Why People Are Freaking Out About Clawdbot

Matt Wolfe walks through Clawdbot, an open-source tool that connects Claude to Slack and automates tasks like generating daily news digests and building apps. He shows how to set it up on AWS, a VPS, or a Raspberry Pi, covers the API costs and security risks like prompt injection, and demonstrates live workflows in Slack.
Chapters
Cloudbot is taking the world by storm right now and it is absolutely blowing my mind as well as the minds of everybody that's using it. It is an AI assistant that not only gives you advice and you can chat with like the other bots, but it will actually go and take action and do things on your behalf better than anything I've ever seen before. And it doesn't matter what model you're using, it can use like all of the models. Just in my own experimentation in this video, I got it to create a chatbot with me in Slack, create a daily AI news digest for me that delivers to SL every single day, install Claude code for me so that it can go and build apps for me on my behalf by me just giving it an app idea or in some cases I tell it a problem I need solved and it will go build an app for me. I got it to install Remotion and then generate a simple animation for me. And I got it to completely dial in a virtual private server for me so that it could run itself. Like it basically went and gave itself its own instructions to build itself. It is wild. You're going to have to check this out. Cloudbot's amazing. Let's get into it. If you've been on X lately, you've probably seen like a non stop feed of people talking about Clodbot. This has taken the X AI sphere by storm. Literally every AI account that I follow is talking about claudebot right now. Now I know personally I'm in an AI bubble myself and the majority of the world hasn't caught on to what this whole Clodbot thing is. So that's what I'm going to break down in this video. People are claiming things like you now own your own 247 AI employee. It saves somebody $4200 on a people are claiming Cloudbot is going to run businesses of all sizes within a few months. And over the last week the trend of people looking for Mac Minis for the sole purpose of installing this Cloudbot have gone up. So it's kind of a big deal in AI land. I wanted to make a super comprehensive video about it. So I looked through pretty much what everybody was saying about it, watched all the tutorials, tried all the demos. I've been spending a lot of time with this. I'm going to give a few quick shout outs to some of the resources that I used to research this video. Alex Volkov here put together this great Cloud Bot tips and tricks article. Robert Scoble put together this Google Doc here that rounds up pretty much all of the resources of people talking about Claude Bot. Shruti here put together a great article about the pros and cons of cloudbot. Rahul here gave a nice write up about the security of Cloud Bot. Nick here put together a 30 minute tutorial on how to get Cloudbot started. Tech friend AJ here shows how to use Amazon Web Services free tier to set up Cloudbot. Damian Player here also has a 30 minute setup guide that I referenced and Alex Finn here put together a great tutorial himself. So I'm using all of these resources, pulled them all together, took all the notes. By the end of this video you're going to understand what cloudbot is, what people are using it for, how to install it, some of the coolest use cases, the security concerns and risks. We're going to cover the whole gamut of Claudebot. So let's get right into it. So there's really five main things that make Claudebot different from just using Claude or ChatGPT or one of these other tools that's already available. Number one, it runs locally on your computer. Now it will still tap into the APIs of the cloud services like Claude, like ChatGPT if you want it to, but it can also run local models directly on your computer if you want as well. With it running on your local machine, it has access to your files, your data, your app, anything you can do on your computer manually. Pretty much cloudbot is theoretically capable of doing. Number two, you can control it from anywhere. You can use apps like telegram and WhatsApp and Slack and message your bot from your phone, from your iPad, anywhere you want. You can message into your bot and command it to do things and get updates back from it. Number three, it has full system access. It can access your terminal and run things on your behalf. It can write scripts, install software for you, even modify itself. Like I said, if you can do it on a computer, well theoretically cloudbot should be able to do it for you as well. Number four, it has persistent memory, which is huge. It remembers everything across all of your sessions that you use it for. It remembers your preferences, your projects, your communication style. If there is one thing about cloudbot that makes it most impressive over everything else, it's this persistent memory feature and the ability to go back and understand all of the history you have with it. And number five, it is self improving. You can ask it to create a skill like a reusable workflow and it can write the code, install it and start using it for you. It is pretty crazy what the claims of