I Tested OpenClaw (Moltbot) for 72 Hours, and It Will Revolutionize Your Workflow |

Note: The tool mentioned in this article, formerly known by various names including Clawdbot, Cloudbot, and Moltbot, has been rebranded to OpenClaw. This article has been updated to reflect the new name and installation instructions. For the latest information, please visit the official website: openclaw.ai.

I have seen the future in the last 72 hours. I tested a tool that has completely changed my life. It was called Cloudbot, then Moltbot, and is now OpenClaw. It’s an AI employee that works for you 24/7. It can control a computer, possesses infinite memory, and can do almost anything a human can do on a computer. I was expecting this technology in two to three years, minimum, but it’s already here. And honestly, it’s frightening.

In this article, I will show you exactly how it works, how to configure it for yourself, and present several use cases that will certainly change your life. Towards the end, I will also explain why I am genuinely a bit scared of what will happen when the whole world discovers this tool.

What is OpenClaw?

OpenClaw is a free, open-source tool that you install on a computer. The concept is simple: you ask it to do things, and it thinks, responds, and acts directly on the computer. It has total control, which is what makes it so powerful and life-changing.

You interact with it via messaging apps like Telegram, WhatsApp, and many others, which are very easy to configure. For example, I set mine up with Telegram. I can send it tasks anytime, anywhere, even from my bed when I remember something.

It acts on demand, but it can also act autonomously. You could instruct it: “Every morning, send me a summary of what happened in the AI world.” It will find the information, synthesize it, and send you what’s relevant.

Key Features of OpenClaw:

It Remembers Forever: This is crucial. Every time you give it a task, explain something, or correct it, it remembers its mistake and learns. We have an AI that self-corrects, logs its issues, and applies that knowledge the next time. It’s like a real employee on your computer who learns and doesn’t repeat errors.
Infinite Memory: It stores everything in local files on the computer, not in the cloud. Its memory is limited only by your available disk space.
It Learns and Self-Corrects: As mentioned, you can tell it, “You made a mistake here. Fix it.” It will rectify the issue and won’t make the same mistake again.

The tool has a rich documentation site and is open-source. The developer has made it available for free, which is fantastic. To give you an idea of its recent explosion in popularity, the number of stars on its GitHub repository has skyrocketed in less than ten days. It’s truly insane.

A Live Demonstration

To give you a clear idea of its capabilities, let’s walk through a demo.

Yesterday, I tasked my bot with accessing my proposal management tool, PandaDoc, and downloading my latest invoice. If you’re an entrepreneur, you know the pain of downloading all your invoices from various tools every month. It’s a real chore.

The first time, it struggled a bit but eventually corrected itself and saved the process to its memory. Now, I’ll ask it to do it again.

Me to the bot: “Can you go and download my latest invoice from PandaDoc?”

I’ve given it access to my 1Password vault via a skill. The bot immediately navigates to the correct payment information page without asking for clarification. Yesterday, there was back-and-forth, but today it knows exactly where to go. It knows it needs to scroll to find the invoice.

It initially reports that the page isn’t scrolling, but I’ll let it figure it out. It searches on its own. It says the element exists but isn’t visible. “Let me find it,” it thinks. It scrolls, downloads the invoice, and places it in my Downloads folder. It succeeded on its own.

Me to the bot: “Can you save this process to learn for next time, or have you already done that naturally?”

It confirms it will save what we just did to learn for the future. Just like an employee, it improves with each task.

Now, for a step I haven’t done before. I’ll ask it to take this invoice and upload it to Dext, my communication tool with my accountant.

Me to the bot: “Can you please upload the invoice to Dext?”

The bot doesn’t know what “Dext” is. I’ll tell it to look in 1Password. When it accesses 1Password, it requires my authorization for security.

It successfully retrieves the credentials and opens Dext. It enters my login and password. This is its first time on this site, so it doesn’t know where to upload documents. It needs to click the “Add Documents” button. Let’s see if it figures it out.

It sees a previous PandaDoc invoice I had already uploaded. It finds the button, clicks it, and locates the document drop zone. It uses the file upload function and successfully sends the invoice. It’s now uploaded to Dext. How amazing is that?

The bot even waits for the processing to finish, noticing the spinner icon. Once the spinner stops, it confirms the task is complete.

Now, I could potentially tell it: “Every 24th of the month, go to PandaDoc, download the invoice, and upload it to Dext. You know how to do it now.” I never have to tell it again. This one use case saves me time and mental energy every single month.

