(Credit: Zooey Liao/PCMag/OpenClaw)
Most AI agents claim they can do it all—and most of them run into the same limitations. OpenClaw promises something different: a fully customizable AI sandbox that runs locally and can control your browser, apps, and files. On paper, it sounds like the ultimate power tool. In reality, it's a lot messier.
After spending time testing OpenClaw firsthand, I ran into steep costs, serious security concerns, and a setup process that feels more like a project than a product. More importantly, I struggled to find a reason to keep using it at all. Here's what stood out.
What Makes OpenClaw Different Than Other AI Tools?
OpenClaw is technically an AI agent, but unlike others, such as ChatGPT Agent or those in AI web browsers, it runs locally on your device. Beyond that, OpenClaw is more a collection of features and tools you can set up to work with an LLM of your choice as an AI agent than it is a ready-to-use AI agent like ChatGPT Agent. It's much like the difference between building a gaming PC and buying a PlayStation: A PlayStation isn’t as flexible or powerful as a gaming PC, but you can set up and start using one in minutes for a lot less money.
I can easily enable ChatGPT Agent on my desktop or mobile device, ask it to find me a recipe, and have it add the necessary ingredients to my Instacart cart. But if I want to do the same with OpenClaw, I need to install it, acquire and input an API key for my LLM of choice, install a skill to enable OpenClaw to control my web browser, start my OpenClaw instance, and then ask it to do the same thing. Even then, depending on my model choice (and usage limits), OpenClaw might require additional tweaks to actually work.
(Credit: OpenClaw/PCMag)Why use OpenClaw at all, then? Well, OpenClaw can do countless things other AI agents can’t, which you can learn more about on OpenClaw’s Shoutouts page. For example, it can control your home’s air purification system or remove watermarks from downloaded videos. Since OpenClaw can access the internet, control your computer, and run on your local hardware, everything from editing images to setting up a fully functional website and more is theoretically possible.
The prospect of OpenClaw sure sounds tempting, but four main problems emerged in testing that killed my enthusiasm for it.
1. Setup Is Where Things Start to Break Down
The first hurdle with OpenClaw is its complex setup. Before you even start installing it, you have to decide whether you want to put it on your personal machine or host it on a server. Going with the server route adds additional steps to the process. Installing OpenClaw means choosing the right model, finding API keys, optimizing prompts, setting up skills, tweaking settings, and more. You also need to be comfortable using a command line or, at least, navigating awkward, dense interfaces.
I don’t mind the complexity too much, but OpenClaw's general cumbersomeness turned me off from the start. Although setup guides can help, they don’t necessarily cover what you need to do to configure OpenClaw for your specific project. I found many guides that focused on making OpenClaw into a personal assistant you could talk to with your phone, for example, but I was more interested in using OpenClaw to control my computer.
2. The Hidden Costs Can Add Up Fast
OpenClaw doesn’t cost anything, but it's like a car without an engine until you connect it to an LLM. You can use pretty much any LLM you want with OpenClaw, but you can’t use many models for free. What free models are available have serious limitations. For example, after just a single prompt, I ran out of my free Gemini usage. Meanwhile, I couldn’t use Anthropic’s or OpenAI’s models for free at all. I also immediately hit Nvidia’s free Kimi model’s context limit, and I couldn't get the free OpenRouter model to perform the tasks I wanted.
(Credit: Hostinger/PCMag)The bottom line is that you will probably need to pay for an LLM. Depending on what you do with OpenClaw, usage costs can get expensive. For example, Hostinger estimates that setting up OpenClaw on a private server costs anywhere between $6 and $200 per month. That’s a huge range, but it’s practically impossible to know exactly how much OpenClaw costs until you figure out exactly how you will use it.
3. The Security Risks Are Hard to Ignore
According to Microsoft, OpenClaw’s potential to compromise its host environment, expose important information, and susceptibility to malicious memory modifications make it “not appropriate to run on a standard personal or enterprise workstation.” Even then, OpenClaw can make costly mistakes, like when a Meta researcher accidentally deleted their own emails. I ran OpenClaw on a dedicated machine to avoid security issues, and I didn’t use any important accounts with it, but you might not have that luxury. And if you do use it on your personal computer or with accounts you actually care about, you risk things going very wrong very quickly.
4. The Hardest Part: Finding a Reason to Use It
If you figure out how to use OpenClaw cheaply, get through the setup, and address its security concerns, you still need a good enough use case to make all that worth it. If you look around online at what most people rely on OpenClaw for, it’s often incredibly niche, such as controlling a 3D printer or visualizing your morning briefings in a scene. I couldn’t come up with a super compelling reason to use OpenClaw regularly, though you might have something in mind from the start.
More Experiment Than Everyday Tool
If you have a free Saturday, a little spending money to blow, and a spare PC, messing around with OpenClaw can be fun. You might even stumble upon something useful it can do. However, I won't blame you if you can't look past its complexity, costs, and security risks. I don't intend to keep it running, either.