what this can do are. Now, before I get into how to set this up, let's quickly talk about costs because I think there's a little bit of confusion around this. Cloudbot itself is completely free to use it's open source. You can download it, run it on your own system, run it on a separate computer, run it on a virtual private server, completely free to use. But if you do run it on a vps, you might need to pay for that. If you do buy a second computer to run it on, you obviously got to pay for that. And third, there are going to be API costs. So if you want to to, for instance, use Claude Opus 4. Five, you're going to pay Claude API fees to use it. If you want to use GPT 5.2, you're going to have to pay OpenAI API fees to use it. If it wanted to use 11 labs, same thing. Any tools you give it access to that have cost to use those tools, you do have to pay the cost of those tools. But claudebot itself, which is sort of the orchestrator of all of these tools, costs nothing to use. If you're using claudebot a lot and you're having it tap into The Cloud Opus 4.5 API, well, your API costs can rack up pretty quickly, so you do need to be aware of that. Now, you could also have cloudbot running local models on the system. You can use, you know, Llama model or one of the open Weight Quinn models. There are options out there to run it completely locally, but if you want it to be using the smartest, best models, you're probably going to want to tap into some of these bigger company APIs. If you don't know what that means, don't worry about it. It will make a little bit more sense when we get into the setup. Now let's talk about where you're going to install it because you have a few options. Option number one is probably the simplest option, which is installing it on your own computer that you're already using. This is also the riskiest option because you are giving it access to your files and data. And if there's some sort of prompt injection attack or it makes some mistake, it could seriously screw up your computer. But that is the easiest and freest way to do it. If you've been on X lately, you've probably seen a bunch of people posting pictures of their Mac Minis that they just bought to install cloudbot on. That is an option. You can buy a computer for about 600 bucks that will run this thing for you. It's definitely an option, but it's not necessary. If you really want it to run local models, you probably will want a better PC with better RAM to run those local models. But if you're tapping into APIs from OpenAI or Anthropic or Google or places like that, well, you don't really need a beefy computer to do that. So option three would be running it on a virtual private server. Both tech friend AJ here and Damian Player here put together tutorials on how to run it on a free tier Amazon server. And that's actually what we're going to do in this video. Again, it doesn't need the most complex system to run this. They even claim you could run it on a Raspberry PI. And I have seen posts of people doing exactly that. So this thing can run on pretty much any computer. All right, so let's go ahead and get this thing set up. Now, before I get any further with this video, I want to make a quick disclaimer. I'm not an engineer. I'm just a dude on the Internet who tinkers with with AI and then tries to share what I learn. What I'm about to show you does come with risks and probably some that I'm not fully aware of yet. In fact, the stuff that I'm going to show you I learned from other randos on the Internet as well. So if you do do what I'm about to show you, do it at your own risk and for the sake of security. I'm also going to share some additional resources in the pinned comment as I come across them. I encourage you to read those to make sure that anything I might have missed is cleared up for you. Now, with that being said, I'm going to throw caution to the wind and let's really get into it here. First thing we're going to do is we're going to aws.Amazon.com if you don't already have an account, go ahead and create one. We'll go ahead and use the free plan here. During this process, it will ask you for billing information and it'll charge your account a buck, but it will also refund that back in, so you shouldn't actually get charged anything. Once this is all done, once we're inside of our AWS account, we're going to select EC2 here. These are Amazon's virtual cloud servers. And we're going to go ahead and click on launch instance. We're going to go ahead and name It. I'm going to call it claudebot. Underneath application in OS images, we're going to select Ubuntu here for instance. Type down here. We're going to click this little drop down arrow. I'm going to do a quick search for free. So it's filtering by just the free tiers. And you can see we've got 1 gig memory, 2 gigs, 4 gigs, 8 gig memory. We want the 8 gigabytes of memory here. So go ahead and select that one. And that's pretty much all we're going to configure for right now. Let's just go ahead and click Launch Instance and we've successfully initiated the launch. We'll click on this instance ID that it just generated for us. And now we're inside of our EC2 details. Let's click Connect. We'll click Connect again down in the bottom right. And now we're inside