How to Set Up OpenClaw: Step-by-Step for Beginners

Written by Ty Sutherland in AI Agents

a close up of a sign on a building

OpenClaw has 250,000 GitHub stars. To put that in perspective: it’s the most-starred software project in GitHub’s history. It went from zero to 60,000 stars in its first 72 hours. Jensen Huang stood on stage at GTC and compared it to Linux and HTTP, then asked the audience: “What’s your OpenClaw strategy?” Tencent built AI products on it for WeChat. NVIDIA built an enterprise layer on top of it. The Chinese government restricted state agencies from using it. And it was created by one Austrian developer — Peter Steinberger — as a hobby project, first published in November 2025 under the name Clawdbot.

If you haven’t set it up yet, this guide will walk you through everything you need to get started — what OpenClaw actually is, how to install it, how the skills system works, what API keys you need, and critically, what the security risks are that you should take seriously before you expose it to the open internet.

What OpenClaw Actually Is (And Why the Hype Has Some Substance)

OpenClaw is a locally-run AI agent framework that turns your existing messaging apps into an interface for AI. Instead of opening a chat window inside ChatGPT or Claude, you talk to your agent through WhatsApp, Telegram, Signal, Discord, or iMessage — whatever you’re already using. The agent runs on your machine, uses your own API key, and does things rather than just answering questions.

The “lobster way” tagline and red lobster mascot are charming distractions from what’s actually interesting here: the architecture. Most AI agent tools are wrappers around a single model with a fixed set of capabilities. OpenClaw is more like an operating system for agents. It has a skills system — modular directories with a SKILL.md file that define what the agent can do. There are 100+ built-in skills, and a community registry called ClawHub where people publish their own. You install a skill, and your agent can now do that thing. Think of it like npm, but for agent capabilities instead of code packages.

It’s MIT licensed, completely free, and brings-your-own-API-key, which means you’re not paying a subscription to OpenClaw — you’re paying directly to whichever model provider you use. It works with Claude, GPT-4o, DeepSeek, Gemini, or local models via Ollama if you want to run entirely offline.

The name has a strange history worth knowing. It launched as Clawdbot. Anthropic sent a trademark complaint — presumably over the “Claud” sound-alike — so Steinberger renamed it Moltbot on January 27, 2026. Three days later, it became OpenClaw. The project moves fast, and so does everything around it.

What You Need Before You Start

Before you touch the installation, get these things in order:

  • A machine to run it on. OpenClaw runs locally. This means a laptop, desktop, or a small server you control — not a hosted service. A modern Mac, Windows machine, or Linux box all work. A Raspberry Pi or home server works too if you’re comfortable with that setup.
  • At least one API key. You need access to one of the supported model providers: Anthropic (Claude), OpenAI (GPT-4o or newer), Google (Gemini), DeepSeek, or a locally running model via Ollama. If you’re just starting out and want to minimize cost, DeepSeek’s API is significantly cheaper than OpenAI or Anthropic at current pricing — though model quality trade-offs exist depending on task type. If you want full local and free, get Ollama running with a model like Llama 3 or Mistral first.
  • Node.js installed. OpenClaw requires Node.js. Install the current LTS version from nodejs.org if you don’t have it. Run node -v in your terminal to check if it’s already there.
  • A messaging app you want to use as the interface. Telegram is the most commonly documented option in the community because the Telegram Bot API is clean and well-supported. WhatsApp and Signal require bridge setups that are more involved. Discord works well if you’re already in that ecosystem. iMessage works on Mac.
  • Basic comfort with a terminal. You don’t need to be a developer, but you need to be able to clone a GitHub repo, run a few commands, and edit a config file. If “open terminal and type git clone” sounds unfamiliar, spend 20 minutes with a basic terminal tutorial before proceeding.

Installing OpenClaw: Step by Step

The installation process is straightforward once your prerequisites are in place. Here’s the core flow:

  1. Clone the repository. Go to the official OpenClaw GitHub page (search “OpenClaw” on GitHub — it will be unmistakable given the star count) and clone it to your local machine: git clone https://github.com/openclaw/openclaw.git. Navigate into the directory.
  2. Install dependencies. Run npm install inside the project directory. This pulls in everything OpenClaw needs to run.
  3. Configure your environment. Copy the example config file — usually .env.example — to .env. Open it in any text editor. This is where you paste your API key, choose your model provider, and configure your messaging interface. Take your time here. Do not skip the security settings.
  4. Set up your messaging interface. For Telegram: go to Telegram, search for BotFather, create a new bot, and copy the bot token into your .env file. The OpenClaw documentation walks through each messaging platform. Telegram is the fastest path to a working setup.
  5. Run OpenClaw. Execute npm start from the project directory. If everything is configured correctly, you’ll see confirmation in the terminal and your bot will come online in your chosen messaging app.
  6. Test with a basic command. Send your bot a message. Ask it something simple. If it responds, you’re running.

The whole process takes 15–30 minutes for someone comfortable with a terminal. It can take longer if you hit issues with Node versions or messaging app bridge configuration — the OpenClaw GitHub issues tab and the growing community on Discord are your best resources when things don’t work. Once you’re up and running, there’s a lot more you can do with OpenClaw beyond the basics covered here.

The Skills System: Where OpenClaw Gets Interesting

A base OpenClaw install is useful but not impressive. The skills system is where it gets genuinely capable.

Each skill is a directory containing a SKILL.md file plus whatever code or configuration the skill needs. The SKILL.md file defines what the skill does, what commands it responds to, and what context it gives the underlying model. When OpenClaw loads, it reads all installed skills and builds a capability map. Your agent now knows it can do those things.

Built-in skills cover a range of common agent tasks — things like web search, calendar access, file management, running code, and sending emails. The ClawHub registry is where the community publishes skills beyond the defaults. By March 2026, this ecosystem has grown substantially, with skills for interacting with APIs, automating workflows,

Ty Sutherland

Ty Sutherland is the Chief Editor of AI Rising Trends. Living in what he believes to be the most transformative era in history, Ty is deeply captivated by the boundless potential of emerging technologies like the metaverse and artificial intelligence. He envisions a future where these innovations seamlessly enhance every facet of human existence. With a fervent desire to champion the adoption of AI for humanity's collective betterment, Ty emphasizes the urgency of integrating AI into our professional and personal spheres, cautioning against the risk of obsolescence for those who lag behind. "Airising Trends" stands as a testament to his mission, dedicated to spotlighting the latest in AI advancements and offering guidance on harnessing these tools to elevate one's life.

Recent Posts