If you've decided to run your own AI assistant on hardware you control, this OpenClaw setup guide takes you from nothing to a working, always-on assistant. No vague hand-waving, no assumed prior knowledge — just the honest path from zero to running.
There are two ways to get there. You can install OpenClaw yourself on hardware you already own (the DIY path), or you can skip the setup entirely with a ClawBox that arrives with OpenClaw pre-installed and configured. Both end in the same place: a private assistant you own, running on your own machine, answering you over Telegram or a web UI. This guide walks through both so you can pick the one that fits your time and patience.
Before you start: hardware and requirements
Before you install anything, make sure your hardware can actually run it. The short version: OpenClaw needs at least 4GB of RAM, Node.js 18 or newer, and an always-on internet connection. It runs on Linux, macOS, or Windows.
That's the floor. For a setup that's genuinely pleasant to live with, aim higher: 8GB or more of memory, an NVIDIA CUDA GPU or a Jetson Orin Nano if you want fast local AI, an NVMe SSD for storage speed, and a machine you can leave running 24/7. An assistant that's only awake when your laptop is open isn't much of an assistant.
If you want the full breakdown of what to buy and why, read our OpenClaw hardware requirements guide before spending money. It'll save you from over- or under-buying.
Option A — Install OpenClaw yourself
This is the DIY path. It's the right choice if you already have suitable hardware, you're comfortable on a Linux terminal, and you enjoy owning every layer of your stack. Here's the high-level sequence. For the exact commands and current configuration details, always follow the official OpenClaw documentation — it's the source of truth and it stays up to date in ways a blog post can't.
- Get suitable always-on hardware. A small home server, a mini PC, a spare desktop, or a Jetson Orin Nano. Whatever you pick, it needs to meet the requirements above and stay powered on.
- Install a Linux OS. Ubuntu is the well-trodden choice and the easiest to find help for. Get it installed, updated, and connected to your network.
- Install Node.js 18+ and the prerequisites. OpenClaw runs on Node. Install version 18 or newer along with the supporting tools the docs call for.
- Get OpenClaw and follow the official docs to install and configure it. This is the core step. Download OpenClaw, then work through the official setup instructions to install dependencies and create your configuration. Don't guess at flags — follow the documentation.
- Connect your first channel. Link a messaging app (Telegram is the usual starting point) so you can actually talk to your assistant. More on this below.
- Choose your AI models. Run models locally on your own hardware, route to a cloud provider (Claude, GPT, Gemini) using your own API keys, or mix both. Local keeps everything private; cloud gives you more horsepower on demand.
- Set it to run as a background service. Configure OpenClaw to start automatically and stay running so it survives reboots and keeps working when you're not watching. An assistant that dies on restart isn't done yet.
Realistically, the DIY path is an afternoon if you know Linux and a weekend if you're learning as you go. Both are fine. The reward is total control. If you want a deeper walkthrough of building on this kind of board, our Jetson Orin Nano AI assistant complete guide goes further into that specific setup.
Option B — Skip setup with ClawBox
Not everyone wants to spend a weekend on terminal work — and that's a perfectly good reason to buy the shortcut.
ClawBox is the pre-installed option. It's an NVIDIA Jetson Orin Nano Super delivering 67 TOPS of AI performance with 8GB of unified memory and a 512GB NVMe SSD, sipping just 7–15W of power. It ships with Ubuntu 22.04 and OpenClaw already installed and configured, built by ID ROBOTS Ltd.
You plug in power and network, and it's running. No OS install, no Node setup, no service configuration. The work in Option A is already done for you. It's a €549 one-time purchase — you own the hardware and the software, with no subscription. For a lot of people the math is simple: a weekend of your time is worth more than the price difference, and you skip the part where something goes wrong at step 4.
Either way you end up with the same OpenClaw. ClawBox just removes the setup friction.
Connecting your first channel
Once OpenClaw is running, you need a way to talk to it. This is the moment it stops being software and starts being an assistant.
The most common approach is to link a messaging app you already use — Telegram, WhatsApp, or Discord — so you can chat with your assistant from your phone like you'd message anyone else. Conceptually, you create or authorize a connection between OpenClaw and the app, and from then on your messages route to your assistant and its replies come back in the same thread.
If you'd rather not connect a messaging app at all, you can use the built-in web UI and talk to it from a browser. Many people run both: web UI at their desk, Telegram on the go.
Adding skills and automations
A running assistant that only answers questions is underusing the hardware. The point of OpenClaw is what it can do once it's live.
- Browser automation. OpenClaw drives a real Chrome browser, so it can navigate sites, fill forms, pull information, and handle multi-step web tasks the way a person would.
- Scheduled and cron tasks. Set it to run jobs on a schedule — a morning briefing, a recurring check, a nightly report — without you lifting a finger.
- On-device voice. Talk to it and have it talk back, using local Whisper for speech-to-text and Kokoro for text-to-speech. The audio stays on your hardware.
- Email and calendar. Connect your inbox and calendar so it can read, summarize, and help you manage your day.
If you want ideas for stringing these together into something genuinely useful, our guide to building local AI workflows with ClawBox automation shows real examples.
Frequently asked questions
How hard is it to install OpenClaw?
If you're comfortable with a Linux terminal, it's a manageable project — install an OS, install Node, follow the official docs. If you're not, the steps are doable but the learning curve is real. That's exactly what ClawBox exists to remove.
Do I need to be technical?
For the DIY path, yes — basic Linux and command-line comfort helps a lot. For ClawBox, no. You plug it in and start messaging. Pick the path that matches your comfort level, not the one that sounds impressive.
How long does setup take?
The DIY install is roughly an afternoon if you know your way around Linux, or a weekend if you're learning. ClawBox takes minutes — power, network, done.
Can I move my setup to ClawBox later?
Yes. It's the same OpenClaw underneath, so starting DIY and moving to a ClawBox later (or vice versa) is a normal path. Plenty of people experiment on a spare machine first, then buy a ClawBox for the always-on production setup.
Do I have to use cloud AI models?
No. You can run models entirely locally for maximum privacy, route to cloud providers with your own API keys, or mix both. The choice — and the keys — stay yours.
Why run my own assistant instead of using a hosted service?
Privacy and control. Your data stays on hardware you own. Our self-hosting AI complete guide to privacy and control makes the full case.
Ready to get started?
You've got two honest paths to a running OpenClaw assistant. If you want the hands-on route and own suitable hardware, follow the official OpenClaw documentation and work through Option A — it's genuinely satisfying to build.
If you'd rather skip straight to a working assistant, a ClawBox arrives with everything done for €549. Plug in power and network and you're running.
Either way, you end up owning your assistant outright. Get started at clawbox.tech and pick the path that fits you.