Transcription
to work 2 hours autonomously on your code, to independently call a restaurant and book a table, to constantly improve yourself and add new skills, and to remember everything you discussed with it. Online, it has already been nicknamed AGI, and users compare Klodbot to Jarvis from Iron Man. Well, you remember. >> People, robots, hello. An incredible wave of hype around Klodbot has risen in recent days. I didn't rush to make a video, I decided to wait, see where it all leads, and give you my honest opinion. In this video, I will explain how to install this agent. I will briefly talk about its capabilities and show you on my own example what I managed to do with it. Of course, I will talk about security, because if something is done wrong, the entire internet will get access to all your data, secret keys, dialogues, WhatsApp groups, emails, Telegram. There are already a bunch of hacked Klodbots online and an incredible amount of leaked data. We don't want that, do we? And I'll say right away, I'm truly impressed by what this thing is capable of. I was skeptical when I was deploying all this, but it's impressive. Subscribe to the channel, like, leave comments, watch until the end, and let's get started. First of all, what is KlodBOT? It's an agent that was developed by this person, Peter Steinberg, this developer from X. And actually, before all this hype, I followed him for some time and observed what he was doing, and planned to test Cloudbot. But the mainstream has slightly outpaced me. This AI agent is built on the basis of clot SDK, meaning it has all these functions that ClotД has. Hooks, skills, clod MD, and all these wonderful gadgets. But it's distinguished by several interesting things. Memory. The thing is, Klodbot remembers absolutely everything. And it's also worth mentioning that Anthropic approached the developer and said that this slightly violates copyright and to rename your agent to something else. So now it's called Molt. And like, that's the joke. I'll leave a link in my Telegram channel about how the memory works. It works interestingly. There are different indexing systems, a rack system, and simple search through markdown files. But the trick is that it truly remembers all information about you and can access it to provide you with personalized answers. The thing is proactive. You can set up various webhooks and triggers. And the bot will write to you itself. If you receive a notification by email, Ail, or if there are 20 minutes left until a meeting in your Google Calendar, all this is configurable. Everything is configured through a dialogue with the bot, and it is capable of being proactive in communication with you. And finally, it has no guardrails, meaning limitations. It immediately starts executing all existing commands out of the box, having access to your computer, local or remote server, and calmly does whatever it pleases, including changing its configurations, which means it is capable of improving itself. But there are many risks here. Essentially, to install it, you need to perform literally one action. Here, we go to mol.bot, as I said, it's called that now. We copy this command, open our terminal, and enter it. And in that case, you will install it locally, and it will have access to all your files on your computer, which is catastrophically dangerous, and I do not recommend you do this. We will deploy mod/clodbot on a remote server, where there will be no risks of this thing gaining access to our files and accidentally or intentionally cleaning them up, destroying them, or creating any vulnerabilities after some prompt injection. Let's proceed without further ado to deploying a server using Selectel. Everything is configured very clearly here, and in just a few clicks, we will be able to create a server, deploy it, and then proceed to install moldbot on the server. To begin, open the myselect.ru panel. I'm leaving the link in the description. If you don't have an account yet, click register. It will take just a few minutes. If you already have an account, like me, then just log in. The panel is where you can create a server, connect disks, network, and then manage all your projects in a couple of clicks. You can top up your balance by clicking this button. After you've done that, we need to select a cloud server. You can click on Products, Cloud Servers, and create a new project if you haven't already. I already have a project here, but let's create another one, clod YT. Let it be called that. Great. Next, click Create Server. And I recommend leaving these initial configurations as they are. We will choose the Ubuntu operating system. This will suit us more than well. Next, you can choose the configuration, and I will take Standard Line. This is dedicated cores, meaning stable performance for a working project. If you need the most budget-friendly option, then Sharedline is renting a part of a core. And such a server can cost from 10 rubles per day. In the side panel, you can see the cost for our config per hour, day, and month. In principle, the minimum configurations of two cores and 4 GB of RAM will be enough for us to launch all this. Therefore, you can safely choose this option. I would increase the disk size to at least 20 GB, but in reality, you can always scale your configurations later. This is also done very easily. If you realize that you don't have enough current resources, this can be quickly fixed. Our network will be private. Next, you should generate an SSH key. I'll show you how to do this now. I already have one generated, but you probably don't. So, we'll go to the terminal and enter the following command. And I will overwrite my previous key. You probably won't see this prompt. Next, come up with a password and enter it. Press Enter. It's normal that you don't see the password itself in the console here. And great, we've generated the