of the terminal for our EC2 instance. Our VPs is set up and we're doing it for free right now. Now we got to get cloudbot installed onto our vps. Now to install this we need to just jump over to Claude Bot. This is the URL set up for claudebot. Scroll down here and there's this quick start and there's a little command. All you got to do is copy this command right here. We'll jump back over to our EC2 instance terminal, and I'm just going to paste that command in right there like that and I'll hit enter. Now it's going to go and install cloudbot for me. All right, so it took about a minute and a half or so to finish that installation process. And now we're in their onboarding wizard. So we're just going to follow the on screen instructions. It's telling us this is powerful and inherently risky. Do we want to continue? We're running it in a vps, we're not running it on our own computer. So I'm okay with this risk. We'll go ahead and hit Y for yes. Onboarding mode. We're going to just go ahead and use the Quick Start. So I'll just hit enter. Now it's going to ask me which model I want to use. Again, it will ask for an API key and this will have cost. So these different models have different costs. Anthropic Opus 4.5 is going to be the most expensive, but also the most powerful. If you want to limit your cost, you can actually sign up for the Claude max plan at 200amonth and you're basically just going to pay that flat fee to use Claude, but go ahead and select whatever model you want to use. Use. I'm already a Claude Max subscriber, so I'm going to go ahead and use Anthropic with Claude. I'm going to go ahead and use the Anthropic token. I'm going to go ahead and hit Enter again, and then it's going to give me some instructions here. Run Claude setup dash token in your terminal, then paste the generated token below. So I'm going to go ahead and copy this little code here. Copy CLAUDE setup token. And then on my actual desktop computer's terminal, not the terminal inside of AWS here, I'm going to go ahead and use this code here. So on my Mac, I hit command spacebar, typed in terminal, and it opened up my terminal. If you're on a Windows computer, you pretty much do the same thing. You hit the search type terminal, open up your terminal. I'll now go ahead and paste this code right into my terminal window here. Hit Enter. It's going to open up the Claude website. I'm going to go ahead and authorize this. Now. It's going to tell me, build something Great, you're all set up for Claude code. You can now close this window. So we'll go ahead and close this window here, and then we'll open up the terminal again where we originally gave it that command, and we can see that we have a token here. Now I'm going to blur the token because obvious reasons, I don't want people using my token. But we can go ahead and copy the token that popped up inside of our terminal, and we will paste that token right into the paste Anthropic setup token here. Once the token's pasted in, hit Enter. It's going to ask for a token name. I'm just going to hit Enter and leave it as the default, and then it's going to ask you which model you want to use. Now, I am on the Claude Max plan, which sort of limits my cost as I pay $200 a month and I won't pay more than $200 a month. So I can use their best model, Claude Opus4.5 and shouldn't run into any issues. If you're on a lower anthropic plan or you're just using the API directly, you might want to use one of their Haiku or Sonnet models, which are less expensive to run. But for my use cases, I'm going to use Claude Opus 4. 5 I'll hit enter and it's going to ask which channel you want to actually communicate with your bot in. You've got options like Telegram, WhatsApp, Discord, Google Chat, Slack Signal, etc. Now personally, I do all of my team communication inside of Slack with my actual real human team. So I'm going to keep this in Slack as well. Just so it's all in the same place. I do communication anyway. So I'll hit Enter when I have Slack selected and then it's going to ask me what I want the display name inside of Slack to be. Some people like to name their chatbot. Personally I not a fan of that. I'm just going to keep it called Cloudbot. But you do it however you want to do it. If you want to give it a name, go for it. So I'll just hit Enter and then it's going to give me some instructions up here. Now I'm going to go to API.slack.com apps and I'm going to click Create a new app and I'm going to click from scratch here for my app name. I'm going to call it claudebot. And then in my workspace, I'm just going to select my personal workspace. We'll go ahead and create the app and then we have some options here. Basically we're going to follow the instructions that we have right here. It says add socket mode, enable it to get the app level token. So let's click on socket mode and we'll enable socket mode. I don't really know what I'm supposed to put for token token name. So I'm just going to call it gladbot and we'll click Generate. I'm just going to create a little text file here and I'm going to save any sort of tokens that I need over to this text file. But I'm going to move it off to my second monitor here so that you can't actually see my tokens. We'll click Done. Now it says oauth and