Как настроить Claude Code за час и получить второй мозг для решения любых своих задач

Oh, yes, everything went great. And even the cart started working right away. Generally, it's just some kind of space. Super, super. Today is a simply fantastic day. Friends, hello everyone. With you is Timur Ugula. This is Neromasterkaya. Here we continue to study what is arranged in neural networks and what good, useful, bright, kind, pure things we can do. And before I introduce the guest, I will traditionally say a big thank you to our friends from KOS. KOS are our information partners who have really coolly integrated into our project and help us. We help them. And we have such a win-win. So, our guest today is a wonderful person. Now we will bring this wonderful person to the air. Sergey Shima. Firstly, he is the author of the II Masters Me channel. A magnificent channel. You need to go there and read. And it's cool there. And I write little, I only read. But know this, I always read. And he is also the director of TDI AI Creative Agency and a person who will tell us today about such a scary story as Claude code, code Claude, Jean-Claude. In general, what can be done with this if you are not a no-coder, not a programmer, and generally not a techie? Hello, Sergey. >> Salute. Hello. Sorry, glad to see everyone. Literally, literally 5 minutes ago, well, maybe 10, there was a conversation with my close friend. He is a humanitarian. Absolutely a humanitarian. A real marketer, a humanitarian marketer. >> You said that as if it were something bad. >> It's, well, I'm the same, everything is fine. But he is not very deep into neural networks. And we had an interesting conversation on the topic of what, listen, well, how is it all about you, well, about this technical stuff. You will now show how to do something, and then I still won't be able to repeat anything on my task. Is that so? >> Yes? >> I answered him the same way. >> If you don't need it, well, yes. Our task today is to show why it is needed, and to lead by the hand step by step so that it is easy for the little ones. >> Super. And the question, will there be a recording? The recording will be. Visible, audible. That's it, we are fully on air. Come on, I'm handing over the word to you. Tell me, please, what are we going to do now. >> Okay. Yes, thank you again everyone for joining or watching us in the recording. Today I will try to tell you what code is. I will not bring you theory, I will try to explain and show with examples as practically, practically, uh, aesthetically as possible, how I use it, what tools there are, what settings, how to connect it, where to connect it, because many people have a problem with how exactly to use it, especially in countries with restrictions where they don't allow us to use it. And even in my country, where I am in Uzbekistan, everything is okay. But sometimes I also use proxies, because, well, it's easier to communicate with it through some separate server so that it runs there, and nothing remains on my computer. In general, I will lead you by the hand, show, tell, forgive me, show, tell. As usual, when I was preparing today, I thought: "Damn, what to tell? I opened it, I installed it, and then I started thinking, I'll show this here, this here. And I still got a big board. Is my screen visible? Can you display it? >> Now it will be, yes? >> Here is our today's uh canvas, right? We will start from the top. First of all, whoever wants to jump in faster, you can scan the address of this board with your phone. Uh, it's too long to write it down somehow. >> Wait, wait, I'll scan it now and attach it to the comments. >> Yes, you can throw it into the chat. Uh, if anyone is interested, subscribe to my Instagram. Well, nothing special is happening there. I'm not very good-looking. >> Banned in the territory of the Russian Federation. >> Okay. Yes, well, I also banned myself there. I'm not good-looking, it's not interesting for me to look at myself and this beautiful life. A beautiful life surrounds me. There is a link to my, uh, Telegram channel here. And I must also, as a matter of obligation, well, since I am the guest today, talk about my website, which I assembled for New Year, also with the help of code, where I packaged, in general, my product. For three years now, I have been involved in education for business, not in terms of how to write prompts, but more for large corporations where people want not so much implementation, but alignment, so that everyone understands and communicates in the same language. And where I don't explain theory, but we analyze working cases. That's why I like to come to businesses, listen to their problems. Here is the link, you can go and see. And I like to communicate with people and help them find themselves in this madness, in this FOMO that is happening around AI, because because I have seen too much, tried too much, and I don't need to chase after shiny things like magpies, but I offer only what works. Okay. I have completed the mandatory program, yes, I have told about myself, given the board. Look how many people are hanging out here. Let's try to dive into what awaits us today. First of all, I will be talking about code today, not about its capabilities, but specifically about how to launch it, how to use it, how to configure it, what to pay attention to. Moreover, we will talk about both local launch and server launch. It's not difficult at all, guys. It's really not difficult at all. I will show everything step by step. I will tell you about my system, which is called Edgarain. This is my second brain, which allows me to structure