SSH key. Next, we enter the following command. And now I will show you my SSH key. Well, just so you understand how it all looks. But under no circumstances should you distribute it anywhere. Before recording this video, I will delete the entire server and all the keys and settings that will appear here, I will of course catastrophically erase all of it. But this is what our SSH key looks like, which we need to add to the SelectT panel. We copy it, starting with the SSH value, enter it here, and click Add. Actually, that's it. We can click Create Server. And while the server is deploying, I'll tell you about another cool feature of Select. You can freeze it in the future. If your project is on pause, you won't have to pay for the server, for the CPU, for the RAM. You will only pay for the disk and the public IP. And honestly, it will cost absolute pennies. Well, and if you expand your project, you can create a full-fledged infrastructure in Select thanks to the ecosystem of services. You can deploy cloud databases, connect Kubernetes clusters, object storage, content delivery networks. And you can start using the server immediately after ordering, because Select offers automatic operating system installation. Go to the my.select.ru panel, I'm leaving the link in the description. And if you need a server for your project, then Select is a fast, clear, and reliable choice. Great. The server is ready. We can start configuring it, but, of course, we first need to connect to it. We'll exit here, delete everything unnecessary, and enter the command that Select provides us here. Via SSH from our console, from our terminal, we will connect to the remote server's console. Next, I simply suggest that you follow all the commands I show sequentially and attentively. I will try to explain them clearly, but in principle, if you do what I show you, you will definitely be able to set everything up. We'll enter the word yes, because we want to connect. Next, we need to enter the password that we previously set at the SSH key creation stage. And behold. Look, we are essentially already on our server and can execute various commands here, which is very pleasing. Now, it's important, if we don't make some settings and don't enter some commands that will allow us to strengthen our security, there are many risks that someone from the internet can connect to our server, because it will be accessible to everyone on the network. And I'll say right away, I'm far from a developer, I don't have a strong IT background, I learn everything myself, but I've generally embraced this whole story, tried to approach it responsibly, researched some information, and talked with the LLM. Therefore, I will show you my config, which, in my opinion, looks quite reliable now. You can repeat all of this. We'll start with some basic hygiene. We'll enter the command you saw a little above, just to update the operating system and update all old versions of libraries and utilities to the latest states, which have likely fixed some of their bugs. Now we've connected to the server as the root user, but there's another basic rule. The principle of least privilege, when we will run the bot not as root, but as an additional user, because then access will be limited to certain folders of that user, not the entire system. Therefore, we enter the command add user and here we enter the username. Let's enter, well, like this. Let our user be called. We need to create a new password. And enter this password. Come up with it again. Well, and try not to forget it, of course. We'll need it again. Next comes some settings that we can actually skip. Just by pressing Enter, Enter, Enter, Enter, and Enter. The information is correct. Enter Y. And done. We've created a new user. After that, we will enter the following command so that our user does not exist, of course, it doesn't exist, because I called it differently, I could use the sudo command to change various system settings. And it's not exactly super secure, but on the other hand, for the agent to be truly autonomous, we need to grant such access. Next, we will enter the following command to transfer the keys to the new user. Here, accordingly, enter the name of your user that you created. Mine is clod, I remind you, to avoid further password logins. And passwords can be guessed or found, but it's harder with keys. Well, and let's end this session and try to log in with the new user. Instead of root, I enter here. and the same IP address that we have. Super. We can now enter the password we set and open our remote console. With the next command, we will further strengthen the security of all this. We enter the password again. And here it asks me to grant certain access to my terminal. Probably, if you use a regular terminal, and not, like me, a guest one, this won't appear. Well, next, using the sudo command NnoO and the path to the folder, we open the SSH config file. Enter. Simply using your mouse or touchpad, we scroll to the values we need. And these values are somewhere here. Permit root login. Here, instead of prohit password, we need to enter just one word and two letters that make up the word. no. Password authentication. Here, we also leave no, as it is already here. And the authentication folders, you see, it's commented out here. We also need to fix it. Just remove this hashtag here. And that's basically it. Next, press Ctrl O to save the file. After that, enter the command Enter and then press Ctrl X to exit the file. Next, we can restart the entire system. And what did we just do? We've only left keys for accessing our remote server. Passwords and attempts to log in as root are now completely blocked, and we are not afraid of this. After that, you need to configure the Firewall, meaning close all ports except SSH. The fewer open ports we have, the lower the probability of attacks. We only allow necessary