OpenClaw Self-Hosting Guide for Beginners: Every Install Path Compared (2026)

Transparency note: this article includes an affiliate link to Hostinger. If you sign up through it we earn a small commission at no extra cost to you.

OpenClaw is an open-source personal AI assistant that runs on hardware you control and answers across the messaging apps you already use — WhatsApp, Telegram, Discord, Slack, Signal, iMessage, and 15+ others. Because it's local-first and MIT-licensed, the OpenClaw self-hosting story is the whole story. There is no managed cloud to sign up for; you decide where it lives.

That freedom is also the biggest source of beginner confusion. Should you run it on your laptop? A Raspberry Pi in the closet? A Mac mini? Docker? A VPS? This guide walks through every realistic install path for OpenClaw, the honest trade-offs of each, and the single default we recommend for most first-time self-hosters.

If you already know you want a VPS, skip ahead to the step-by-step Hostinger deployment guide. If you're still choosing, keep reading.

What Self-Hosting OpenClaw Actually Means

Self-hosting OpenClaw means running the OpenClaw gateway (a Node 24 process) on a computer you own or rent, with your data, memory store, and API keys living on that machine rather than in someone else's cloud. The gateway connects outbound to messaging platforms (Telegram, Discord, and so on) and to a model provider of your choice — Anthropic Claude, OpenAI GPT, or a local model. Your conversations don't route through an OpenClaw-branded service; the project doesn't operate one.

Practically, that means three things matter: the machine is online whenever you want to reach the assistant, it has enough RAM to run the gateway plus Docker sandboxes for skills, and its filesystem is backed up somewhere you can recover from. Everything else is configuration.

What You'll Need Before You Start

• A machine that can run Linux, macOS, or Windows (WSL2) with at least 4 GB of RAM available to OpenClaw.
• Node 24 (or Node 22.16+) and Docker. The onboarder installs most of this, but you'll need Docker available separately for skill sandboxes.
• An API key from Anthropic or OpenAI, or a local model server like Ollama reachable on your network.
• Bot credentials for whichever messenger you plan to use first — Telegram BotFather is the easiest starting point.
• About an hour, plus a bit of patience for the first pairing flow.

The Five Realistic Ways to Self-Host OpenClaw

Option 1 — Your laptop or desktop

Best for: evaluating OpenClaw, short experiments, and anyone who works from a single always-on desktop.

Why it's tempting: zero extra cost, maximum privacy, no new machine to manage. Install is a single command:

npm install -g openclaw@latest
openclaw onboard --install-daemon

Why it breaks in practice: laptops sleep. When the lid closes, the gateway stops, pairings go stale, and some messengers (WhatsApp, iMessage) don't reconnect cleanly. You also can't easily expose a long-running browser-control task if your network changes every time you move between coffee shops. For anything beyond tinkering, you want an always-on host.

Option 2 — Raspberry Pi or mini PC at home

Best for: homelabbers who already own the hardware and enjoy tinkering.

A Raspberry Pi 5 with 8 GB of RAM can run OpenClaw for a single user. A used Intel NUC or Mac mini works even better — you get more CPU, no ARM compatibility quirks, and usually enough RAM for browser control. The install is identical to the laptop path, just over SSH.

Trade-offs: your home IP isn't always-on the way a datacenter IP is. ISP outages, DHCP renewals, and the occasional router reboot all interrupt the gateway. You'll also fight residential IP reputation for some commercial APIs, and port-forwarding any webhook gets awkward fast. This path is right if you already love self-hosting; it's a bad first project if you don't.

Option 3 — Docker on anything

Best for: users who want isolation and easy teardown, regardless of host.