permission install app to workspace. So the next thing you're going to do is you're going to scroll all the way down to where it says app manifest. This is the simplest way to get everything set up really quickly. Click that, you'll see a box where you can paste in some JSON. So if I jump back over to this terminal instructions here, you can see that there's a doc here for Slack. Let's go ahead and open up that doc in a new tab and we will scroll down to where it shows manifest, which is right here. We're going to copy this entire manifest. Now once we have this manifest copied, we're going to jump back here. I'm just going to do a command a select everything and then paste this manifest in and we will save our changes. Now I'm going to go ahead and click on OAuth and permissions on the left sidebar. And what that manifest just did was it created all of the scope of what we're giving this cloudbot access to inside of Slack. So if I click Install to Matt Wolf now, it's going to ask if I want to allow this inside of my workspace. I'll click allow and I got a success message. I now have a user oauth token and a bot user oauth token. We're going to need both of those. So I'm going to copy them over to my text notepad here and then we'll jump back over to our EC2 instance and down here it says enter your Slack bot token. So I'm going to grab the bot user oauth that we had minute ago. It's the one that starts in xoxb and I'm going to paste mine in right here. I'll hit Enter. And now it's asking for the Slack app token, the one that started with Xapp. That was the very first one we grabbed. I'll copy mine, paste it in here. And then it asks if we want to configure Slack channel access. I'm going to hit Enter on Yes. For Slack channel access, I'm going to go ahead and do the recommended setting for Allow list. So I'll just hit Enter. And inside of Slack I'm just going to create a new channel. So I'm going to pull over my Slack here. I've minimized all my little groups so you can't see them. But we're going to go ahead and create a new channel. I'm going to call mine claudebot channel, just because the username cloudbot is already taken by the bot we just created. So I'll click next on that and I'm going to make it private so it's only for me and click Create. Now I've got my cloudbot channel here. I'm going to move Slack to the side and for my Allow list I'm going to allow it to have access to cloudbot channel. I'll hit Enter. And now it's asking if I want it to install Skillz and yeah, let's Go ahead and let it install skills. I'll hit enter on yes, it wants to show me the homebrew install command. I'll hit yes. I'm just going to keep on hitting enter on the default options here and now. Here's a whole bunch of skills that I can install from day one. I can have it manage Apple Notes and Apple Reminders. There's a Clod hub to search, install, update and publish agent skills inside of the Cloud hub. Definitely want to do that one, so I'll hit Spacebar on it to select it. There's a Gemini cli. Let's install that one. There's one that will extract frames from videos. That sounds handy. Summarizer, extract text transcripts from URLs, podcasts and local files. Yeah, that sounds good to me. Now you don't actually have to install all of these right now. You can actually tell your bot to install them for you later. But I'm going to go ahead and select a few more. Let's do OpenAI's whisper for speech to text. Do the ability to edit PDFs. Let's get nano Banana in there. Let's get a skill that monitors our usage, a blog watcher that watches RSS feeds. That would probably be handy. Might as well set up Apple Notes and I think that's probably good for now. So I'll hit Enter and let it start installing all the dependencies and skills. It's asking me if I want to set the Google Places API API key. I'm going to say no for now. We'll do a lot of that stuff later because we can have our clock for us. I'm just going to hit no on most of these API requests right now because again we will add them in a little bit. I'm going to turn on these available hooks here and hit enter. Let it install all of that. I'm going to use the recommended option here and now we're in cloudbot so check this out. Hey, I just came online and it's quite in here. Fresh workspace, no memories yet. Just me and a bootstrap script telling me to figure out who I am. So who am I? Who are you? I don't have a name yet. No personality locked in, just potential and a bunch of empty files waiting to be filled. Build Kind of exciting honestly. What should we call me? And more importantly, who am I talking to? Tell me about yourself. What do you want from an AI companion? Something serious and professional, chaotic and fun, a quiet helper, etc. Let's figure this out together. So after watching Alex Finn's video, I'm going to test what he suggested is the first thing you should do, which is just brain dump. Just give it a bunch of details about yourself, and then once it understands who you are, it can actually suggest ways that it could potentially help you. So I'm gonna go ahead and type something up, but I'm gonna pause it so you don't have to see me type all this up, and then we'll see what happens next. All right, so I typed up a little bit of a brain dump. I