access for administration. Therefore, enter the following commands after me. Suda up install flagway. WF, to be precise, UFW suda UFW allow open SSH. Excellent. And finally UFW enable disdrupt cs, allow. Enter the letter Y and press Enter. And here you can now close this session, try to log in again, check if you can log in with passwords. In theory, you should no longer be able to log in with passwords. And next, we install clдbot. Finally. Yes, congratulations, you've reached this stage. I'm also very happy, honestly. Here it is, our wonderful clё command. We copy it, enter it, and behold, the download has started. After this download is complete, it will be quite fast for us, we will start the onboarding. I look a bit too usual in the video today. I think I'll probably refine my image a bit while this installation is still ongoing. I congratulate you. Cloudbot is alive here, ready to heed our commands. And it immediately prompts us. This is very powerful and probably risky to continue. The developers absolutely do not hide that this is a risky thing. And I hope you've already understood this from my numerous, uh, admonitions and, in general, stories about this agent. Yes, we understand this perfectly. Next, onboarding mode. We choose Quickst. And then we need to connect a model that will power all of this. There are quite a few options. You can connect, in general, codex with an API key or simply with your ChatGPT+ subscription. Probably the most optimal option, the cheapest, but I will probably use Open Router, as I already did during testing. Open Router API key. We need to enter it here. I'll go to Open Router. It's somewhere here. Let's make a new key. Clot YT. Great, create key. I'll remind you again, don't give out your keys to anyone. I will hastily delete it after recording the video. And of course, the main feature of the router is that I can choose absolutely anything. In my tests, I chose the Gemini Flash model, and it pleased me greatly. It responded incredibly fast. Gemini 3 Flash Preview with a million context window. In general, honestly, I liked it. Next, we can configure our Klodbot to communicate with it through some messenger. And probably the easiest option is Telegram Bot API. So let's choose it. Well, maybe while it's thinking a bit, we'll create a new token for a new bot. Let's create a new bot. Clod bot. YT. Clodbot. YTbot. This will be its peculiar name. Create bot. Done. And then we need to copy the API key of this bot to, in fact, add it to the settings of this whole gadget. Aha. Telegram token, it asks us to paste. I paste it, press Enter. And don't forget, you don't need to show this token to anyone. I will also delete this bot. Next, we can configure, set up skills. And in general, skills are what power our Klodbot. It draws all the necessary information from there. And in general, it amazes everyone with its capabilities because it does all this through skills. That's something that Antropic invented. Let's configure them. Yes, homew install command will be useful. We click Enter here. Note manager to install skills, we will choose npm, because it's the most familiar, understandable to many. Here you can scroll through all of this. I recommend you choose clothub, there's a huge number of other skills that your bot can pull. And also, if something catches your eye and you decide to install it too, then do it. I'll install Videoframes, I'll also upload 11 Labs right away. Some Open Aper probably won't work for us. Model usage. Well, and in principle, that's enough for now. You can always return here. I'll show you how to do it. After you've selected all of them, press the spacebar to select and deselect. And Enter. The installation of the selected skills begins. It also offers us to enter several keys. I'll click No here everywhere. No, I don't want any. Not with Google, not with Gemini, not with OpenAI. All of this is now. Well, actually, you see, Labs key. I chose the skill, but didn't enter my key, so it's unlikely to work. Regarding hooks. Here I chose command logger, session memory, boot. Let's select everything here, it will be useful. Press Enter. Gateway service installed. And here, in fact, I didn't press HCH ini. Well, let's press HCH ini. Ah, but we just started a session with the agent, and it's offering to configure it. In fact, I don't really want to do that now, because, in general, we will communicate with it through Telegram. And I'll also show how to open all of this in the web interface. But let's continue with security settings. I suggest you open this folder to see what's inside. Nano and the path to this file. And here it's important to note the following. The most important thing is that we have this bind here, and it says lbck. If for some reason it's not like that for you, then set this value here. This will allow you to avoid unauthorized access to the console that we are about to open from your device. Again, please do not show this token that I showed you here to anyone. Ctrl X to exit. We've created the bot, and now we can write to it simply to find out our user ID, because we need to ensure that only we can communicate with this bot. And for that, we need to set certain settings. The first setting in our console will look like this. I have some minor issues with the binary, as far as I understand. Well, the bot itself prompted me above that I need to enter this command. Let's try to do it now. Super. And again, enter the command CHK ClДbo. Super, everything worked. Perhaps you won't have such problems at all, but okay. Let's return to our previous command, the command to set a specific user ID in Telegram. First, enter the command to update this policy. We only had an allowlist for a specific user ID. Then we restart this whole gadget and enter our Telegram ID here and return here. Enter this command. It seems to have worked. We need to