ClawdBot is out of control

Do you remember that whole idea about one person building a billion-dollar company with no employees, no human workers? I think the clock on that has just started.

This beefy fella is Peter Steinberger. Andre Karpathy posted a huge post just a few hours ago about where coding is heading. Not just coding, just software engineering in general. He's asking what it's going to feel like in the future, like playing Starcraft, playing a factorial, playing music. Someone responds, "Tenx engineer can be a one-man army. There could be soon a one-person billion-dollar company." Andre responds tagging Peter Steinberger saying, "You can do it."

So Peter, he came back from retirement to mess with AI and help build the lobster to take over the world. After watching this video, I think you'll agree with me this is going to be the year of the lobster. Oh, also Peter is polyagent moras. Polyagent moras, meaning that he likes a lot of agents. He he does it with a lot of agents at the same time, if you know what I mean. And this is the thing he built, Cloudbot.

Now, obviously, this isn't sponsored. This is an open-source project. It's hackable, meaning anybody can contribute to it, use it. It's it's basically free. You might have to pay for the AI frontiers, their models, their tokens, API costs, but this thing is free. If you search for it on Google Trends, it's got one of the wildest like search curves that I've ever seen. It's just flat and then it's just vertical. Like the entire world is like, "What is this thing?"

So in this video, let's take a look at who this person is. Who is this man? What in God's name did he just build and what would it look like if Arnold Schwarzenegger was born in in modern times and instead of going into bodybuilding decided to go into AI? Here's a quick clip from Unicorn Mafia where Peter talks about his project Cloud Code. This is where I believe it was just first revealed to the world. And at the end, I'll show you how to use it, how to install it, and how to get cranking on the thing because I mean, really, you should be using this. This is a a glimpse into the future as you'll see. Take a look.

>> I'm Peter. I'm originally from Austria and my latest project is Cloudbot. I was playing with for a while like cuz I I use cloud code or at least I use cloud code also to manage my computer to do like I don't know random file management, everything. So, what would happen if we up the game a little bit? Plot is kind of like I don't know, Jarvis, maybe a little bit like Her. Uh, if you've seen those movies, it's it's an agent that runs on either your computer or in the cloud. You can install it on everywhere. It's just TypeScript and it can access pretty much anything you want. So, Cloudbot has access to my emails, my calendar, my file, uh, through Tailscale, basically my whole network, my cameras, the temperature of my bed. Uh, it can order food for me. It can pretty much do everything. Uh, it can also do MCPs if you're into that stuff. But why not show you? Say hi.

All right. So, so my preferred we support right now WhatsApp, Telegram, Slack, Discord. Somebody submitted MS Teams and I regretfully merged it. Uh, and also El Salo, um, Matrix and it has a whole plug-in architecture. So, basically, you can install an npm and add your own messaging service. So, there's more to come, but let's just say let's just make Claude say hi.

>> Hello everyone. I am Claude, the space lobster AI assistant. It is an absolute pleasure to meet you all at demo day. Remember, we don't exterminate, we exfoliate.

>> Yeah, I'm a Doctor Who fan and I went with a space lobster scene. So, it came up with exfoliate. And that's where we go. So, it also has access to my cameras. I can just ask you, show me the cameras in my London apartment. Any interesting emails? Oh, but actually this is much better cuz the beauty is you can access it from anywhere. Oh, yeah. I see. Um, I put them a little bit into private mode so you don't see too much. Also, my cleaning lady keeps like moving them to the background every time she's there. Uh, can you show us a picture of ourself? How do you see yourself? You can also do voice messages, of course, because it's much nicer. Uh, and maybe also a video. I think you made one before and I I Q.

So, so even you can you can hook it up in a group chat.

>> Here's how I see myself. A majestic space lobster foliating through the cosmos. Hard shell, soft heart, ready to exfoliate.

Oh, it was complaining because it was creating a video and apparently lobsters are like not safe for work. So, it was like it was like it was it was keeping blocked because lobsters are too dangerous with their claws. So, now it made a really cute little lobster and was like, "At least I got that." There's of course agents.md with all the basics system prompt. I I don't use cloud code underneath anything. I I use my own harness so I have more control over the identity loop. Then we have an identity file that's basically everything about me. Then we have a soul.md which is kind of like the core values and like how the how the agent should act. Then it also has an identity and the really cool thing is if you if you go on the website and you just want to try it out. Let's see if that works. I speak fluent bash star customer and aggressive tab completion energy. It starts with a blank canvas. Um, I put a lot of care into the the first run experience so you don't get a random AI that's boring. You can make a a sassy space lobster or smart fox or I don't know what we're going to do. Let's see. Let's see if it crashes.

