

• What to grab: 1 vCPU, 2 GB RAM, 20+ GB SSD, an NVMe volume, and an IPv4 address.
• Where: any provider with a low entry tier works. I run mine on Hostinger's VPS — their launch VPS plans start around $5-6/mo, which is the cheapest honest price for a machine that never sleeps.
• OS: I use a plain Linux image (Ubuntu 22.04/24.04 or Debian 12). Skip "managed" stacks for the base — you'll be installing Node anyway.

After provisioning you get a root/SSH login. That's all you need for this step.

Step 2 — Open the right port and lock the firewall

OpenClaw's Gateway listens on a local port. You only ever connect through SSH, so keep the default SSH-only exposure:

1. In your provider's control panel, open one port for SSH (22).
2. Do not open OpenClaw's port to the public internet. Run it behind an authenticated reverse proxy (see Step 5) instead.

Then on the box, install ufw and lock it down:

sudo apt update && sudo apt install -y ufw
sudo ufw default deny incoming
sudo ufw allow 22/tcp
sudo ufw enable

Step 3 — Install Node and OpenClaw

OpenClaw is built on Node. Install a current LTS Node, then OpenClaw:

# Node LTS (via NodeSource)
curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
sudo apt install -y nodejs

# OpenClaw (follows its documented install)
curl -fsSL https://openclaw.ai/install.sh | bash

Confirm the version, then start the setup wizard:

openclaw --version
openclaw setup

The setup wizard walks you through: model provider, API key, the channels you want (Telegram / WhatsApp / etc.), and where OpenClaw keeps its workspace (its config, memory, and operating instructions). Let it use the default ~/.openclaw workspace on the VPS.

Step 4 — Connect your AI model (bring your own key)

OpenClaw is model-agnostic — Claude, GPT, Gemini, or fully local via Ollama. For a 24/7 personal assistant I recommend a hosted model provider (reliable, no GPU cost on the VPS):

1. Get an API key from your chosen provider.
2. Put it in OpenClaw's config / environment — never hardcode it in a file you'll commit.
3. Test with a one-shot prompt to confirm the round-trip works.

Cost control: the VPS is a few dollars/month; your AI usage is the variable cost. Pick a model tier you'll actually use, and set provider spend limits so an agent that loops can't surprise you.

Step 5 — Keep it running 24/7 (systemd)

Don't just run it in a terminal that dies when you log out. Make OpenClaw a service:

# /etc/systemd/system/openclaw.service
[Unit]
Description=OpenClaw Gateway
After=network-online.target
Wants=network-online.target

[Service]
User=youruser
ExecStart=/usr/local/bin/openclaw gateway
Restart=always
RestartSec=5
Environment=NODE_ENV=production

[Install]
WantedBy=multi-user.target

Then:

sudo systemctl enable --now openclaw
sudo systemctl status openclaw

Restart=always means if it ever crashes, systemd brings it right back — the whole point of a 24/7 assistant.

Step 6 — Reach it from your phone (reverse proxy + auth)

The Gateway has a browser Control UI. To reach it from your phone over the internet safely:

1. Put it behind Caddy or nginx with a free SSL cert.
2. Front it with basic-auth or OAuth so the world can't open your assistant.

Caddy makes the TLS part automatic:

openclaw.example.com {
    basic_auth {
        youruser <hash>   # run: caddy hash-password
    }
    reverse_proxy 127.0.0.1:<openclaw-port>
}

Point a cheap domain (or a subdomain of one you have) at the VPS's IP, and Caddy issues the cert. Now your phone reaches the assistant at https://openclaw.example.com — encrypted, and only you can in.

(A simpler alternative: SSH tunnel — ssh -L 8080:localhost:<port> your-vps and open localhost:8080 in your phone browser. Zero extra setup, no public exposure at all.)

Step 7 — Point your channels at it

Now the satisfying part. Connect the chat apps you already use:

• Telegram: create a bot via BotFather, drop the token into OpenClaw. Message the bot — that's your assistant.
• WhatsApp: link a dedicated number (the two-phone setup OpenClaw recommends: one number just for the assistant, so your personal chats stay private).
• Slack / Discord / Signal / iMessage: OpenClaw supports these through its channel plugins.

From that moment, your "assistant" lives on the VPS and is reachable from anywhere.

Keeping it healthy

• Back up the workspace (~/.openclaw) — that's the memory and config you can't get back. A nightly rsync or tarball to object storage is enough.
• Watch disk: logs and sessions grow; add a small rotation.
• Update deliberately: OpenClaw moves fast, but test an update in a clone before rolling it onto the box your assistant runs on.
• Credentials: keep API keys in env vars or a secrets file with 600 perms, never in the repo.

The honest bottom line

The VPS itself is the cheapest part — a few dollars a month on a provider like Hostinger (referral code OVDAMRMASPIM). The real work is the setup: Node, systemd, a locked-down firewall, and a reverse proxy in front of the Control UI. Do it once and you have an always-on, model-flexible, private assistant that meets you in the apps you already use. That's the whole pitch of self-hosting OpenClaw, and it's genuinely achievable in an afternoon.

This guide is part of the OpenClaw how-to collection on ClawDesk — 90+ guides on installing, configuring, and hosting OpenClaw. Ready to host it? A cheap VPS via Hostinger OVDAMRMASPIM is all you need.