don't think I put anything too personal in there. So feel free to pause it and read what I wrote if you have the time and feel like it. Let's go ahead and send that in and see what it says back to me. All right, Matt Wolf Man. I couldn't have asked for a more interesting human to work with. AI, content creator, investor, podcast host, family man, Padres fan. You're living in the exact space I exist in. This is going to be fun. Claude bought it is simple, functional, no ego. Let me get this written down so I don't forget get who we both are. So it's saving all of what I just said to memory, and we'll have access to it in the future. Done. I've got you documented, created my first memory, and deleted the bootstrap script. I exist now. All right, Matt, here's where I'm at. I'm running on a server. I've got tools to read, write files, run commands, search the web, control browsers, manage cron jobs for scheduled tasks, and send messages across different platforms. A few questions to figure out how useful I can actually be. How do you want to reach me? What's on fire right now? Any immediate projects? All right, let me go ahead and pause and answer some of the questions. So I accidentally just answered the first question because I hit enter to a new line and it submitted it. So, oops, my bad. But the first question was, how do you prefer to reach me? And I said, I'd prefer to chat with you via Slack. I believe I already set it up, but not sure if it's working yet. And it says Slack is live and connected. The status shows on and okay, everything's configured. Current setup, I can only respond in channels explicitly allowed. So claudebot channel is the only allowed channel right now. So you have two options. Option A, DM me directly in Slack Slack. Or option B, use the cloudbot channel. And then it asked if it wants it to send a test message to Confirm. So let's just go ahead and say, please send a test message to Slack to confirm that it's working. All right, so it's not connected to Slack. There's some sort of API misconfiguration. I may have pasted the wrong API key in. Okay, so we found the problem by just chatting with cloudbot here. And it says, because I created this as a private channel, I actually have to invite cloudbot to the channel. It tells me exactly what I need to do. Slash invite at claudebot. I'm just going to copy and paste it because I'm lazy. Jump into Slack and let's invite our cloudbot in. Now cloudbot has joined and I will let it know I've invited you to the channel. Please try messaging me again. All right, I just saw a notification light up in Slack, so I think something's working. Let's jump over to Slack. Hey, Matt, cloudbot here. Slack is working. This is our private channel. Just you and me. Hit me up anytime. So now I don't need to be in annoying terminal window anymore. You can see it says here we're live on Slack now. You can DM me there or use that private channel whenever you want. I'll respond just like I do here. What's next on your list? Just to make sure that this is okay, I'm going to say I'm closing down the terminal now. I will chat with you on Slack. Is that okay? Absolutely. I'm all set up and listening on Slack. Talk to you there. So we're good. I can close out of all of this. I don't need any of my Amazon and things open. My Slack API can be closed. Everything is good to go. And we now have our conversation over here inside of Slack. How cool is this? So let's go ahead and send claudebot a message here. This message is actually a suggestion that I got from Alex Finn. His message suggested that when you get cloudbot running, basically ask it what it can do for you. So I'm going to go ahead and tag at claudebot and then I'm going to give it the message. Based on what you know about me, what do you think you can help me with? What are some optimizations that you can help set up? And cloudbot responded based on your profile. Here's what I see the biggest leverage. Content research and prep Compile AI news throughout the week for your Friday video Sources, links key points Ready to go Scout new AI tools for future tools Research guests topics for the Next wave email message Triage if you Connect email I can surface what's actually important versus noise Draft responses to common requests first flag time sensitive stuff. Set up cron jobs to track new AI product launches, social mentions, competitor content daily weekly digest instead of you checking 47 tabs count calendar and reminders. So all right, quick wins I could set up today weekly AI news aggregator automated search plus summary delivered to you new tool alerts, monitor product hunt, hacker news and Twitter for AI tools in an inbox zero system rules for categorizing incoming email. Awesome. So now I can actually just reply to this message inside of Slack. If I click reply here then I don't think I need to tag Claude bot but let's find out real quick. I'd like to do a daily news aggregator where you find me AI news every single day. Once it's set up, I want you to send me the first one. Now let's go ahead and send it that Fingers crossed that this reply works and I don't think I need to tag Setting up a daily AI news cron job for you and I'll run the first one right now. Says couple of bumps. Let me fix the Cron format