Clawdbot: POWERFUL AI Assistant on your Phone

Over the last few days, I've created an AI assistant that lives on my phone that can manage my calendar, summarize emails, maintain shopping lists, and more. But don't just take my word for it. Let me show you. Hey, Curly. What's currently in my shopping list?

Great. Can you now add some eggs, some toilet paper, and also maybe some zucchinis?

Let's do another one. Hey, Curly, can you put an event on my calendar for Sunday at 12:00? I'm going on a coffee date with my friend.

>> Yay. And this technology was made possible due to a new exciting open source software called Claudebot. And that's spelled C L A WD because the creator Peter Steinberger seems to really like lobsters. Anyway, in this video I'll be going over what Cloudbot is, how you can set one up for yourself, some useful things to know to make Cloudbot even better, and some caveat from my personal experience of using this Cloudbot. Now before I dive in, I am expecting some degree of

technical knowledge is because setting this up will require either a virtual machine or a remote computer. But without further ado, let's get started. Now, what exactly is Cloudbot? Cloudbot is an open- source software that allows you to build personal assistance using AI. Now, the way that it does this is that it connects your local computer with your clients like your WhatsApp chat or Telegram or Slack with an AI model. Now, the best way to show you what Cloudbot is capable of is to go

through its features. So, let's go through that right now. So, first and foremost, Cloudbot runs on your machine. And that's important because eventually you're going to want to configure your Cloudbot with various API keys and integrations. and you don't want that information living on somebody else's computer. Secondly, Cloudbot is in the way that I view it the glue between your clients which are your chat apps and your AI that you want to use. So when it

comes to your clients, you have multiple options. You have WhatsApp, you have Telegram, you have Discord, Slack, Signal and more. And these are very good because you can talk to them in one platform like WhatsApp for example. and then you can resume your conversation on Slack and it will be completely seamless. The third feature is persistent memory. And persistent memory is really important because you want your personal assistant to remember certain things about you. So that could

be your name, uh your preferences, uh your location, your style of communicating. You can even configure your chatbot to have a different name and persona as well. And these are all things that gets stored into your computer for future reference. So, in my case, I've actually renamed my bot to be called Curly from Good Omens, and it has a persona of a devil who is sarcastic, witty, and it's just a really fun vibe to to work with an personal assistant that I like to work with. Now, the

fourth feature is browser control. And browser control can be a little bit tricky just because navigating websites or any bot whatsoever is quite tricky. There's a lot of websites that have anti-bot guards in place and that makes it hard for bots to navigate websites just because it can't do certain things that a normal user could. So I would say in my personal experience the browser control is a little bit hit and miss but perhaps some other people have figured

out ways around it. The fifth feature is full system access and from what I understand this is optional because you can install cloudbot in a docker container too but the way that I have set cloudbot up is through a virtual machine. So it has full system access and it can modify its own file as well which is really cool. The way that Cloudbot explains it is that it's kind of like doing surgery on yourself. So it can change its own configuration file and restart its own system. So it's very

autonomous in that way and that's what makes it very powerful because it's now able to modify some of its own file, give it self skills and give itself some uh integrations and that kind of stuff. Of course at the end of the day you still have full control. you. It only does the things that you ask it to do. And in the case that something goes wrong, you're always able to either back up and restore or you can just kill it all together. And finally, we have skills and plugins, which in my opinion

is really what makes Cloudbot way more useful. And you can give your Cloudbot skills like the ability to transcribe audio. So now instead of having to type to it, you can just talk to it using your voice directly. Or you can use things like CLIs like um Google Maps. So you can search up nearby restaurants or you can even give it access to your Gmail accounts, your Google calendar, so it's able to give you summaries of your emails from yesterday or look up your

your schedule and make your schedule for you. So these are all really powerful things that you can do once you augment your AI with certain functionalities. Okay, now's the time to install Cloud Bond into our machines. The way that I'm going to do it is through Proxmox VM and I'm going to set up an Ubuntu server. Now, you can do this on your cloud computer as well. Um, but for the sake of this video, to keep it short, I'll be speeding through the setup of the Ubuntu server.

So, we've completed provisioning our virtual machine. It's been set to Ubuntu server with one core and 4 GB of RAM. The next step for us is to install the prerequisites for Cloudbot and they're fairly short. You just need to have Node version 22 plus. So, let's let's install Node version 22. And the easiest way is to use NVM. And we're going to install 22.

Great. And check that our node is installed correctly. Good. Next up, we'll want to install PNPM. And the easiest way is to run core pack uh enabled. And if you just run any PMP command, it should install it. version. Good. Now, to install Cloudbot, they do provide an installation script, but in our previous attempt at running this installation, it actually ran into an error and and so we're going to stick with the simple installation version, which is the global installation. We're

going to install with PMPM.

