Key Takeaways
- Moltbot (also known as Cloudbot or OpenClaw) is a viral, open-source AI agent from Austria that is installed locally on a computer and is designed to autonomously handle tasks.
- The bot can run on various devices – from a Mac mini to an older Windows PC or a Docker container.
- The central security problem: in productive use, Moltbot receives access to emails, calendar, and files – a considerable security risk that almost nobody fully audits.
- Open source does not automatically mean secure: who actually reads through the entire codebase and every update patch for the extension skills?
- As a safer alternative for Mac users, Thorsten Jekel recommends Claude Code and Claude Co-Work from Anthropic, which only access explicitly approved folders.
- Microsoft 365 with the paid Co-Pilot version offers many of the desired functions (calendar, email, summaries) within clearly defined security boundaries.
- For automation, proven tools such as Zapier, Make.com (formerly Integromat), and n8n are the recommended alternative to standalone AI agents.
- n8n is a Berlin-based company with a more modern technology stack; it can be hosted with German providers (e.g. Hetzner, Hostinger) or in your own data centre.
- Guardrails for AI productivity: always check data privacy and data security (cybersecurity) first, then evaluate the productivity gains.
- Conclusion: "Think first, then act on the technology" – free tools can become costly when security risks are not carefully assessed.
Introduction
Welcome to another episode of TJ's Technology Tuesday, where we always tackle current and fundamental questions about how you can use technology simply and effectively. You know I am a strong believer that productivity emerges beyond the hype – and one very hyped topic right now is "Moltbot".
Moltbot – an AI bot that claims to do everything
For those who have not yet heard about it, let me show you one of the typical messages currently circulating heavily on the internet: "Moltbot is an AI bot from Austria that is suddenly causing a stir worldwide." I have even had customers call me and say: "Jekel, you and your funny Co-Pilot and all that stuff – we want to do it ourselves. We want to do it securely. We want to do it with Moltbot. It can do everything." So the question is: what is the idea? The idea is – and perhaps you have already come across this – that you can install Moltbot on a separate computer. It does not necessarily have to be a Mac mini, by the way, because many people say: "Well, then I need to get a Mac mini for that." The Mac mini is, in general, one of the most cost-effective machines for AI in terms of price-to-performance ratio, because its architecture is simply very well suited for AI processing and it offers an excellent price-to-performance ratio.
But you could just as well install Moltbot on a relatively old Windows PC. You need a bit of memory, but you can install it – even on a Raspberry Pi, though it would be a bit slow. You can also run it inside a Docker container. I am not a huge fan of that approach, and I will explain why in a moment. Perhaps before I get into the critical topics, let me first clarify: what exactly is it? Do not be confused: it used to be called Cloudbot, then Moltbot, and now it is called OpenClaw – various names, because the company Anthropic, which makes Claude, was not very happy that the name came quite close to their own brand. So now it is officially called OpenClaw: a viral, open-source AI agent. That means the idea is that it operates autonomously. It can do various things, it can remember information, and it can be extended through so-called skills. Great concept. In other words, I install this thing on a separate computer, and ideally – and this is where I start to get a little nervous – I give it access to my files, my calendar, my emails; it answers my emails, and so on.
Data security and cybersecurity are a massive issue!
And that brings me straight to what gives me real pause: the topic of security. Because if I look at this – Heise, for example, which I consider a very reputable source – it is just one of many sources pointing to massive security problems. When I have a brand-new technology that is open source, that means anyone who can read code can see exactly how it works. But who among those who have installed Moltbot or OpenClaw actually reads through the entire codebase before deploying it? It is a bit like the general topic of "terms and conditions" – which nobody reads either. As you can see in one comment: "Nobody reads the terms and conditions." So right at the foundation, there are already massive security vulnerabilities that can be closed – but only if you know what you are doing. And then there are the skills, the extensions. It might be that the first skill is perfectly fine, with nothing suspicious in it. But then updates arrive, and of course you would need to check the code again with every single update.
Does anyone actually do that? Nobody does. So you are left with an extreme security risk built right in. And simply granting open access to your email inbox – email is the most critical asset because it can be used to reset all your passwords. Quite apart from the fact that many people have stored their credentials and email passwords in subfolders of their inbox. So this is something where I say: "Yes, productivity – absolutely. But for me there are always two guardrails: on one side, data privacy, and the other, even more important one, is security – data security." That is why, when a customer asked me about this, I said very clearly: No, I am not doing it with Moltbot, because the security risk simply cannot be adequately contained. So I am really not a fan. If you want to explore going in that direction at all, and especially for those who work with Macs, there is a very interesting alternative concept: Claude Co-Work.
What is Claude Co-Work?