everything that happens to me very well and coolly. I have brought it to Telegram, so here is the Telegram icon. I can even show how it works at the start, so that you are a little interested. Here it is, uh, right here. Uh, I asked it to output a couple of stupid messages so that some sensitive information is removed. Here. Uh, it has, well, what I will show now is a slightly more upgraded version, which I will release soon. Because now you can communicate with it as an agent. Before, there was just a forwarding of some information. For example, you could, and you still can, send voice messages. Let's try. Hello. Tell me, how are you doing? We are currently recording the broadcast, which I told you about, so do not share sensitive information, do not show what projects I am working on, but just tell me how many tasks were completed last week and how many tasks I have this week and how much this aligns with the goals. Do not say what my goals are, do not say what these tasks are. Just list them and give the percentage of alignment. Here I sent him a voice message. He should process it, in theory, if I haven't broken it. Possibly, I broke it. >> Does he compare the task with some task manager? >> No, he compares the task. I will tell you later, no task manager, it's just a file system. And how it is structured, I will tell you. There is nothing there. In principle, there is, of course, connection to external data. I will also tell you about them, but the base is a stupid file system. It's a folder with many, many folders, many, many files. He navigates through them, uh, he interacts with them. And I will tell you how this happens, so that you understand it, because for some it seems like magic, but it's not magic. Okay. Maybe it will think for a minute or how long to answer, because it's not ChatGPT after all. It needs to search for something first, then collect, then answer. We will return here, we will definitely return to this screen. Let's start with the very base, how to install code, because for many this is also a problem, what it is, how to install it, how to communicate with it. All you need to do is type in Google. I'm telling the truth, today I will show everything specifically. Plot-code. Uh, yes, by the way, he answered. See, he wrote: "Here is a picture without details, last week, the fifth week of the year. 38 tasks were planned, about 70 were completed. Many tasks have rolled over to this week, as many as 15. This week we have 78 tasks. You see how much I have. But these are really big projects. Right now, 17 tasks are still in the backlog. Well, you see, the alignment with goals is 87%. In principle, you see, he answers voice messages. He answers uh, text, if I describe it to him. It's also convenient because I throw everything that comes to me into it. Let's meet. I throw in pictures, posts that I like. And he structures it later. But we will get to that, because for now it's a rabbit out of a hat. And let's start with how to install it. Here we find, in essence, the first link. Code tries to outbid. Here. Yes. This is the documentation. The address is clot codcod.com clotcд docs and here there is, well, of course, everything is in English, don't worry, sooner or later almost everything will be in Russian. There is this link, which looks like this. It looks like this, I won't read it, I'll read it badly anyway. I click to copy it. And what do we need to do, friends? We need a terminal. A terminal is this thing. If you are on a Mac, you can find it here when you type terminal app, it looks a little different here for me. I can make it bigger with commн plus. Here it will open somewhere. It might look a little different. It's already customized for me, but in general, it will just be a black field, a cursor, and nothing more. And you need >> it looks scary. You will get used to it very quickly, because all the fun happens in it. You paste this link, which I just downloaded. I pressed Ctrl V, Command V, and press Enter. Uh, I won't do it on my computer right now, because everything is already installed for me. But it will guide you through a dialogue, and you will have almost no errors. If you encounter any error, well, let's, okay, let's press Enter. It's not scary, it won't reset mine. It will take some time. It will be, I don't know, maybe a minute for someone, three for someone, five for someone. But if you encounter an error, it stops somewhere, you don't understand what happened, or you don't even understand if the work is finished or not, you can copy all of this, if you saw an error, highlight it, copy it, take it to ChatGPT, throw it in there, I don't know, into the simplest model. I'll remove everything now, yes, and ask what it is and how to help me. Just copy any errors. Even ChatGPT helps for free. And here it says what it is. So, uh, something is happening, a street for downloading, trolling. In general, it describes everything. If there was an error here, it would say how to fix it, what needed to be redone. So don't be afraid. The fact that it's hanging is normal. It really works like this. It can hang for a while and then, then it will pass. Even if you don't have a VPN, then, well, you see, network code, even if you don't have a VPN, then don't worry, it will install. You will need a VPN or proxy when you communicate with it. See, it says installation complete. Uh, that's it, great. Absolutely nothing, not a single request was made. Well, because my environment was already prepared. By the way, if you are on Windows, dear friends, if you are on Windows, then I strongly recommend typing these three letters, which are called VSL, and