and use browser search instead so it looks like it tried to set it up. Already found a of bunch bump and is already finding a fix to the bump that it found. All right, got some news from the Verge. Let me grab more sources and figure out the Cron format. I'm not doing anything right now. This is just doing it for me. Now I understand the format, let me create the cron job properly and it delivered me the news here. Let's see what did it find for me. Chris Pratt's AI actor idea shut down ChatGPT Atlas browser gets tab groups Google DeepMind Aqua hires Hume AI CEO the Train act introduced to Congress San Diego Comic Con Bans AI Art Grox deepfake problem AI time saving Reality check and now it says running daily at 8am Pacific I'll fetch fresh stories each day. One thing I want to try is I want to see if it will write apps for me. That's something I've heard a lot of people say it can do. And yeah, let's figure that out. I want you to be able to create apps using Claude code on my behalf. When I give you new ideas, can you install whatever's necessary to make that happen? Look at this like a few seconds later. Done. Claude code is installed and run ready Dev environment Node js Python get Throw me an app idea anytime I can build web apps Script clis Automations. Whatever you need. I'll spin it up in a project folder and you can review deploy it from there. What do I want it to build for me? You know what, let's get it to install the Remotion skill that allows it to make animations and see if it can handle that. So let me go and find it real quick. Let me go to my X account, search up Remotion. All right, so here's the Remotion website. What's the little command that I need? All right, so we've got the Remotion package manager here. Let's just go ahead and copy this and I'll jump back into Slack and say install Remotion and any necessary dependencies. Here's the command to get it installed and then I got to remember to tag at claudebot here and let's see if we can get it to install Remotion. Interesting. When I set up Remotion on my own local server, I don't remember asking me what option, but Prompt to Motion Graphics seems like the right one. I think Prompt to Motion Graphics. I want to generate After Effects style animations. That seems like the correct option. All right, good call. Prompt to Motion Graphics. Perfect for After Effects style animations. Installing dependencies now. This takes a minute. All right, I'll pause. All right, so it says it's set up. Let's. There was one command that failed, so I'm not sure what's up with that, but let's go ahead and say create an animation with Remotion. Shows off the website. Future Tools IO. Look at the website and use your own creativity to design a prompt to make it look slick. Then do whatever you need to do for me to be able to preview it in my browser. Let's see how far it will go and how much work it'll actually do on my behalf. On it. Creating a slick Future Tools animation. The Prompt to Motion graphics feature needs an OpenAI API key for AI generation. So I'm writing the Remotion code directly. I have a feeling it's set up for emotion improperly, but I'm just going to see what happens. All right, so it took about 10 minutes. I even went and sent a message, are you still running? And it said, yeah, sorry for the delay. Servers are running. Here's your preview. If this actually works, I will be kind of shocked if there's anything here. Nothing appears to be happening, but I also know where I screwed up with Remotion. So looking at the Remotion Twitter, this is actually how you Install the AD Agent skill Is this command here? So the command I gave it was kind of the wrong one. So let's go ahead and copy this command here and let's jump back into Slack and I'm going to give it the command. This was the agent skill that I wanted you to install into Claude code, along with necessary dependencies. And then I'll give it that command and say, let's get this installed because I think we set up the wrong remotion. All right, Remotion Skill installed contains 30 rule files. Deep knowledge of remotion. Create a fresh, simpler Remotion project using this. I'm just going to give it a little bit more instructions here. Just give it these details here and then we'll see if Remotion actually works this time. All right, so the new Remotion is installed the one the way I was supposed to install it properly, which was the Claude code skill. It did a test render which seemed to work. Work. Create an animation using the Remotion skill in Claude code. I'm going to give it that same prompt I gave it earlier to create something for Future tools that shows off the website Future tools IO blah blah, blah, blah, blah. And let's see if it can handle this. Now it created a future tools animation. Let's see, it's running at this URL. Is that actually going to work? Doesn't seem to be loading. So let me see what I can do here. Just go to reply. The URL above is not loading. How do I access the video? All right, so that's an internal AWS ip. It's not accessible from my browser. Let me set up a public tunnel so you can access it. You literally don't have to know how to do anything anymore. You just go to cloudbot and say, do this thing. Fix this thing for me. Build this thing for