Hi everyone — we are probably the biggest OpenClaw fans in the world. 😄 We were also one of the earliest teams to build a cloud-computer platform around OpenClaw. In this post, we want to share some advanced ideas about AI harnesses, agents, and automation systems. Even if you are not one of our users, you can still apply many of these ideas to your own workflow. Building a website has never been easier In the past, launching a production-ready website required developers, designers, hosting, deployment, integrations, testing, and a lot of time. Today, the entire process can potentially be compressed into one message: “Build a production-ready demo website for this business.” No coding. No manually setting up every tool. No spending weeks connecting different services. Inside SignatureClaw, we pre-installed and configured the MCPs, deployment tools, service-layer workflows, design skills, upload systems, and other capabilities required to keep building continuously. This means the agent does not only generate code. It can design, build, deploy, test, improve, and continue working 24/7. However, even if you are not using our platform, the bigger lesson is still extremely valuable: You can package your service layer, deployment workflow, design skills, upload process, and business logic into tools, MCPs, skills, or reusable agent workflows. The technical implementation is another topic. Today, we want to focus on the higher-level idea: How do you design a system instead of completing a task? From the screenshot, you can see that a production-ready website can already be compressed into one instruction. For example: “Build a demo website for a local fitness business.” But that is only Level 1. The next level could be: “Research a profitable website niche, build one new website every day, and send me the results.” It is still one message. But now the agent is no longer completing one task. It is operating a repeatable production system. Then we can go one level further:

I used Openclaw to build a custom Chrome extension that enables my WireGuard VPN only for specific work sites I select. I was getting so tired of routing ALL my web traffic through the VPN—it kept getting me blocked or flagged on everyday sites. Now I have a clean split-tunneling setup right in the browser. It's super useful and saves me so much trouble!

I find it incredibly difficult to decide whether I should run OpenClaw via my existing Gemini subscription, via Open Router, or, for example, via a separate subscription with Moonshot to use Kimi K 2.5. Which language models do you use, via which platform, and for what tasks?

I built and published a live Pokémon game website without writing a single line of code. Anyone can play it, and you can even save it as a shortcut app and play it directly on your iPhone. https://sc-zhn4pruh.vnc.signatureclaw.com/pokemon-gba.html Here’s how you can do it — with or without SignatureClaw. For SignatureClaw users, OpenClaw already comes with the necessary skills and tools to vibe-code and publish a website or game. Each user gets a unique HTTPS URL. Once you tell OpenClaw to publish your website, it handles everything automatically. If something doesn’t work, you can email their team and they’ll help you troubleshoot. For other OpenClaw users, you can set things up manually: 1. Install Nginx 2. Buy a domain and set up an HTTPS certificate (or use a service like ngrok if you don’t have a domain) 3. Disable all unused ports except 443 4. Vibe-code your web game and link it to your website folder That’s it — you can now publish your game without coding!