Specifically, Claude from Anthropic. I once asked Claude itself: what is Claude Co-Work? Let me make the text a bit larger. It is a beta product, a desktop tool that is currently available exclusively for Macs. Here on the Mac you can do quite a lot. That means you can also grant direct file access, you can run sub-agents, similar things to what you might want to do with Moltbot, and you can work across longer time periods. And what existed before this – and I ask myself: what is Claude Code? – was essentially its predecessor in a way. There is the possibility to work directly in the terminal – a terminal environment, where many people say: "That gives me the creeps, it looks like something for programmers." In the terminal I can simply get things done directly. That sounds technical at first, but I last programmed 40 years ago. So I am not a programmer. As a schoolchild I programmed in assembler – I was quite good at it back then. But what I still know is that I still have a solid basic understanding of how programming works.
And in this terminal, one example of what I can do is this: I have a folder structure, I drop something into a folder, and then I can always tell the terminal what to do with it. Let me give you a few examples. I dropped something into a folder – specifically, a video from a podcast episode. And then I told Claude Code in the terminal: "I would like you to turn this into a LinkedIn carousel post with ten slides. At the beginning or end, please always include a photo of me that I have also uploaded to that folder – with a black background, text in the style of the Apple website, and so on." After three or four iterations, it did this beautifully. And then I said: "Now build me an automation so that whenever I drop a new video into this folder, it automatically creates a carousel file and includes a draft text suggestion." And that works brilliantly. So I drop in a video, it says it is starting a process that is also in that directory, waiting for me to drop in a video.
Then it creates the carousel post – a PDF with ten pages, writes the accompanying text, closes the application again, and I can go into that directory and copy out what I need. The great thing is that I only gave it access to that one folder, not to all my folders. Whereas with Moltbot, you can configure everything if you know how to do it – but by default it is far more open. Now, Anthropic has said: Claude Code is somehow a bit of a challenge for some users, so for those who find the terminal a little daunting, there is now – currently for Mac users – the option to download the application and get the classic Claude chat interface. There is also Code mode, which lets you go properly into the terminal. Or there is the middle ground: Co-Work. What Co-Work can do, for example: I can say, if I grant access to a folder, I told it: "Help me organise my downloads folder – analyse the current contents, rename them consistently, organise them into sensible subfolders, name those consistently, and optimise the approach as needed."
Then I gave it access to my downloads folder. And it beautifully sorted and structured my entire downloads folder. These are exactly the kinds of things where – when people talk about Moltbot or Cloudbot – it is also not trivial to set up properly. It is genuinely difficult to get the whole thing secure. And when I look at it, many of the things you want to do with complex agents, I can do today in voice mode with Co-Pilot simply in a conversation. I can do them with a Researcher that has access to all my information. I am a huge fan of the Researcher because I can simply access everything. And even when I am in normal voice mode, I can talk to M365 if I have the paid version. We often say: "I do not want to spend money." I assume, however, that your own services are also not offered entirely free of charge. I think offering good services for a fair price is not unreasonable. And Moltbot – yes, it is free, it costs nothing, the skills cost nothing too. But honestly, for me the security risk is simply too high. If you do want to explore automation, there is something worth looking at.
Proven automation tools such as Zapier, n8n, or Make.com
We have talked about these here a few times already. One option is working with Zapier. Let me pull up the page again. The idea of Zapier is that you have a data hub here. I use it myself, for example, so that when someone signs up for my email newsletter, they are automatically added to my CRM system. And these are the classic "if this, then that" automations that have been around for a very long time with Zapier. So you do not need to over-complicate things. The second option is Make.com, formerly known as Integromat. With that I can build these kinds of workflows. Honestly though, I have increasingly become more of an n8n person. Why? Because professionals who dive deeper into this topic – and you have surely seen these workflows on social media – say: "This is the newer technology stack." Make has been around for a very long time, growing out of the acquisition of Integromat. I once had the head of development at Make as a fellow speaker at a conference, and he openly admitted: "n8n has the newer technology stack."
Make.com has the great advantage that it is a little easier to use. I would say it sits somewhere between Zapier and n8n. You can get something up and running relatively quickly. n8n, on the other hand, looks simple at first glance but is already a bit more complex. The idea behind it, though, is pretty brilliant – nicely visualised here: you have a trigger, then an agent, and to that agent you can attach multiple things. You can attach a model – Anthropic, i.e. Claude, for example. You can attach a memory module so it remembers things, meaning next time it knows what you did before. You can even connect things like your employee directory. You can attach a calculator, for example – because AI models do not actually calculate. They output the most probable result. But then there is a calculator module. So you can say: "When it detects it needs to calculate something, instead of going to the large language model, it goes to the calculator."