reading what I moved somewhere else. Here. And read what Windows Subsystem Linux is. This is, in essence, you run Linux, a system on your Windows and work with it in the terminal. Why is this important? Because because everything you do, code does, on Mac and Linux, almost everything is files that can be read. Can be read. Can be written or created, can be edited, can be searched, can be deleted. And these commands, these are the tools that code possesses. Everything that is under the hood is a fairly smart LLM model, like ChatGPT. Well, it's a model. And it has a set of tools that it knows how to call. What does it mean to call a tool? It's to type a command in the terminal, for example, C. This means display a file for me, the file name, and you will see what is written inside that file. We won't learn commands now, all this You will need to learn only four commands. I advise you to learn, even essentially three. That's it, never memorize anything else. Three commands. The first command is called CD - it's change directory, change directory. You type CD, space, and the address where you want to go. For example, you always have some folders on Windows, the address is directly visible, C, colon, slash, something further. Copy it on a Mac, if you work in Finder, for example, let's create a folder in my projects, somehow here, by the way, it's always a big problem on a Mac to create a folder. Let's call it NNN. I don't know, NnN let it be, yes. Okay. Here is our folder. When we entered here, not even entered, just right-click on the folder and press option. Option Alt, yes, and you will see Copy NNN here. If you can't see it, let me zoom in. Does it zoom in, Timur? >> Yes, yes, yes. Excellent. >> Here, if you press option, you will see Copy nns path name here, that is, the path to this folder. You click on this thing, you will understand now why I am showing this, because this is really a command that you need to learn. Here. And then, for example, I want to go into this folder, start working inside it. And for this, I press CD and paste this path. You see, the full path that leads to this folder. I press Enter. And, as you can see, my address has changed. I am inside the Projects folder, inside the NNN folder. All files. And you can't, well, that is, you can run code directly from the home folder, where you have all the files, but I don't recommend doing this, because you are giving it access to everything. This is bad, because sometimes it can make a mistake and delete something unnecessary. Always create a separate directory in which you want to work. For each separate project, you should have a separate directory. For example, if you want to create a personal knowledge base, please create it in a separate directory. I did this for testing. I dumped a huge number of markdown files into this folder. These are text files where who knows what is stored. My notes. I exported them from my notebook. These are just my notes over a long time. And now I want to go into it. I have a slightly different browser here, so there is this copy pass. Copy path. Let me change the directory. Change directory. And paste. As you can see, I have moved into the folder with my notes. That's it, this is a command you need to remember. The second command is the same, only if you want to go up one level, go back, like change two dots. If I do this, enter two dots, you see, I will just go to the Downloads folder. Before, I was in Downloads test folder, and now I am just in downloads. It's that simple. There are two more commands that I want you to know. They will help you a lot. You will always be aware. Sometimes the address is not written here. I'm telling you, this is my customized environment. You don't know where you are. Type the command PVD. This is print working directory. Like, tell me where I am. And here is the address, it says: "You are here, it's easy." The last command, honestly, the last command is called ls ls - it's list, that is, display a list of files and folders. And - this is an additional parameter that displays them beautifully in a list and shows all hidden files. And now, for example, I want to see what I have in the downloads folder. In any folder, I type LS LA. And you see, everything, everything, everything that I have in this folder has been displayed. You need to look here, here, here, in this column, because folders are marked in blue, files in white. I need to find my folder, which I had before. Here it is. Test folder MD. I can copy its name and go into it again. CD test folder MD. Everything is okay. One more bonus command, which you will sometimes use, because I don't like this mess on the screen. Just type clear, wipe clear, that's it, and everything is clean again. That's it, no more commands. The last, or rather, not the last command to launch code is code. Just like this, code. And this cool little guy will launch. >> A red little man, >> yes? A red little man. The folder you are working in will be displayed here, the model will be displayed. But on the first start, unfortunately, you won't have this. You will immediately have a connection dialogue. I will do it on my end now too. When I press the login command, you will have this dialogue. And here you will need to choose, don't even read these two things, they are not needed. You need a subscription to Claude. Any subscription Pro costs $20. Max subscriptions are already $100 and $200. This is not interesting for us. This is for large companies. If you have a subscription, you have it, you choose this option. I move with the arrows back and forth. Choose option one. It opens your