This this project kind of took off. Um, if you if I can Oh, yeah. This is my GitHub. If you want to find me build way too much projects. I kind of I kind of had to build an army of CLI because I very much believe CLIs are better than MCPs. For example, while Mr. I forgot the name. Cloudflare talked about CLIs.

>> Matt, I just used MCPorter, one of my projects to build a CLI out of an MCP because turns out agents are really good at calling CLIs. CLIs scale really well. You can have like a few thousand or 100,000 on your computer and it still works and it's progressive disclosure. So I just have to know that there is a CLI. Then I call help on it. Then I get like what the CLI can do and then I call it. That's kind of like how Unix works. And you can just package every MCP in the world through McPorter into CLI. And yes, I totally got the idea from Cloudflare when I built it. They also did random [ __ ] like in the terminal. Why not? Uh, YOLO here. This is like the one of the Cloudflare MCPs through McPorter. You get a nice TypeScript API that your agent can call. But that's not the point. This was the the the speech you saw was through 11 Labs. So I built like a CLI for that. Uh, I have a CLI to summarize. So the model can like be context smart. I built a CLI for Google because freaking Google has no CLI to actually manage all your stuff. Uh, I guess I have to do everything myself.

This project is really interesting because usually when I do stuff on Twitter, I get a lot of response, but nobody really got it because it's like a a new product defining category. Like nobody really had personal agents before and and we're still too too much boxed into our thinking that coding agents are for coding when they're like really clever beasts and can do anything. For example, I tell you a story. I started this. It didn't it wasn't called a Cloudbot. It was called WhatsApp Relay because I just wanted something where I can like type in little commands from when I'm not at my computer to run at my computer. And I hacked it to get in like two hours. Very simple. But I I usually use prompts when I when I do anything. So I wanted to have image support. So I built something where if somebody sends an image from WhatsApp, I just pass on the file path and the the agent will be smart enough to like load itself. And then I was in Morocco and I was playing with it a lot. It was fun. And at some point I wasn't thinking and just sent a voice message. And then 10 seconds later the model just replied with the answer. And I'm like, "How the [ __ ] did you do that?" And the model was like, "Yeah, you sent a file, but there was no file ending because I didn't program it." So I looked at the headers and I saw that it was OPUS. So I used FFmpeg to convert it to Wave. And then I wanted to use Whisper, but you didn't have Whisper installed. But I found this OpenAI API key. So I sent it to OpenAI, translated it, and then here I am. And I'm like, "That's why I kind of saw like the potential like these beasts are like just incredibly smart at general purpose problem solving."

My god, you have no taste. What else can you do? Ah, the app is shitty. What else can you do? Tell us a little bit. Show us. Give us the full rundown. Maybe also tweet that you're a demo day. Of course, I have an embedded browser as well. So, so because giving a model full access to your whole system and your network is not crazy enough, I thought, can we up it a notch? So, I added a heartbeat function that just calls the the agent in every configured model like every 10 minutes get a heartbeat with the prompt that says "Surprise me." And the model can can either decide to send a token back like "heartbeat, okay?" So, like you don't see anything, or it can check up on you. So, I have like client apps, so the model knows if I'm working on my computer. Sometimes it looks at my screen and roasts me. It can also click because I have like I built PeekAbove so it can take over my full screen. Like sometimes I have like code editors open. I go to food. I'm like, "Is this done already?" And it would check my screen and type in "continue" for me. It has memory. So I would say still primitive. It's like I marked on file a day and then some vector search over it. But it works damn well, I would say. Here you can see it's now tweeting. This is its own browser. So I use like I use I started with Playwright but I hacked it and made it a lot better. So it happily checks the "I'm not a bot" check boxes. Then come on, you can find the post button. It checked me in on my British Airways flight because I I did a I did the keynote at a conference yesterday. I was sweating a little bit because to check in it actually had to find my passport on my Dropbox, get the key out and it did all of that and then it was like bitching about the shadow DOM. So, here we are. Did a tweet. Please don't do a second one. What else do we have here?