Oops. And we're going to need to run setup. Okay, let's try that again. Okay, at this point we have Cloudbot installed and all we need to do is run Cloudbot on board and it will walk us through the steps. Going to go with the quick start mode and for our O provider we're going with anthropics cloud API. Now you can use your OOTH for your cloud code subscription as well, but recently there's there's been some development from Anthropic and they don't really want you to be using your cloud code

subscription for third party applications like this. So in the meanwhile, we're just going to stick with an anthropic API key. We're going to use a default model here and we're going to set up Telegram. Now they do support WhatsApp, Discord, Slack, and whatnot. Um, but for our demonstration, we'll be using Telegram. And the onboarding process is nice. It provides you with the instructions on how to set up your Telegram bot. So, let's open up our Telegram. I have here

bot father opened up. And the command to run here is new bot. And let's give it a name. We're going to call it Jarvis. And we need to provide a username for our bot. Now, this has to be unique. And so we're going to go with Jarvis Overseed AI bot. And so it's created for us an access token. And what we need to do is grab this token and paste it back into our application. Now this Telegram allow from section here is going to be your unique identifier. But we currently don't have

it right now. So we're just going to put in some placeholder numbers. And we're going to skip the skills for now. The skills are really handy for allowing your your your bot to have additional abilities. And I can show you what this looks like. And they have a lot of different skills that you can install, but they do take a lot of time. And I found that sometimes the installation doesn't quite work exactly. So I'm just going to skip them for now. Skip the rest of the application

configurations for for skills.

Okay, now the onboarding is completed. We just want to check if Cloudbot has been installed to our Damon. Now the Damon is important because you do want this bot to be running 24/7. So to check the Damon, we're going to run clotbot Damon status.

And perfect, it is currently running and it should be responding to our messages now. So, we're going to go back to our Telegram. We're going to click on the link that Botfather has created for our Jarvis bot. And that should open up the chat to Jarvis. Click start. And Jarvis should start responding to messages now.

So, turns out there was an issue with the onboarding wizard where it wouldn't let me set a policy for pairing. So it defaulted to using the allowed user ids only. But in our case, what we wanted is the pairing mode. So this is a good example for me to show you how to edit the configuration file for your Cloudbot. And to do that, we're going to SSH into our computer again. And we're going to CD into the Cloudbot directory. And inside of this directory, you're going to find that there is a

JSON configuration file. So we're going to edit this file. And I've already recorded this and unfortunately that recording was corrupt. So, so this already has the solution in here. But let me just show you where you need to change this. So under channels, you're going to want to change the DM policy to pairing. And once you do that, make sure you save and exit. You're going to want to run bot damon restart. And this command will restart the server for you. and apply

the configuration. Now once that is completed, I was able to go back to the Jarvis bot and I was able to ask it a question and that prompted a pairing code. Now this pairing code then tells you the the command you need to input to to authenticate with your system. So copy that, paste that into your terminal and then grab your pairing code and append that at the very end of your statement and enter that in. Now, I already I've already paired the application in, so it

it's going to error out, but afterwards, I was able to text Jarvis just fine, and it it was able to respond back to me. Now that we have Jarvis up and running, the next step is to make Jarvis more useful. And the way that we do this is through context. Now, of course, you can use Jarvis as is. And right now, it's currently running Cloud Opus 4.5, so it's going to be very useful just talking to it in general. But if you really want to get the extra go the extra mile and make it a lot more

useful, you want to give it more context. And there's a few ways you can do this. The first thing you you want to do is set up your profile. So right off the bat, Jarvis is going to ask you who am I, what am I, and all all of these profile based questions. And you can give it a unique name, you can give it a unique persona. And there's a few ways you can do this. First, you can just talk to Jarvis directly. Um, like I mentioned earlier, it is able to do surgery on itself. So, it's able to

update its own code as you go along and it's it can even restart its own server. It's really handy for that. Or to give you a little demonstration, if we go back to our terminal here, we can actually um we can cd into the claw directory and this directory contains the this the configuration markdown files for your for your bot. So if we look at soul.md, this talks about how the persona of your bot is configured. Uh if you look at the the tools.md can talk about the tools that it has and

as your bot evolves and talks to you and it gains more abilities, it's going to update these files and it's going to become more and more useful and more of a humanlike bot. The second thing you can do is give it explicit skills. And there's a few ways you can do this as well. You can first rerun um cla bot onboard. And if you update the onboard, you can go back to that uh skills section and you can check off the skills that you want to install to your application. Or the other way that you

can do it, which is really handy, is the Cloudbot hub. Now, this hub has a CLI tool that allows you to install tools really easily. And the one that we used earlier was notion. So, let's take a look at notion again. And you'll see the notion show up here. And it gives you a little explanation of how it works. But basically, it's an it's a markdown file that cla uh the cloudbot reads and so it understands what it needs to do to make the skill work. And the easiest way to install

this is to go back to that cloud hub page, copy the command, and inside of your cloud directory. Now, this is really important. Uh there's two directories here. We have the home cloud directory, which has your your personalization files, but then there's also the system files as well. And that's where we had a look at the cloud. JSON file. So inside of this dot clot do cloudbot uh directory, these are your shared configuration files across all your bots if you go that way. Um and if

you put your skills inside of here, it's going to be shared across all your bots. Uh but in our case, we don't want to do that. We just want to go to our custom bots directory, which is the /cloud. And inside of here, we want to run mpxcloud install. And in our case, if we go back, um what was it? We just want to take a look at what what the slug of our skill was. And in our case, it's just notion. So we can type notion in here. And we're going to install claude hub. Claude hub.