OpenClaw ships a Dockerfile and docker-compose.yml, with a sandbox variant and an optional Podman setup script for rootless environments. Cloning the repo and bringing it up is straightforward:

git clone https://github.com/openclaw/openclaw.git
cd openclaw
cp .env.example .env   # fill in your keys
docker compose up -d

Why it's great: everything runs in containers, including the skill sandboxes. Upgrades are docker compose pull && docker compose up -d. Rollbacks are a tag change. You get the same result whether the host is your laptop, a home server, or a VPS.

Why to be careful: Docker needs proper persistent volumes for ~/.openclaw and .env, otherwise you'll lose your memory store on the first container replace. Read the compose file before your first up -d.

Option 4 — A small VPS (the recommended default)

Best for: almost everyone who wants OpenClaw to actually be useful rather than a demo.

A budget KVM VPS solves the always-on problem cleanly. A 2–4 vCPU, 8 GB RAM plan on a provider like Hostinger, Hetzner, Contabo, or DigitalOcean is enough for a single-user assistant with browser control and a handful of skills. We walk through the full install on one specific provider in the Hostinger VPS deployment guide, and compare several options in the best VPS for OpenClaw roundup.

Why this is the default: it's cheaper than most people expect (often under the cost of a single streaming subscription per month), it removes every reliability concern laptops and home networks create, and the hardening story is well-documented. You trade "data on your own hardware" for "data on rented hardware under your admin control" — an acceptable compromise for a single-user assistant.

If you don't already have a VPS, a Hostinger KVM plan is the cheapest reliable starting point. That's an affiliate link; it supports these guides at no extra cost to you.

Option 5 — Managed container platforms (Railway, Fly.io, Render)

Best for: developers who want zero Linux admin and don't mind paying a premium.

Managed platforms hide the Linux layer entirely — push a repo, get a running container. They're pleasant to use, but you pay roughly 2 to 4× what a comparable VPS costs, and some don't support the persistent volumes OpenClaw needs for its memory store. Verify volume persistence before committing.

How to Pick the Right Path

Four quick questions:

1. Do you need the assistant online 24/7? If yes, rule out the laptop option. If no, anything works.
2. Do you already own a Pi / mini PC / home server you trust? If yes, use it. If no, renting a VPS is cheaper than buying a machine.
3. Do you want container isolation? If yes, go Docker on whatever host you chose.
4. Are you comfortable with Linux, SSH, and a firewall? If no, pick a VPS host with a one-click Ubuntu image and a browser SSH console. Hostinger checks both boxes.

For most readers, the honest answer is "I want it always-on, I don't want to buy hardware, and I'm new to Linux." That's a small VPS with the global OpenClaw CLI, managed by systemd. It's the step-by-step walkthrough we link to throughout this guide.

Step 1 — Prepare Your Host

Whichever path you chose, get to a Linux (or macOS) shell with your user ready and internet access. On a VPS that means ordering the plan and hardening SSH. On a home machine that means opening your terminal.

Install Node 24 and Docker:

# Node 24 (NodeSource)
curl -fsSL https://deb.nodesource.com/setup_24.x | sudo -E bash -
sudo apt install -y nodejs
sudo npm install -g pnpm

# Docker Engine (Ubuntu example)
sudo apt install -y docker.io docker-compose-plugin
sudo usermod -aG docker $USER && newgrp docker

macOS users can use Homebrew (brew install node@24 docker) or Docker Desktop.

Step 2 — Install OpenClaw

The onboarding CLI is the friendliest path. It handles workspace creation, model selection, and your first channel pairing:

npm install -g openclaw@latest
openclaw onboard --install-daemon

If you'd rather build from source or pin a specific commit, the alternative is:

git clone https://github.com/openclaw/openclaw.git
cd openclaw
pnpm install
pnpm openclaw setup
pnpm gateway:watch

Config lands at ~/.openclaw/openclaw.json; secrets go in .env. Keep both out of version control.

Step 3 — Pair Your First Channel and Model Provider

During onboarding, pick your model backend (Anthropic or OpenAI keys paste in directly; a local model needs its base URL) and one messenger to start with. Telegram via BotFather is the simplest — you get a token in under a minute, and the gateway picks up messages within seconds.