So there are various modules, and the clever part is that you attach them to the agent, and the agent itself decides: "When do I need what?" And you can combine this with well-structured automated workflows. So as always, there are various options. n8n has the charm of being a Berlin-based company. I will deliberately note here – I suspect that if the cheque is large enough, it might well happen that they get acquired by OpenAI or Google at some point. And I would not hold it against the founders, because if they never have to work again and have new capital for their next passion project on the other side, I do not find that unreasonable. Right now it is a Berlin company. You can host it either with n8n themselves, you can run it on German hosting providers such as Hetzner or Hostinger. I have it on Hostinger myself as one instance of n8n. Or you can even self-host it in your own premises. So there is also the option here – with n8n offering a much more, let us say, secure structure, including in the area of Claude Code.
Conclusion
The skills you can add to it are vetted by Anthropic. Yes, there may not be as many, but they are vetted. The same applies to n8n in that area. Against that backdrop, the same principle applies here: think first, then act on the technology. And if you need more information on this, or if you need C-level sparring as an executive thinking about what to deploy in your organisation, please do not hesitate to get in touch. Your sparring partner for executives.
Thorsten Jekel.
Frequently Asked Questions
What is Moltbot and what is it used for?
Moltbot – now officially called OpenClaw – is an open-source, viral AI agent from Austria that is installed locally on a computer. The idea is that it autonomously handles tasks, remembers information, and can be extended through so-called skills, for example to answer emails or manage calendar entries.
Why was Moltbot previously called Cloudbot and is now called OpenClaw?
The tool has gone through several name changes: first it was called Cloudbot, then Moltbot, and now it carries the name OpenClaw. The reason for the most recent change was that Anthropic – the company behind the Claude AI model – was not willing to tolerate the close resemblance to the name "Claude".
On which devices can you install Moltbot / OpenClaw?
Moltbot can be installed on a Mac mini, an older Windows PC, and even inside a Docker container. The Mac mini is frequently recommended because its architecture is particularly well suited for AI processing and offers an excellent price-to-performance ratio.
What security risks does using Moltbot involve?
In productive use, Moltbot receives access to emails, calendars, and files – and email is particularly critical because it can be used to reset passwords. In addition, the source code reportedly contains massive security vulnerabilities according to reputable sources such as Heise, and almost nobody fully reviews every update patch of the extensible skills.
Why is open source alone not a security guarantee?
Open source means the code is theoretically visible to everyone – but in practice, almost nobody reads through the entire codebase before deploying the tool. Especially with the extensible skills, you would need to check every update to see whether new malicious code has been introduced, which virtually nobody does.
What is Claude Co-Work and why is it an alternative to Moltbot?
Claude Co-Work is a beta desktop tool from Anthropic, currently available exclusively for Macs, offering similar functionality to Moltbot – including file access and autonomous sub-agents. The crucial difference is that you can deliberately limit Claude Co-Work to access only specific folders, and the skills are vetted by Anthropic – which significantly reduces the security risk.
What can Claude Code do and how was it demonstrated in the video?
Claude Code is the terminal-based predecessor tool to Claude Co-Work, allowing you to automate tasks directly from the command line. The video demonstrated how a podcast video is placed in a folder and Claude Code automatically generates a LinkedIn carousel PDF with ten slides – complete with a photo, styling, and a draft text.
Which automation tools does Thorsten Jekel recommend as alternatives?
As proven alternatives, Thorsten Jekel recommends Zapier, Make.com (formerly Integromat), and n8n. While Zapier and Make.com are easier to use, n8n scores with a more modern technology stack and the ability to host it with German providers or in your own data centre.
What distinguishes n8n from Make.com and Zapier?
n8n is a Berlin-based company with a newer technology stack than Make.com – as even the former head of development at Make confirmed at a conference. n8n allows you to connect an AI agent to various modules – such as a language model, a memory module, an employee directory, or a calculator – and can be operated in a privacy-friendly way with German hosting providers such as Hetzner or Hostinger.
Is it worth paying for Microsoft 365 Co-Pilot instead of using free tools like Moltbot?
According to Thorsten Jekel, yes: the paid version of Microsoft 365 Co-Pilot provides access to calendar, email, and summaries within clearly defined, vetted security boundaries. Moltbot may be free, but the incalculable security risks make it unacceptable for productive use in organisations in his view.
Tools & Resources Mentioned
- AI overview on digital4productivity.de – Further information on AI tools and their secure use
- Microsoft 365 on digital4productivity.de – Information on Co-Pilot and productive use in organisations
- Zapier – Automation platform based on the "if this, then that" principle, e.g. for newsletter-CRM integration
- Make.com (formerly Integromat) – Visual automation platform, easier to use than n8n
- n8n – Berlin-based open-source automation tool with a modern stack, hostable with German providers such as Hetzner or Hostinger
- Heise Online – Cited in the video as a reputable source for security reports on Moltbot/OpenClaw