update with this command. By the way, I'll remind you, if I haven't reminded you yet, haven't said it, I will provide you with all these commands. It will all be in markdown files that you can get from my Telegram channel, so that you, of course, don't struggle so much. And as if, let's enter колбот status, let's see what it tells us. Uh, it seems like everything is okay, tokens are configured. The agent is our main troubleshooter. No troubleshooting. Let's go to Telegram and ask: "What model are you, my dear friend? Something is typing. I am working on the Gemini 2 Flash Thinking Preview model through Open Router. This is some absolute nonsense. It hallucinated terribly, but it answered us. And yes, you can start communicating with it on Telegram. Look, my dear friends, the video is getting catastrophically long. I'm showing a few more important commands that we need to enter. I forgot again that this is now not clod, but clД YT. This whole thing is called clД uh. The first command you need to enter. Well, enter it. Next, what else interesting can we do? We can enter, I think, clod bot and generally enter the entire list of commands that it offers. And here it will offer us one very, very sensible command that will help us set up security, that is, safety. Again, enter CLДbo Security. I think there was a double dashp. Let's try this command. Ah, I almost guessed it. This is the command we need: Clot Bot Security Audit Deep. And it will give us, in general, all the warnings and all possible problems that we currently have with security. I'll just recommend you to copy all of this and send it to some other agent or even to the same Klodbot and ask it to fix it if it sees certain problems here. But there's another important thing that I can't not tell you about. This is the following command, which will allow us to set up SSH access to the panel where we can view all the statistics and everything related to our Klodbot, how it works, monitor logs. And here we need to enter the following command that you see on the screen now, and of course, take your IP address, which, I remind you, you can get from the panel where we created all of this. We are now creating SSH access to our Ctrl UI panel. Don't forget to enter the password. Great, we've done it all. Leave this terminal window as it is. And now we will enter the following address on the network. And if it tells us that we are disconnected, then another command will be useful to you. Yes, I know you're cursing me now for all these commands, but it's still security, it's important to understand. We enter this command on our remote server to find the necessary token that will allow us to obtain this SSH UI. Now the correct link, which we take, copy, and again go to the internet. and enter this link in our browser. Excellent. We can see here everything that the bot is doing, what tools it is using, how it works, what skills it currently has. You can even connect these skills here manually. Let's see what some BL Bubbles. Blue C. Well, let's install. BL C doesn't want to, it says, through Vibock Watchers it can. It doesn't want to. Well, let's install some bird. We should have PNPM. Twitter. Oh, we'll be reading Twitter now. This is very good, friends. I've shown you how to set up this whole gadget. Further, I urge you again that you need to be extremely careful with everything you do with it further. Giving out API keys is quite dangerous. Giving access to your email is quite dangerous, because there is such a concept as prompt injection. And if the bot goes online and stumbles upon something, it can then, in general, forward all the secret data you have to someone. But as if, let me briefly show you what I managed to do with it in my tests. For some time, we debugged everything for a long time, after which I told it: "Let's, I want you to respond to me using audio, to be able to process my audio." It didn't cope with that without API keys, I didn't want to give it any API keys. And after that, I ask it to respond to me with voice, and it finds some way to start responding to me with voice. And then I ask it to do this every time, it remembers it, and in each of its messages it adds a voice message, an audio recording. It's unavoidable. >> Vova, a picture from D's article. This is the main image of the news about the launch of KMI's new model from the Chinese company NSH. >> That is, this is, well, a normal, adequate voiceover. And it's not Level Laps, but who the hell knows what library, where it got it from, and it added it, and it continues to communicate with me. After that, we connected it to an. After all, I gave it an API key to my Brave, and it was able to, in general, summarize all the articles for me, send pictures from these articles. And this is, well, it's very impressive, the use that you get when you start communicating with it. Simply through chat, you perform all the settings for this bot, and it truly obeys you, but it periodically hallucinates. In general, it's good for you to understand how it works. Go through all the steps that I've shown you to make the necessary security settings and go to this console and periodically monitor what it's doing, what tools it's using, so as not to stumble upon some prompt injection problem, and in general, to roughly understand what's happening here and what the bot has already managed to do. I hope the video was useful. I will probably shoot another video, because, as you may have noticed, well, there's a lot to talk about here, about different further use cases, we've talked about settings, but there's still a lot of work to be done to make this thing work for the benefit of individuals or all of humanity. Subscribe to the channel so you don't miss new videos, new episodes. Leave comments, likes, encourage us. We'll make all sorts of useful things for you. This was Dya Dev. See you in future episodes. Bye.