Yeah, I'll show you how to create your own. So, I recommend you only use like a really good model like Opus 4.5 has pretty damn well prompt injection. I hacked on this for a while. I saw everyone that I showed it in person was like hooked and on Twitter I got very immediate response. I was like, "What can I do to actually make people believe?" If you see the graph uh here, can you guess where where I figured out how to get the graph? I made a Discord and I just hooked my my bot into the public Discord. That was before I implemented sandboxing. So, I was sweating a little bit, but turns out Yeah, this is this is our Discord. We grew within like 10 days to almost 5,000 people. There's a whole bunch of of AI researchers and and CEOs in there like the PlanetScale and and Convex, a whole bunch of people who do models. So, it's it's a really it's a really cool place. Was a really crazy one. Yeah. But so, and people tried to trick it into like getting getting stuff out, but it was basically laughing at them the whole time. So, I wouldn't say this is this is safe. I'm sure eventually this will be prompt injected because nobody has a great solution for that yet. But to me, this is this is just insanely useful. I use it every day, not just to manage my life, but also as a co-worker. Like, I give it an open-source repository that's like, "Check this out. Play around with this." And and it tries the CLI, tells me, "Oh, no, there's a bug. I had to compile it." Blah, blah, blah. This is not better than this. It also like helps me to to keep track of my fitness goals. It has access to my health data. So, it it it knows when I'm when I didn't sleep well enough. Uh, it will [ __ ] at me. It will [ __ ] at me if my weight goes up. This is far better than any fitness trainer because the more information you give those little monsters, the more tailored it can be. Like it knows exactly that I'm a demo day right now. And it will keep remembering that at least bits of it if I remember it. So, I'll show you a little bit the first run experience because I'm quite proud of it.

Now, you all have to look away because I have to find the entropic token. And I know those those are in high demand. Entropic tried to cut us off because they now have their the Teemo the Teemo one co-work. Here we go. Enjoy. Then you can configure the channel. I built a lot of scales. It installs a a launchd demon on Linux on on Mac OS. It also works with Linux. Windows works a VSSL works better, but those works. And then the first run experience is for of course you can do it in web but I like the terminal more. I also built a TUI and and then it basically says, "Hello, who am I? What am I? What are you?" "You are Gollum." Let's see what's coming there.

>> Yes, he likes it. Precious. It feeds us.

And then it also picks an emoji because like it's it's kind of do this thing like I have the the space lobster emoji. Every every creature gets their own emoji and it I guess the the swampy creature is good. I like the swampy creature. And I'm Peter and I'm a demo in London right now. So there's like 100 people watching you. Yeah, it's bugging out. Let's see what it's saying. We like Oh yeah, the ring emoji. 100 presses in London. So this is basically the the startup. It has like a bootstrap.md again. Everything is very simple. You don't need high-tech for that. And you feed it with some information. It it will create the identity files, the soul files, and it's also allowed to update them. Uh, you're you're recommended that you put it in in Git, so it backs up its own its own work directory. It also has Chromes. It also has sub-agents, so everything that's a little longer, it can spin up sub-agents. It can it can remember you like I use it as an alarm clock and it has control of my my son speakers and also my my computer. So, so it can both check on how much did I sleep and then like slowly change the lights because I use Philips Hue and it has control to the lights as well. It can check on the camera if I'm actually awake or not. Some people say I'm building Skynet here but with with with claws and oh yeah, it also built a game. So, but [ __ ] you. So, this is my own app trying to trying to get me cookies. Make sound effects the space invader game. I I integrate quite deeply with with Tailscale. So, it has a has one of the features also is a web server. So you can just put files in a server. It'll automatically be distributed. So the model has doesn't have to do too much and you can totally use it as an experimentation. There's also native apps. Uh, you can put it on the wall like it builds a morning report with a custom custom dynamic interface where I have buttons. When I click a button, a prompt is sent back to the model and it updates the interface. Uh, that stuff is still a little bit slow, but it's a transitionary problem. Uh, I think this is a glimpse into the future.

>> All right, let's claw into Cloudbot and see exactly what this thing can do, which is a lot. But first and foremost, it comes with a warning. Like a legit warning. There's a reason why this thing wasn't made by the Frontier AI Labs, by Google, by Anthropic, by OpenAI. This is an open-source project. The reason why Labs won't release anything like this anytime soon is because I mean there's going to be a lot of people that are going to be absolutely wrecked by this thing. They're going to do something that they didn't know was dangerous. And it is dangerous and data is going to get lost. They're going to get hacked. All sorts of bad stuff is going to happen. As a lot of people have pointed out, there's going to be disasters coming. We'll touch on that a little bit later. Basically, what Cloudbot is is it's your 24/7 AI assistant employee that will take a stab at anything you want it to. You hook it up to all your tools, your calendar, your email. You can even run it on your sort of main computer where you have all your files and whatnot. And the sheer amount of stuff it can do is amazing. It can automate a lot of things. It can be incredibly incredibly powerful.

