OpenClaw Demystified: From Installation to Architecture
OpenClaw is one of the most interesting open‑source AI projects to explode in popularity today. But to understand it, it helps to know the story of how it got its name — because that story also reflects the community momentum behind the project.
The Naming Journey
- Clawdbot — The project was originally released in November 2025 by software engineer Peter Steinberger as Clawdbot, inspired by the Anthropic “Claude” model and the project’s lobster mascot.
- Moltbot — After a trademark concern over the similarity to “Claude,” the community decided on Moltbot for a short period. The term “molting” evokes a lobster shedding its shell and growing bigger, symbolizing growth.
- OpenClaw — The final name that stuck is OpenClaw, chosen because it clearly reflects the project’s open‑source nature and its lobster‑inspired identity — and it cleared trademark checks.
In this blog, I’ll share my OpenClaw setup experience. Honestly, there are many good setup guide across the internet such as using mac mini, using a hosted services, by passing many complexity. I like to do at ground zero and build things from scratch and also don’t want to invest on a $800 Mac mini. Instead of a step-by-step guide (that would make this article huge), I’ll highlight key pre-requisites, workflow, and gotchas so you can get started efficiently. As such, you You would need to have some good experience with Ubuntu, npm, nvm, pnmp and Node.js. At least, you need to know where to pull the repo.
Pre-requisites:
- A cloud Ubuntu VM — with 2 vCPUs and 16GB RAM. This provides enough resources to run OpenClaw and its AI models. You can spin off Azure, AWS or GCP if you want.
- SSH client — I used PuTTY on Windows and ZOC on Mac to connect to my VM.
- Node.js environment — Install Node.js and npm, and verify their versions. Using nvm to manage Node versions is highly recommended, as OpenClaw works best on Node ≥22. I am using Node 24.
- pnpm — OpenClaw uses pnpm for dependency management, so install it globally after setting up Node.js.
- Root access — Enable root access for your ubuntu VM. For port forwarding and daemon installation, ensure root login is enabled on your Ubuntu VM.
Installation Overview
Once your environment is ready, the installation workflow looks like this:
- Git clone the openclaw repo at https://github.com/openclaw/openclaw.git
- Next step, cd to the openclaw directory and build the dependencies using npm/pnmp.
- Install the Openclaw globally using — npm install -g openclaw@latest.
- And, then, run openclaw onboard — install-daemon #this will install the gateway as a long running process in the background.
If step 4 is done correctly, you should see the following screen:
Answer yes and proceed to setup the configuration. I shall use quick start to learn and also allow me to make changes later.
Next choose the model, select openai API key. I chose openai/gpt-5-mini.
Select the channel, I am using whatsapp. A QR code would be given. You would then use the QR code to pair with your whatsapp link and phone number.
Once you are done, the settings are saved to workspace and you would be asked to add the skills. I choose not to do now and will add it later for security reason.
Once down, you should see this screen:
If you want to see the openclaw dashboard, you would need to do port forwarding using this:
ssh -N -L 18789:127.0.0.1:18789 root@ipaddressofyourUbuntuVM. You must run this at the desktop where you launch either Putty or ZOC to console to the VM. At your desktop, browser, launch 127.0.0.1:18789
And you should be able to see the openclaw control UI
If you don’t see this immediately but see some connection error. It’s fine. Just got to overview and input in the gateway token. If you miss it during the setup step, you can actually ask openclaw for it in whatsapp. Here is the message. First, I ask for the token and then ask him to introduce himself. I give “it” his name: “wiwiki68”.
Notice something weird? It is sending using my own whatsapp account. That’s why the message is on the same side (right). The reason for that is that OpenClaw doesn’t inject a bot to whatsapp. Remember, you scan a QR code with your phone to authorize OpenClaw. As such, it is essentially automating your own account to send messages to you. As not to feel weird, you can use another whatsapp account to talk to it.
Openclaw Architecture
Openclaw is actually a headless process running on your VMs. It has full permission to read your files, application and launch browser. The openclaw gateway is written in typescript and it is a long running process listening to the messaging channels. Openclaw enforce a serial execution by default. As such, every user i sassigned a specific lane.
OpenClaw is a powerful open-source AI assistant that’s safe to use if you stick to trusted skills and take basic precautions. Treat it like any flexible tool — use isolated accounts, isolated environment, limit access, and you’ll have a secure, versatile AI agent at your fingertips.