And now it's installed. Now if we do ls on here, we're going to see that there is a skills directory. We go inside of there. And there's a notion in there. Sorry, cd notion. And let's just uh print the uh the file here. And it's going to provide your bots with instructions on how to integrate with the notion API. Now, with notion, you're going to have to create that um notion integration bot on their end as well, and that's going to be beyond the scope of this tutorial. But this is exactly

how you would create that shopping list item. Um that's how I did it. And the reason why I use notion is because I wanted to be able to see the shopping list on my end as well along with sharing it with other people that I might have. But you can always use your thought bots internal memory to keep uh shopping list or any sort of list whatsoever. It's just that in our in my case I wanted that list to be collaborative. So that's just notion. The other tools that you're going to

find really interesting is most likely Gmail and Google calendar. Now, for those guys, uh there is that built-in onboarding step where you can integrate with it. And they do require a little bit more um hands-on work just setting up the API on the on the on the Google side of things. But once you do get that set up, it's really really easy and it's really handy. Uh but if you would like to see more uh hands-on approach on how to get that set up, please let me know

in the comments down below and I can create a video for that as well. Okay, so now that we have Clawoth Hub set up on our phones, on our Telegram, I do want to go over a final section here and those are the caveats. Um, of course, it's a brand new piece of software, so it's going to have some issues. And this is to be expected. They've, I think, published the software only a few months ago, and since then, there's been some rapid development in the repository to

make this even better for you. But of course, I personally ran into a lot of issues getting this set up, especially the first time. And during the course of this recording, there's been a lot of pushes as well. So that's been constantly been changing. And that in part has made recording this video a little bit difficult. But on a as a whole, that's a really good sign because it's a very active repository and there's a lot of communication on the Discord server as well. So if you need

some help, um the Discord server is very good. But but yeah, it's an early stage software, so it's going to have bugs. It's going to have some issues and the documentation is constantly changing. Uh sometimes it's a little bit outdated. And I found the documentation jumps around here and there. And so it was a little bit hard to get my head around the first time, but once you have it going, it's actually a really good system. And I've been using my personal Crowley uh Cloudbot, and it's been a

very useful application to have in my pocket. Well, the other thing to consider with Cloudbot is your usage. So currently I think a lot of people are using the OOTH subscription for cloud code but we don't know what the future is going to hold for that. So if you do go the route of using API tokens the usage is quite high when it comes to cloud bots. So you are going to be racking up the cost quite a bit if you are using the more pricier models like Opus uh 4-5. So there are alternative

models that you can use. You can even bring your own model if you're able to deploy something locally. Or as an alternative, you can always opt for the cheaper anthropic models or I've been hearing news about the new miniax models that are coming out or that have come out and those are significantly cheaper than the Opus model or the anthropic models. So those are some ways that you can kind of bypass these cost issues. But I just wanted to let you guys know that there are considerations for for

cost if you are using this as an API tokenbased system. And the other thing is that because this bot is able to work on its own, it has its own heartbeats. It is going to acrue a lot more cost in the long haul if you have your bot running more uh more frequently 24/7. So that's another thing to keep in mind. And finally, and this is probably the most important, um your bot is going to make mistakes. Uh don't think that it's going to be this flawless AI assistant

bot. Um it has made mistakes on my on the things I've asked it to do. Generally, it's pretty good. But I what I suggest you to do is don't trust the bot to be perfect from the get-go. One of the areas that I've personally found that the bots's been struggling with was time zones. So, when I first asked it to do certain things on a on a routine or to set a reminder for something, it got the time zone wrong. But after a few times, I realized what was happening and

bot. Um it has made mistakes on my on the things I've asked it to do. Generally, it's pretty good. But I what I suggest you to do is don't trust the bot to be perfect from the get-go. One of the areas that I've personally found that the bots's been struggling with was time zones. So, when I first asked it to do certain things on a on a routine or to set a reminder for something, it got the time zone wrong. But after a few times, I realized what was happening and

I told it to keep a note about the time zone differences and it's been flawless after that. Again, just don't give it the full trust at the start. do monitor what it's doing and only after a little while, after a couple of uh back and forth, you're going to find that the accuracy is going to improve a lot more. And that's it on Cloudbot. Um, I hope you guys found this video useful. Um, I personally found Cloudbot an amazing piece of software and I've been using my

Crowley assistant pretty much all day, every day for the last seven days, and it's been amazing um, seeing what it can do and what it can come up with. Recently, I've been playing around with using Crowley as my research bot. So, it's going out and doing some research and coming back to me. It's been really good for that. So, yeah, do let me know down in the comments below if uh if there's anything I missed, if there's anything you wanted me to talk about more or if you like content like this

where I talk about some of the new technologies that are coming out. Uh please feel free to like and subscribe. It does help out the channel quite a bit. And yeah, I'll catch you in the next one. Bye-bye.