I started building Stumpy because I believed everyone would have an AI assistant they use every day. But I had a hard time getting the idea across. I’d say "you’ll have an assistant that can manage your email, set reminders for you," and people would say "I already have Gmail filters. I already have Google Calendar." "It can do other stuff." "What?" "Anything!" Nobody got it. I couldn’t sell the idea.

I could see the potential anyway. I sprinted for a couple weeks to get it ready. Friday I filed for my LLC, worked all weekend. Monday morning I’d add payments and launch.

Monday morning rolls around. Here comes Clawdbot.

In case you’ve been living under a rock: Clawdbot is an open-source AI assistant that runs on your machine. You can text with it through Telegram, WhatsApp, whatever. It’s made by a fantastic hacker named Peter Steinberger, and it is taking the world by storm.

People love it. And most surprising to me—everyone gets it. Suddenly the whole world sees how useful AI assistants are: they can do things in your life, oh my gosh, they can be AI employees. Yeah! Welcome to the party!

So it’s pretty similar to Stumpy, but let me talk about a few key differences.

You need a machine that you can leave running at home. If you have a laptop that you take places, it’s not going to be running on that. In fact, if you have a machine at all with anything sensitive on it, that’s a no-go. You’re going to want a second machine that you dedicate to running just Clawdbot. Maybe you’re one of the people picking up $600 Mac Minis for that purpose.

The second big thing is that since it’s running on your machine, it has access to everything: all your documents, your bank statements, your medical records, your browser history. So when you connect it to your email or Telegram or anything, you are one prompt jailbreak away from giving the world access to everything on your computer. It’s kind of a scary situation.

Stumpy agents run in the cloud. They’re Cloudflare Durable Objects, and optionally they have Fly Sprites attached for use as sandboxes, so they can still install arbitrary software and do work for you like building software or browsing the web. But this is running in an isolated environment; it only has access to what you explicitly give it access to.

Here I should acknowledge a shared vulnerability. When you connect either Stumpy or Clawdbot to a communication channel like Twitter or email, it opens up a potential for the agent to be manipulated and for your data to be exfiltrated or for its behavior to be hijacked for the attacker’s purposes. You need to be very thoughtful about when and how you connect these channels. Both services have that kind of vulnerability in common. That said, for the main communication channels we’ve built in protections, so no one can contact your Stumpy agent unless you configure it, and the platform can’t be used for spam.

A third area where the two platforms are different is ease of use. Clawdbot is a difficult and technical setup. It’s not a product for the general public. It’s something that fairly techy people can set up as a hobby.

Stumpy on the other hand is the easiest thing to use in the world. You can set up an agent in about one click. Then you configure it by talking to it. You can use it a little for free or you can pay to use it a lot. Life doesn’t need to be much more complex than that.

Overall, to be honest, I love that the world has Clawdbot. It has established the market. Everyone is hearing about this idea of having AI assistants in their lives. It’s setting people up to appreciate that kind of product, and it’s also teaching people about security and the care they need to take as they set up agents like this. It’s also a super cool project! I love projects like this and the people who create them. So I just want to throw some congratulations at Peter right now for building something amazing and putting it out into the world. Nice work dude.