So, really fast, here's some things that people have managed to do with Cloudbot. So, here's Alex Finn. He just wanted to see what would happen if he texted Henry. That's what he named his Cloudbot to make a reservation for me next Saturday restaurant. When the OpenTable reservation didn't work, it used 11 Labs scale to call the restaurant and complete the reservation. As here, 99% of people have no clue. So, I hooked up my Cloudbot to 11 Labs. And yeah, I was pretty impressed how quickly and easily it's able to use the 11 Labs API to create a voice. So, while I haven't tried this particular thing, I man, I believe it. So since Cloud Code can be connected to 11 Labs to produce AI voice. It's connected to your calendar. It's able to it also I think asks to connect to Google Places. So I set the Google API there. So it's able to get data on actual restaurants and other businesses and locations. Stuff like this is very possible and it's only going to get better. So as Finn is saying here, Cloud Code can learn to do anything. You just ask it to learn it. You don't need to explain it. It'll figure it out. Few days ago he told it to learn how to talk. Here's an 11 Labs key. Learn how to use it. Same thing for being able to use a Twilio. I want you to be able to make phone calls for me. Please build skills for Twilio so you have a phone number. 11 Labs so you have a voice and a Deepgram so you can listen. Tell me anything you need from me. I literally just posted that into the Cloudbot. And uh, let's see. Let's see where it takes us. Maybe it'll be done by the end of this video. What's interesting is saying any particular voice you want me to use. I can clone yours if you want with a little smirking face. This is why we need AI safety to be a serious priority. I don't want this thing cloning my voice and just running off of it. Apparently, some people are using it to create CloudCode tomagois. So, it's an actual like a little representation of a Cloud Code doing stuff. So, people are using it to check incoming spam, remove spam, incoming messages. You can have it sorted by urgent and non-urgent. Sync to Google Places, create issues on GitHub, generate a nice PDF summary of car conversations. You can use it to impersonate yourself in a group chat with friends. It has its own one password vault it can read and write to. So it can save all the passwords. You can give it a bunch of passwords for specific things you want to have access to. It can give you a daily morning brief. Researchers people before meetings and creates a briefing docs. I think you get the idea. The sheer amount of stuff this thing can do is kind of mind-boggling because you have intelligence plus the ability to connect to all the stuff that you do online and in fact with these voice things actually call real-life people get reservations etc. can be your fitness coach yelling at you if you're not doing enough stuff. It connects to all of your fitness data. It can go on X and do research, find the right posts. It can run 24/7 and getting you all the stuff that you need.

As Ethan Malik said here, "Good night, Claude. I hope these projects are done by morning." I think that in the future the biggest productivity hack will be to make sure you have enough stuff that you've given Cloudbot or Cloud Code or whatever AI assistant we're using. Just before you go to sleep, make sure it has a a full sort of schedule and to-do list that needs to complete by the time you're awake. Give it a week's worth of tasks. Say good night and go to sleep. You wake up to massive amounts of progress, tons of completed tasks, reports, research completed, projects built, etc. It's like Christmas morning every morning.

One of the cool things it asks you to do, you don't have to, but you can hook it up to some messaging service. In my case, I used a WhatsApp. It was just the easiest, most available, but there's tons. You can use Slack, Discord, Telegram, like anywhere where you're chatting. You can have a separate workspace where you're communicating to Cloudbot, telling it what to do. It's responding back to you with what it's doing or taking care of some biological functions, texting back and forth, having it complete stuff for you. Here's at Davekiss. He's saying he rebuilt his entire site via Telegram while watching a Netflix embed notion to Astro. 18 posts migrated DNS moved to Cloudflare. Never opened my laptop.