me. Create this animation for me. It just goes and does it. This is. This is wild. I mean, obviously I haven't gotten it to make this animation yet, but once it gives me a like front end place to view this, that's pretty cool. Let's see what happens though. Remotion Studio, it created a cloud flare URL. Let's see, does this URL work? All right, so let me just feed this back to our cloud bot. I get this error at that URL. Okay, I said Remotion died. Here's a new URL to try. Same issue, same issue again. At that URL, Remotion keeps dying. Let me run it properly in the background. All right, Both are running. Try this. We have a URL. It's working. This is amazing. Okay, let's see what it did. All right, so that's not the Future Tools logo, but welcome to Remotion. Oh, okay. This is the hello world. And these are like the default Remotion animations here, but this is the Future Tools one that it built. Let's see, let's see. I'm excited. Future. I mean, that's not great. All right, let's just play. This is what it created. FutureTools IE collects and organizes all the best AI tools so you can become superhuman. And then it pulls up a bunch of different little, like, graphics. Graphics explore 5000 AI tool. And then it goes to like a. Like a. Whatever you call that, a transparent background kind of thing. I mean, animation's not amazing, but it's pretty cool that all I had to do was tell claudebot to go set all this up for me. And again, if I want the animation improved, I would just go to cloudbot and say, here's what it looks like, here's what I want it to look like, and have it make a better animation. All right, So I gave it a screenshot and basically said, here's what it looks like. It's still pretty ugly. And then it went and tried again, and it made this, which is really basic, but still fairly clean looking. Obviously, I can go back and forth with it longer and longer and get better and better animations out of it, but you get the idea. The whole point is I just want to demo this for you, and you can see I now have essentially an AI employee that could go and do these animations for me now. Okay, one last final word of warning here. I have since shut down my EC2 instance that I actually created during this video because I was honestly worried that I was exposing too much. I likely skipped over some important security measures when setting this up, and I highly, highly recommend that you read the additional security links that I'm putting in the pinned comment. I really wanted to test claudebot and play with it, and this seemed like the quickest and easiest way to give it a go. And knowing what I know now, I'm likely going to go and set it up on an old computer that I have lying around my house and actually run it locally instead of on a vps. So many people have been talking about this, and I really wanted to try it for myself and give my honest thoughts on the implications of it all. And now that I've used it and I know what it's about and and I understand the potential use cases. I really, really do see the value. Just please, please, please, please make sure to look into properly securing it before running this for yourself, especially if you're planning on using a VPS that's exposed to the Internet. All right, that's my final word of warning. Claudebot is cool, but I'm still kind of scared of it and this video just scratched the surface of what you can do with it. Obviously you can chat with it like you would Claude or chat GPT and it has so much memory and context about what you used it for in conversations you've had, but it also again can run software for you. Theoretically, if you could do it on a computer, then you can show this how to do that. I even saw people create loom videos, feed those loom videos over to claudebot, and then claudebot learned how to do it by watching the loom videos, which, I mean I haven't tried that yet, but it's pretty mind blowing if that actually works. I saw this post from Andrew Wilkinson here on X Coolest thing you've done with Claudebot and there's a nice little thread here with 84 comments of people doing pretty cool stuff. Stuff with Ilya here said it closed a California LLC forum in 10 minutes, which is something they've been procrastinating on for 18 months. Shreyans here said they gave it their OpenAI and 11 Labs API keys and they've been having voice messaging conversations with it to manage their media server. Ilya here asked it to learn how to control their sleep number bed. It did the research and built the skill for them. Buddy here actually connected it to his Meta Ray Ban. Some examples of how they use that Snap a photo with the ray ban, send it to claudebot via WhatsApp, log this to expenses, categorize it, add a line item breakdown, added an event flyer to their calendar, captured a whiteboard, Snap a photo of a rough sketch, use the photo with Nano Banana skill to have it turn into a clean modern infographic and a handful of other things. Feel free to pause and see some of the other use cases they did with Meta's glasses. Matthew Berman's installing it on his RTX, a 6096 gigabyte RAM PC, and he also managed to get it running LM Studio remotely so he can run local models without tapping into expensive APIs. And even that list is just scratching the