Command openclaw not found

The install script put the binary on your PATH, but the terminal window you already had open does not know that. Open a new one.

The gateway does not answer on 18789

Something else may be holding the port. Check with `openclaw gateway status` and read the error — it tells you whether it is a port conflict or a service that failed to start.

The assistant does not answer in Telegram

The token is usually wrong, or the bot is talking to a different gateway. Confirm the gateway is running and that you used the token from the right bot.

Modified 2 August 2026

Getting started with OpenClaw

OpenClaw is a personal AI assistant you run yourself. It lives on your machine or your own server, talks to you over Telegram, WhatsApp or the terminal, and connects to whichever language model you choose. This guide takes you from nothing to an assistant that answers.

Official documentation: openclaw.ai

What OpenClaw is

The difference from ChatGPT in a browser is where it runs. OpenClaw is open source and installs on hardware you control. No third party sees the conversations, and you pick the model behind it — Claude, GPT or DeepSeek.

The architecture has two parts. A gateway, which is the engine, and channels, which are how you reach it. The gateway runs in the background and holds the context. The channels are Telegram, WhatsApp, Signal, Slack and a dozen more.

Before you start

Three things need to be in place:

Node.js 22.22.3 or later. Node 26 is recommended.
An API key from Anthropic, OpenAI or Google.
macOS, Linux or Windows. On Windows both native and WSL2 work.

Budget ten minutes. Onboarding detects model access you already have on the machine and verifies it with a real completion before continuing.

Where should it run?

On your own machine

Simplest place to start, and the right call if you mostly want to try it out. The drawback is that the assistant sleeps when your computer does — send a Telegram message with the lid closed and nothing happens until you open it again.

On a VPS — if it should answer around the clock

An assistant you want to reach while your laptop is shut needs a server that is always on. A small VPS is plenty: 2 vCPU and 4 GB RAM handles both OpenClaw and the gateway with room to spare.

Where the server sits matters. If you point the assistant at your mail and files, that data passes through the server — so you want it under EU jurisdiction, not somewhere authorities can request it without you ever knowing.

Inleed — Swedish provider with servers in Sweden and support in Swedish. Data falls under Swedish law.
Scaleway — French and EU-owned. More regions and more room to grow if the project gets bigger.
Hetzner — German, cheapest per GB of RAM. Higher technical bar — less hand-holding in the interface.

All three sit inside the EU. The difference is how close support is and how much you want to configure yourself.

The Inleed link is an affiliate link — we earn a commission if you sign up. Scaleway and Hetzner are not, and are recommended on merit.

Install

Step 1 — install OpenClaw

The install script fetches the binary and puts it on your PATH.

curl -fsSL https://openclaw.ai/install.sh | bash

curl -fsSL https://openclaw.ai/install.sh | bash

iwr -useb https://openclaw.ai/install.ps1 | iex

Step 2 — run onboarding

The wizard asks about model, workspace and channels. The flag installs the gateway as a service so it restarts on its own.

any OS

openclaw onboard --install-daemon

Step 3 — check the gateway

If this answers on port 18789, the install is done.

any OS

openclaw gateway status

Connect a channel

The gateway runs, but you can only reach it from the terminal. A channel makes the assistant reachable where you already type.

Telegram — fastest

Telegram needs only a bot token. Message @BotFather in Telegram, ask for a new bot, and paste the token when onboarding asks. To add it later, run onboarding again and pick just the channel step.

WhatsApp, Signal and the rest

WhatsApp and Signal require pairing a phone number, which takes a few more minutes. Beyond those there are Discord, Slack, Microsoft Teams, Matrix and iMessage. Same gateway, same memory — only different ways in.

Common problems

Command openclaw not found: The install script put the binary on your PATH, but the terminal window you already had open does not know that. Open a new one.
The gateway does not answer on 18789: Something else may be holding the port. Check with `openclaw gateway status` and read the error — it tells you whether it is a port conflict or a service that failed to start.
The assistant does not answer in Telegram: The token is usually wrong, or the bot is talking to a different gateway. Confirm the gateway is running and that you used the token from the right bot.

What to do next

The assistant runs and answers. The next step is giving it tools — read on about MCP, the protocol that lets agents use external services, or about keeping keys and permissions safe as the assistant gets more access.