Your Own AI Assistant, Running On Your Machine
OpenClaw is open source and it connects the model you already pay for, Claude, GPT, Gemini or something local, to the places you actually message from. WhatsApp, Telegram, Slack, Discord and the rest. We install it, connect your channels, lock the machine down around it and hand the whole thing over.
Installing it is a single command and the guides are honest about that. What takes an evening is the gateway staying up after a reboot, channels that stay paired instead of dropping overnight, API keys that are not sitting in a file anybody on the machine can read, and spend that does not quietly run away while you sleep.
Packages And What They Cost
Paid once, with no subscription attached to anything we hand over. Pick by whether you want it running on your own computer or on something that stays awake without you.
Desktop Install
On the computer you already use. Fine for trying it properly before deciding whether you want it awake around the clock.
- Windows, macOS or Linux
- Node installed at a supported version
- Gateway running as a service, starts on boot
- Your model provider connected and tested
- Two messaging channels paired
- Access limited to your own accounts
- Key permissions set properly
- Written notes covering restarts and updates
Always On Setup
On a small VPS, a Pi or a server you already run, so it answers at three in the morning and scheduled work actually happens.
- Everything in Desktop Install
- Running in a container so updates stay clean
- Up to four messaging channels paired
- Sessions stored so pairings survive a restart
- Firewall closed, key based login, root login off
- Gateway kept off the open internet
- Spend limits and alerts set at your provider
- A hand picking the host if you have not got one
Custom Build
Several people using it, a local model instead of an API, or work you want it doing on a schedule rather than only when asked.
- Multiple users with their own access
- A local model through Ollama where the hardware suits it
- Scheduled tasks and recurring jobs configured
- Connected to tools and services you already use
- Access to files and commands scoped deliberately
- Existing broken installs diagnosed and repaired
- Documentation for whoever else has to run it
Added To Any Package
- Each additional messaging channel $5
- Full server hardening beyond the assistant $50
- Moving an existing setup to a new machine $80
- A walkthrough session once it is running $30
- Support after handover, monthly $100 / mo
The software itself is free and open source, so there is no licence to buy. Your only running costs are whatever you pay your model provider for what you use, and the hosting if you go for an always on setup, both billed to you directly and never through us.
Where It Can Actually Run
The real question is whether you want it awake all the time or only when you are. An assistant that answers a message at two in the afternoon and ignores one at midnight is a different thing from one that is always listening, and that difference decides the machine rather than the other way round.
Windows
The desktop app handles most of the setup, and underneath it there is a Linux environment doing the actual work. That is where things go wrong for people, so we configure both halves properly and make sure the firewall prompt on first launch gets answered the right way rather than allowed on every network.
macOS Or Linux
The most straightforward of the lot, since everything it needs is already native. Node gets installed at a version it is happy with, the gateway runs as a proper service rather than a terminal window you have to leave open, and it comes back on its own after a restart.
A Small VPS
Somewhere between five and fifteen dollars a month buys a machine that never sleeps, never gets closed at the end of the day and is not affected by your home internet going down. Messages get answered at three in the morning, scheduled work actually runs, and your laptop battery is not involved.
A Raspberry Pi Or Mini PC
The always on option without a monthly bill, drawing about as much power as a phone charger. A Pi 4 or 5 handles it comfortably for one person. It sits on your own network too, which matters if you want the assistant reaching things inside your house.
In A Container
Running it in Docker keeps it separated from everything else on the machine, makes updates a matter of pulling a new image, and means an experiment that goes wrong is thrown away rather than untangled. Worth doing on a server, less necessary on a laptop you are only trying it on.
A Server You Already Run
If you have a box doing something else, a media server, a VPN, a small site, this goes on it as one more service rather than a reason to rent anything. It asks very little of a machine when it is sitting idle waiting for a message.
What it needs underneath
A recent version of Node, an API key from whichever model provider you use, and somewhere the gateway can keep running. That is genuinely the whole list. It is not heavy software, and the modest sizes people run it on are not a compromise, they are simply enough.
Where the model runs is a separate decision. Most people point it at Claude, GPT or Gemini and pay per use. If you would rather nothing left the machine at all, a local model through Ollama is possible, though that does want real hardware behind it and is worth a conversation before you buy anything.
What Actually Gets Set Up
Nine things, and only the first one is the install. The rest are the reasons a working setup on Monday is still working three weeks later without anybody touching it.
Installed at a version it is happy with
Node at a release it actually supports rather than whatever your machine already had, which is where most failed installs start. Everything kept under its own prefix so it is not tangled into other projects on the same computer.
The gateway runs as a service
Not a terminal window you have to leave open and not something that quietly dies when you close a laptop lid. It starts on boot, restarts itself if it falls over, and you can check whether it is alive without reading logs.
Your model provider connected
Claude, GPT, Gemini or a local model, with the key stored where the software expects it rather than pasted somewhere convenient. Switching provider later is a config change and we leave you the two lines that do it.
The channels you actually use
WhatsApp, Telegram, Slack, Discord, Signal or Teams, paired and tested. Some of these want a bot token, some pair with a code on your phone, and each has its own way of going wrong the first time.
Pairings that survive the night
A channel that drops at three in the morning and reconnects only when you notice is the most common complaint people have with this software. Sessions get stored properly so a restart does not mean scanning a code again.
Only you can talk to it
Access limited to your own accounts and numbers, so a group chat or a stray message from somebody else does not reach an assistant with access to your machine. This one gets skipped constantly and it is the one that matters most.
Keys that are not readable by everything
File permissions set so the credentials are not sitting in plain view of every process and every other user on the machine. On a shared or rented box this is the difference between a private setup and a shared one.
Spend you can see before the bill
An agent left to work through a long task can use a surprising amount of credit, so limits get set at the provider, alerts go somewhere you will read them, and you are told roughly what normal use looks like for your setup.
The machine locked down around it
On a server that means a firewall, key based login and the gateway not exposed to the open internet. On a home machine it means the network prompt answered correctly and the service reachable only from where it should be.
What it can reach, and why that needs deciding
This is an assistant with real access, which is what makes it useful and also what makes the setup worth doing carefully. Depending on how you configure it, it can read and write files, run commands and act on the messages it receives, so where it is allowed to operate is a decision rather than a default.
We go through that with you before anything is switched on, keep it inside a working folder rather than loose on the whole machine unless you want otherwise, and write down what it can and cannot touch so you are not guessing about it in a month.
Send The Error, Get An Answer Today
A screenshot of the exact failure, the resource it involves, and roughly when it started. That is enough to tell you what it looks like and how big a job it is, before there is anything to pay for. Anything already down goes to the front of the queue, and if it turns out to be a five minute fix you will be told that too.
Send us the error