Now you can of course install Cloudbot on your main computer, but that comes with some dangers. It also currently is native to Mac, so it works the best on Mac. A lot of people got to buying these Mac Minis and just having Cloudbot live on it, basically doing everything it needs on that device, which I got to say is kind of insane if you think about it. You got this physical object sitting somewhere on your desk, which is basically a 24/7 employee going about your business, which might be a thing in the future at some point where it gets too complicated. We have too many of these AI agents running around that we have to manage. Having some sort of a visual representation of each might be beneficial. Whether it's a character on screen or a a physical real-life physical object like this, I think it would help our brains kind of keep everything organized. Some people even run Cloudbot on old Android phones. So it looks something like this. Each one is basically running its own version. So you got three agents, at least three because a lot of them can spin off their own kind of sub-agents to run on various tasks. What can you use it for? Twitter research, market monitoring, and Telegram chat daily summaries. Private groups signal flashing to his main phone. So if you have an old phone sitting somewhere in your drawer, turn it into a 24/7 AI server. I'll link this down below if people want to run this, but Mac Mini energy junk drawer budget. Also, a lot of people are just purchasing virtual private servers online and kind of hosting it on there. So, instead of running it on your device or even a Mac Mini or a phone, you just go to one of these providers, set up your Cloudbot on there, you can have many, many different instances, and basically have it run in the VPS.

Here's a person saying, "Cloud disaster incoming. If this trend of hosting Cloudbot on VPS instances keeps up along with people not reading the docs and opening ports with zero authorization, I'm scared we're going to have a massive credentials breach soon and it can be huge. This is a basic scan of instances hosting Cloudbot with open and gateway ports and a lot of them have zero auth." There are some sort of counterarguments to this. This is a ZD. He's saying there's been some noise about Cloudbot gateways being exposed on the internet. He investigated. Here's what he actually found. He says he pulled all Cloudbot instances from Shodan, 450 gateways discovered via mDNS and scanned them for missing authentication. The results there were zero that were vulnerable. Zero vulnerable instances in the wild. The 10 publicly accessible gateways all properly configured with auth tokens. The other 429 aren't even reachable. They're bound to localhost as intended. He's got the full code and report available on GitHub. But it's important to understand that these are still vulnerable to prompt injections as the creator even said himself in that intro little clip that we saw. Somebody at some point is going to get prompt injected. Just that's just the nature of how these things work. We have not found a foolproof system how to defend against it. So kind of watch out for that. Somebody's going to get burned. There's a lot of liability here. That's why the AI labs don't want to be doing this. They don't want to be exposed to something like this. That's why it's open source. It's kind of at your own risk. And overall, I think it's going to move everything forward. It's going to help us figure out how to protect ourselves, etc. But there's going to be a few broken eggs, if you will. There's going to be probably some people that are going to lose out as a result of this. This is very kind of experimental area we're entering.

All right, so let's talk about how to set this up really fast. It's a It's a little bit complicated, especially if you're setting up uh all the skills and all this other stuff that you need. But as always, AI is your friend. Use your favorite chatbot, ask it questions, have it walk you through it. If you haven't done stuff like this before, expect to spend some time on it, maybe even come back a few times. I was not able to run it on Windows, so I dusted off my MacBook. It's the best experience on MacBook. It runs natively. You're going to need some model, some AI model. I used Anthropic, but you can use OpenAI or whichever one you want. I'm sure you can connect whatever you want to it. I only recall seeing OpenAI and Anthropic. I'm sure they have all the other models on there as well. And then you need as they say here, the mouth. So some sort of a messaging platform. You can use Telegram, Discord, WhatsApp or Slack. So I've tried hooking up every single one of these. Discord requires a developer portal setup. So you set up a new Discord, create a bot. You give it some permissions. It can be a little bit difficult again if you're not a developer. If you're a developer, if you're 10,000 hours, most of this will be fairly easy. But for the average person, Discord will be somewhat difficult. Maybe the same thing with Slack. I haven't used Telegram too much. I did have a WhatsApp just because I have to message some people that are only on WhatsApp. That was the easiest one from a technical perspective. If you just want to get up and running, that one's super easy cuz Cloudbot just shows you a QR code. You go into your WhatsApp, then you click on the QR camera button, whatever. You just hold it up and then you say "link this device." It's part of linked to devices and then you're pretty much done. If you ever forget how to install it, you can actually go to cloud.bot and actually has the quick start installation here for you. So you just need to remember this basically. So on your Mac device, just pull up Terminal and just open up Terminal. Then just type in the command to install it. Again, use one of your favorite chatbots if that helps if you're not familiar with this stuff.