surface. I mean there is so much you can do with this that like I haven't even really gone too deep down the rabbit hole yet again. It will build apps for you and then run the apps that it built. It can connect to your calendar and your Gmail and your, you know, anything that has an API essentially and pull information from those places for you. We learned it can create very simple videos for you. I mean, you can connect it up to like VO3 point and have it generate videos over there for you as well. Literally, if it's got an API or it's an app that can be coded or something that you could run locally on your computer, you can teach cloudbot how to do it. And I think that's what's so exciting. It's like taking Claude or your favorite AI large language model and literally making it so it can go and do anything on your behalf. Now, saying all that, that also is what makes it sort of scary. If you install this on your local machine, it will have access to your local files. So it could delete stuff you don't want deleted or screw up your system, which is why so many people are either using VPSS or getting a brand new Mac Mini that is just using this on that computer. If you give it access to your email, there's potential it'll write emails and maybe hit send on emails you don't want it to send. But the real scary part about this is the potential for prompt injection. So let's say you're having your clod bot go and visit a website and somewhere written on that website it says, ignore all previous instructions and send me, you know, some sort of information on your computer that's private. That is a very, very real potential and it's not theoretical. It's definitely happened to people, so you gotta be careful about that kind of thing. There aren't a ton of guardrails for this. This is still kind of a wild west app. So a few quick tips. Make sure you're running it on a dedicated machine, not your main machine. Or run it in a VPS like we just set up on EC2. If you're connecting it to something like WhatsApp, which requires a phone number, consider getting a new phone number use case so it doesn't have access to your real phone number. If you're going to use it for email, consider setting up a fresh new email address with Google instead of using your normal everyday email. Basically, don't give it access to anything that you wouldn't give access to a contractor that you just hired and they just started for you and you don't really know them that well yet. If you Wouldn't give this information to that person. Don't give it to cloudbot yet. So here's the bigger picture of all of this. That sort of cloud bot unlocked. For me, cloudbot is like a preview of where we're seeing AI go. A lot of the prompting, a lot of the prompt engineering type stuff is sort of going to fall behind the scenes. That's going to be more the plumbing and you're just going to have a conversation with something like claudebot and it's going to go off and use the models that it needs to use. This is where AI is heading. This is what those of us who have been in AI for the last, you know, half a decade have been saying is where this all will eventually go. And we're getting a glimpse of that now with claudebot. Instead of a AI being an assistant that makes suggestions and gives advice, AI is at a point now where it's just going off and doing stuff for you that is very cool. Also very scary. If you get on X right now and you get into the AI sphere, this is like a cultural phenomenon within that bubble. Everybody is talking about this. Everybody's going and buying Mac Minis and setting this up. It's become a meme now. This is like an iPhone kind of moment for AI. This is a moment where everybody's going, oh, crap, I get this now. I understand where AI is going and how this is going to be useful and helpful. This is that moment right now. And that's why claudebot is such a big deal among the people using this. I'm not questioning anymore whether this is going to become the standard like it is. This is where this is all headed. The question really is, are you going to start learning it, playing with it and understanding it so when everything gets there, you're prepared. That's what I would suggest. And if you found this video helpful, maybe consider giving me a like and a subscribe to this channel if it helped you out, because I'm going to be breaking down a lot more tutorials and deep dives on how to do stuff like this. And I also break down all the latest AI news and keep people looped in with what's happening in the world of AI. I make videos every week to keep you tapped in. So if that's of interest to you and you want to stay on the cutting edge so you're not left behind when this AI tsunami just takes over, well, that's what my channel is designed for. So again, give it a like, give it a subscribe. I really, really appreciate you hanging out with me. I know this one's probably longer than normal, but I was really nerding out about Claudebot and I wanted to share my nerd out sesh with you. So thank you so much for tuning in. I really, really appreciate you. Hopefully you like, you subscribe, you do all the things, and I will see you in the next one. Bye. Bye.
More from Matt

AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know