Setup and Installation

OpenClaw is a personal assistant that takes control of a computer. Therefore, installing it on your primary machine can be risky. Here are a few safer ways to set it up.

Hardware Options

Dedicated Local Machine: The best setup, in my opinion, is a separate computer next to you, like a Mac Mini or a Raspberry Pi.
- Why local is great: It accesses the internet from your network, so it appears as you. This avoids captchas and blocks from sites like Google or LinkedIn, which often detect and restrict traffic from datacenter IPs.
- A Mac Mini is a powerful option. A Raspberry Pi is a very affordable alternative that works great.
Windows/Linux: You can’t install it directly on Windows. You need to use WSL2 (Windows Subsystem for Linux). On a dedicated Linux machine (Ubuntu or Debian), it runs perfectly.
Cloud VPS: You can install it on a remote server from providers like AWS, Hetzner, or Digital Ocean. However, as mentioned, this can lead to issues with websites detecting the datacenter IP.

For cloud deployments, a managed hosting solution can simplify the process immensely by handling the DevOps for you. I was in touch with one such provider, and they added OpenClaw to their one-click install marketplace overnight.

Installation Walkthrough

The tool was recently renamed from Cloudbot to Moltbot, and now to OpenClaw. The installation process reflects the latest name.

The installation is guided by a command-line wizard.

Launch the installer:

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

Acknowledge the Risks: The first step is to confirm you understand the power and risks of the tool you’re installing.
Choose an LLM: You need to connect it to a Large Language Model.
- Recommended: Anthropic’s Claude 3.5 Sonnet or Claude 3 Opus provide the best results.
- Alternatives: OpenAI’s GPT-4o, or even a local model like Minimax if you want to keep everything on your own machine for security reasons (this requires a powerful server).
Provide API Keys: To connect to a service like Anthropic or OpenAI, you’ll run a command on your local machine to generate a secure token.
```
openclaw setup token
```
This will open a browser window for authorization and provide you with a token to paste into the installer.
Connect a Messaging App: Choose how you want to talk to your bot.
- Telegram: The simplest to configure. You create a new bot by talking to BotFather on Telegram, which gives you an API key.
- Other options: WhatsApp, Discord, Slack, Signal, and many more are available.
Install Skills: Skills are competencies you add to your bot. The base bot has access to a browser and memory. Skills add more power.
- You’ll be prompted to install recommended skills. It’s a good idea to start with the openclaw-hub skill, which is a marketplace for even more skills.
Finalize Setup: The wizard will guide you through the final steps, including pairing your bot with your chosen messaging app.

Once set up, you can close the terminal window. Your bot will run 24/7 on its dedicated machine, always ready for your commands via Telegram.

The Power of Skills

You can see and manage all installed skills from a local web interface. The ecosystem is growing at a phenomenal rate.

Here are some of the incredibly powerful skills available:

onepassword: Access your passwords securely.
gog: Access your Google Workspace (Gmail, Calendar, Drive). You can treat the bot as a team member, have it manage your calendar, or summarize your emails.
self_improving_agent: Allows the bot to note its errors and correct itself. This is a must-have.
github: Interact with GitHub repositories.
cloud_code: If you have Cloud Code installed, the bot can use it to develop entire applications for you. You could ask it to build an app overnight, and it will connect to Cloud Code, write the code, deploy it, and have it ready for you in the morning.
summarize: Summarize content from URLs, PDFs, images, and audio files.
elevenlabs: Generate voice responses.
obsidian: Connect to your Obsidian vault to take notes and act as your second brain.

To install a new skill from the hub, the command is simple:

openclaw install [skill_name]

For example:

openclaw install summarize

The new skill is immediately available for your bot to use.

The Infinite Possibilities

You are beginning to see the sheer power of this tool. You can give it one-off tasks or recurring jobs. It can work in your browser, interact with remote applications, and even develop software for you.

Consider a scenario I encountered with a client who used a proprietary medical CRM with no API. It was a closed system. With a tool like OpenClaw, you can train it to use such an application just as a human would. It learns the interface and can then perform the work autonomously. The possibilities are truly infinite.

Why This Makes Me Nervous

I believe people are in danger. I truly do. I thought this level of automation was years away, but it’s here now. It can easily start replacing jobs. A large portion of the workforce could be eliminated by this technology.

But you cannot stop innovation.

There are two ways to look at this. You can try to fight it, call it bad for humanity, and wait for regulations that may never come or be effective. Or, you can adopt it, use it, reap the benefits, and improve your life, your work, and your position.

The choice is yours. To stay at the cutting edge, you must learn to use these tools.