Keep DM pairing for unknown senders on. This is OpenClaw's default. The reasoning is blunt: your assistant has shell access, and an inbound DM is untrusted input. Letting strangers reach the gateway is an unbounded risk. Our security guide covers why in detail.

Step 4 — Make It Survive Reboots

On Linux, create a systemd unit so the gateway restarts on reboot and on crash:

sudo tee /etc/systemd/system/openclaw.service > /dev/null <<'EOF'
[Unit]
Description=OpenClaw Gateway
After=network-online.target docker.service
Requires=docker.service

[Service]
Type=simple
User=$USER
WorkingDirectory=/home/$USER
ExecStart=/usr/bin/openclaw gateway
Restart=on-failure
RestartSec=5

[Install]
WantedBy=multi-user.target
EOF

sudo systemctl daemon-reload
sudo systemctl enable --now openclaw

macOS users can use launchd; Docker users rely on restart: unless-stopped in compose.

Step 5 — Back Up Your Workspace

~/.openclaw holds your memory store, skills, and configuration. Losing it loses the assistant's memory of you. A nightly tarball to another location is enough for most people:

0 3 * * * tar czf ~/backups/openclaw-$(date +\%F).tgz ~/.openclaw

Add rclone or restic to push those archives off-box weekly. On a VPS, enable the provider's snapshot feature for a second layer.

Troubleshooting Common Issues

• Pairing hangs on Telegram: the bot token is correct but privacy mode is enabled — talk directly to the bot in a private chat, not a group.
• Docker containers constantly restart: your host is out of memory. Check docker stats and upgrade the machine, or disable browser-control sandboxes if you're not using them.
• Model provider rate-limits you: Anthropic and OpenAI both set per-minute and per-day caps on new keys. Raise the account tier if you hit them, or switch to a local model for high-volume automation.
• Gateway runs but doesn't reply: check journalctl -u openclaw -f for auth errors. Expired OAuth tokens (Gmail, Google Chat) need re-pairing through the wizard.

Is It Safe to Self-Host OpenClaw as a Beginner?

Yes, if you stick to the defaults and run on a hardened host. OpenClaw ships with sensible defaults: DM pairing required for unknown senders, Docker sandboxes for skills, and no public listeners unless you add them. Beginners get into trouble when they disable pairing to "just try something quickly," expose the gateway to the public internet without auth, or run it as root. Read the OpenClaw security risks article before your first pairing if you want the long version.

FAQ

What is the cheapest way to self-host OpenClaw? A machine you already own is the cheapest. If you don't have one, a budget VPS at roughly $5–$10 per month plus model API usage is the lowest realistic monthly cost for always-on operation.

Can I self-host OpenClaw on Windows without WSL2? The onboarder supports Windows, but the skill sandbox story is smoother on Linux or macOS. If you're on Windows, running inside WSL2 Ubuntu gives you the cleanest path.

Do I need to expose OpenClaw to the public internet? No. Most messenger integrations poll outbound or use websockets that the gateway initiates. You don't need to open inbound ports unless you're adding a browser admin UI.

Can I self-host without paying for an API? Yes — point the model backend at a local Ollama server. Quality and speed will depend on your hardware; a 7B or 8B model on a laptop with 16 GB of RAM is usable but not comparable to Claude or GPT-5 class APIs.

How do I move an OpenClaw install between machines? Archive ~/.openclaw on the old machine, copy it over, install the CLI on the new one, and rerun openclaw onboard pointing at the restored workspace directory.

What to Do Next

• Ready to put it on a VPS today? Follow the OpenClaw Hostinger deployment walkthrough.
• Shopping providers? Compare options in the best VPS for OpenClaw roundup.
• Still evaluating? Start with what OpenClaw actually is and does.
• Before you paste your first key, read the OpenClaw security and safety guide.