All right, so just I wrote it in there and away it goes. I already installed mine, so how yours appears might be different. It just told me, "I've evolved. Try to keep up." And then the lobster thing. Okay, good. Good job, Cloudbot. If you want to double check everything is running fine and just type in cloudbot-version. So I'll put that in there. Let's see. Pro tip: you have to spell things correctly. You're welcome. There it is. So that's the version that we're running. And then run cloudbot onboard. By the way, I'll just post this down into description or I'll leave a link to a website that that is going to have this. I'll just post it on my blog so that you have the updated commands if needed. But this is what you would do right now. Cloudbot onboard. So type that in and that's going to walk you through that initial sort of onboarding process. So first and foremost, it gives you a security notice. It's saying, "Understand this is powerful and inherently risky." Right? So you'll have to select yes to continue. You can do a quick start or manual. So let's start with the quick start. Again, I already used it. So some of this stuff I already have it in there. So first and foremost, the model provider. We have OpenAI, Anthropic, Minimax, Quinn, Synthetic, Venice, Google, Copilot, OpenRouter, Moonshot, etc. So I went with Anthropic. I think most people will go with Anthropic just because of how powerful Claude Opus 4.5 is. But whatever you select, you know, you can always come back and add more models later. You are going to need the API key from Anthropic's website. So get that and put that in there. After you put that in, you're going to see all the models that that provider has. Likely what you want. By default, it should have a good one selected for you. So if you're using Anthropic, you'll have Claude Opus 4.5. So keep current should work well. This is where some people might run into some issues. So Telegram is great. I think that's the one it sort of recommends you do. WhatsApp just uses a QR link. Again, if you just if you have WhatsApp, you just want to get in there and get going. That might be technically the easiest thing to get started. I've tried Discord. You have to go into their developer portal. You do have to create like a a new bot. You have to give that bot permission. Then you got to give, you know, the secret code to this to Cloudbot. Google Chat didn't work for me yet. Slack works, but again, there's some back and forth API, stuff like that. And I haven't tried anything else yet. So again, the suggested one is Telegram and WhatsApp is just the easiest, the fastest cuz it's just a QR code. Configure skills now. So these are all the skills that you can attach. So they're saying preferred node manager for skill installs. So npm is probably the most common, I would guess. Bun. People say it's really good, but if you have the technical knowledge for it, so I don't know, maybe just hit npm unless you you know what you like and prefer something. Next is install missing skill dependencies. So you have one password, Apple Notes in here. You have a Go Places as well, Nano Banana Pro. So you can get up and running without any of these. So you can go hit space to select skip for now and then just hit enter. But obviously this unlocks a lot of use cases. So at some point do think about setting those up. Enable hooks. So, what I did here, I actually asked the Gemini 3 what what it would recommend. It said just select all three of them and then go. So, I'm I'm trying that for now. We'll see how well that's working and let me know if you have any better ideas. Leave it in the comments if you think that's that's foolish, but that's what I've started with. And so, I already did it. So, it's asking me, do I want to restart? I'll just say skip. And it's saying, how do you want to hatch your bots? Like, where do you want it to appear? Do you want to hatch it in TUI? That's the recommended or the web UI. So if you do TUI, then it just opens up like this right and it says, "Hey, what's going on?" So it's basically this kind of command line interface within a web environment where you can go back and forth of it or a chatbot interface rather and boom, you're pretty much done at least with the setup. There's tons more things that you can add that you can integrate with but that gives you kind of like the basics that you need to get started. So now you can message it through that that TUI that kind of a website that it opened up for you or on your phone if you set up WhatsApp or Messenger or Slack or whatever you did there. You can also start messaging it there and chances are during the process something's going to fail, something's not going to work right. So again, ask your favorite chatbot. It can walk you through it pretty well. I've used Gemini 3 to walk me through it and it does a very good job of of explaining everything and walking me through step by step how to install this thing.

So I kind of love this little meme, these clip that they've made me as a child. Commanding pixel armies. Me now, commanding AI agents, similar to Warcraft and Starcraft, Factorio, etc., etc. Homeworld. That was a good one. Anyways, I'll leave some links down below. Get involved. Start using this. This is this is getting real. Obviously, be careful your security. Be aware of the various attack vectors. Think about what can Cloudbot screw up for you if you know it does something wrong. And also your sort of security. Can there be a prompt injection that happens somehow? Just sort of be aware of the security. We're kind of in the wild wild west era, so it's it's exciting, but also, you know, somewhat dangerous. So, I'm in the process of messing around with this thing. I'll have more videos, more follow-ups later. Let me know what you think and I'll see you in the next.