AI Enthusiasts Are Running Clawdbot on Their Mac Minis, but You Probably Shouldn’t.
I’m a self-described AI skeptic. So far, I don’t see much need for all these AI-powered assistants, nor for many AI-powered features. In my opinion, the most useful apps are unobtrusive—the rest seem more suited to shareholders than real people.
And yet, AI proponents have a new tool that’s garnering a lot of enthusiasm, and it’s now dominating my feeds: Clawdbot. Can this agent-based AI assistant make me believe in its capabilities too? Spoiler: probably not.
What is Clawdbot?
If you’re active in the online AI community, you’re probably already familiar with Clawbot. For the rest of you , here’s a quick overview : Clawbot is a “personal AI assistant” designed to run locally on your devices, as opposed to cloud-based solutions (such as ChatGPT, Gemini, or Claude). In fact, Clawbot runs a variety of AI models, including those from Anthropic, OpenAI, Google, xAI, and Perplexity. While you can run Clawbot on Mac, Linux, and Windows, many users prefer to install the bot on dedicated Mac minis , which explains its viral popularity.
But there are other AI assistants that can be run locally—Clawdbot is unique in that you communicate with it through chat apps. Which app you use is up to you, as Clawdbot works with apps like Discord, Google Chat, iMessages, Microsoft Teams, Signal, Telegram, WebChat, and WhatsApp. The idea is that you “send” messages to Clawdbot, just like you would to a friend or family member, but it behaves as you’d expect from an AI assistant—except perhaps with more activity.
That’s because, while Clawdbot can certainly do what AI bots like ChatGPT can, it’s more designed for tasks that require active participation. In other words, Clawdbot can do things for you, running in the background on your devices. The bot’s official website advertises that it can clean your inbox, send emails, manage your calendar, and check you in for flights—though experienced users are eager to expand the tool’s capabilities.
Clawdbot works with a variety of apps and services you may already use. These include productivity apps like Apple Notes, Apple Reminders, Things 3, Notion, Obsidian, Bear Notes, Trello, and GitHub; music apps like Spotify, Sonos, and Shazam; smart home apps like Philips Hue, 8Sleep, and Home Assistant; and other major apps like Chrome, 1Password, and Gmail. It can generate images, search for GIFs online, view your screen, take photos and videos, and check the weather. Judging from its website alone, it has quite an impressive list of features.
The final important point is that Clawdbot boasts a claimed “infinite” memory. This means the bot “remembers” every interaction you’ve had with it, as well as every action it’s performed on your behalf. Theoretically, you could use Clawdbot to create apps, manage your home, or handle messages, all within the context of everything you’ve done before. In this sense, it would truly be the closest thing to a “digital assistant” we’ve seen at this scale. These assistants have largely been action-oriented—you ask the bot what you want to know or do, and it (hopefully) acts accordingly. But the ideal version of Clawdbot should do all of this for you without you even having to ask.
It’s not just fans who are talking about Cloudbot.
Not everyone is thrilled with Clawdbot, however. Take this user, for example , who jokes that after four messages, the bot booked a spot, and then, after six messages, was able to send a calendar invite, but it cost him $87 in Opus 4.7 tokens. This user concocted a story (at least, I hope it’s a story) in which he gave Clawdbot access to his stock portfolio and tasked it with making $1 million without making any mistakes. After thousands of reports, dozens of strategies, and countless scans of X’s posts, he lost everything. “But it was so wonderful!”
I particularly like this interpretation : “[I] made a tragic discovery using [Clawdbot]. There just aren’t many tasks in my personal life worth automating.” There are also a few snarky comments from users who seem anti-AI, such as this one , which features an unemployed Clawdbot user living in his parents’ basement asking the bot to complete his daily tasks for him.
As with everything AI, there are many thoughts, opinions, and criticisms, especially given how popular this new tool has become. But the main criticism seems to be that Clawdbot requires a lot of resources (in terms of hardware, processing power, and privacy) without offering anything significant in return. Sure, it can do some things for you, but do you really need a bot that books your flights or scans your email? The answer to this question, I believe, is up to each of us, but the “backlash,” if you can call it that, is likely coming from people who would answer “no.”
How to try Clawdbot
If you want to try Clawdbot, you’ll likely need some technical experience first. You can get started with the official Clawdbot GitHub page and the Clawdbot Getting Started Guide. According to this page, you’ll begin by running the Clawdbot setup wizard, which will configure you with a gateway, workspace, channels, and skills. This works on Mac, Linux, and Windows, and while you don’t need a Mac mini, that’s what Clawdbot users seem to use.
Let me disclaimer right away: Clawdbot and its configuration are beyond my expertise, and I won’t be installing it on my devices. However, if you have the necessary knowledge to follow these instructions, or are willing to learn, the developer has provided step-by-step instructions at the links above.
How safe is Clawdbot?
While I probably wouldn’t install Clawdbot on my device anyway, the privacy and security risks involved definitely put me off.
The main problem with Clawdbot is that it has complete control and access to any device you run it on, as well as any software running on it. At first glance, this seems logical: how can artificial intelligence perform tasks on your behalf if it has access to the apps and hardware needed to perform them?
However, the inherent security threat of any such program is related to suggestion injection. Attackers can undetectably inject their own AI suggestions into otherwise benign websites and programs. When your bot scans the text while performing your task, it intercepts the suggestion and, thinking it’s from you, executes it . This is the primary vulnerability of AI-powered browsers , and it can also affect Clawdbot. And since you’ve given Clawdbot control over your entire computer and everything on it… horror. Attackers could manipulate Clawdbot to theoretically send private messages to anyone, run malware, read and write files on your computer, trick you into accessing your personal data, and learn about your hardware for further cyberattacks.
In the case of Clawdbot, these clues can come from a variety of sources. They can come from messages from attackers in chat apps you use to communicate with Clawdbot, from the browsers you use to access the internet, and from plugins you run in various programs—to name just a few possible sources.
The Clawdbot website includes a security guide detailing ways to enhance security when using the service. The developer acknowledges that running an AI agent with shell access on your computer is “risky,” that it is both a product and an experiment, and that a perfectly secure configuration does not exist. However, the system does have built-in security features that serve their purpose and are aimed at limiting access to Clawdbot, where it can be accessed, and what actions it can perform. This may include blocking private messages, displaying links and attachments as “hostile” by default, reducing the number of high-risk tools, and using modern AI models with better protection against malware injection.
However, this whole idea is too risky for me, especially since I’m not sure I even need an AI assistant. I think companies think we want to outsource tasks like calendar management, messaging, and content creation to bots to save time on mundane to-do lists. Maybe some people think so, but I don’t. I want to know who’s contacting me and why, and I don’t want to trust AI to determine which messages deserve my attention. I want to write my own emails and know what events are on my calendar. I also want access to my computer. Perhaps some people trust AI enough to handle all of these tasks on its own—if thinking otherwise makes me a Luddite, so be it.