browser, where it will ask you, if you are registered, it will immediately redirect to authorization. If you are not registered, it will ask you to register, pay, and you click authorize. That's it, everything is ready. As you can see, we return here. It says login successful, that is, I logged in, everything is great, let's do something cool together. I press Enter, and I return to the dialog box. You won't have this. This is my custom setting. It says the model, the context window, how much I have left, and the working folder. All this can be configured. I will tell you later where to look to configure it. But everything else will be done by code itself. Absolutely everything. The very first command you can learn, and code has many commands. To understand what commands code has, press /. And here are all, all, all commands, your commands, its commands, I have many of my own commands and various skills, but in general, you will have a small list of commands. Try to read them at least, what they mean. Here it says, it's a little visible. Here I have many of my custom commands, so don't look. Here is one of the commands you need to remember, also clear. Clear the context window or the compact command, like summarize the previous work. Or the config command. Here, config. A cool command. You can read what it does here. For example, show hints. True means, well, true. FALSE means false. Like, it will show hints. Less movement. I set it to Fils, it's not scary. Skining mode, that is, thinking mode is on. Suggest prompts, on, roll back if something didn't work, on. Verbow output. Well, show it, in short, take a screenshot of this, show it to ChatGPT again and ask ChatGPT what it is. Explain. And it will tell you what each command means if it's hard for you to read. In general, we don't need much here in the team, in the config, except if you are a complete beginner, I would definitely choose the output style here, that is, the output style, I would choose it as a learning style. It will tell you more about what it does, but sometimes it talks too much. I sometimes myself, although I already know well, choose the style called explanatory. Exclatury. And it will always explain, give some insights, what it is doing. It is a very, very responsive tool. To understand what it is. One of the first commands, as I said, is the command called Init. Init. You see, it says Initialize New Cloud MD. This is a file that describes what is inside this folder. When I run this command, it starts spinning there and will write to you. First, it will write to you in English, but if you start communicating with it in Russian, it will answer you in Russian. Very responsive, as I already told you. It will ask for permission. It says: "Can I run this command? Mostly always say yes." Or choose yes and don't ask again. This is normal. For beginners, I recommend always looking at what it asks. If you don't understand something, throw it into ChatGPT again and ask what this command is. It's unlikely to offer to delete the entire system. It's a smart enough thing. It won't offer you that. But sometimes you might see somewhere that it went where it shouldn't have, and ask: "Hey, stop, what are you doing? Don't do that." You can prohibit any command. That is, it will be step-by-step. At the start, please act step-by-step. Don't try to rush it, because sometimes it can lead you where it wants. You see, it read my files, it saw that it's in Russian, and says that this is my vault. Well, that is, the place where my obsidian notes are stored. That is, 14,000 14 files without nesting, uh, without anything, without any Git, it says, nothing is clear. But it writes to this file, clod MD. It creates it and says, what is described here. This is a flat structure. There are a lot of files, what is marked in red, it deleted, in green it added new. In principle, it says: "Can I write the file clotdi?" You say: "Okay, no problem." And by the way, also a small life hack that you saw. Shift Tab switches modes. For example, when I press yes, and now I press Shift Tab. Look here. You see, I am now in the mode of approving all changes. I can switch to planning mode. Then it can't edit anything, it only gives hints and plans. This is a very cool planning mode. First, discuss something with it, plan, then act. Another Shift Tab: no mode. And the mode of accepting edits. Here I click, the modes change. Two main modes: planning and edit acceptance. That's it, you have no more options. In edit acceptance mode, it edits, in planning mode, you communicate with it and ask what to do. Here it says: "Uh, deleted some structure, all files are there, everything is okay. We will come back to this." We will come back, because I am showing how to create your own space. Here could be PDF files, here could be some Excel files, well, it would be harder for it to work, but I just have markdown files here, which have the MD extension. These are ordinary text files, which contain just written information with this markup, which is understandable to the neural network. Let's go back to our board, because we have prepared so far. That is, let me remind you again, Claude has the ability to run commands in the terminal. Commands for reading, writing, editing, searching, and deleting. That is, it can search your database, find something within these files, read and edit them. And then remember these commands. The most important thing is to change the directory. Always work in the project directory. Created, want a new project, create a new folder, go into it, launch code there. If you launch code, simply, we have a signal. Signal Evgeny writes: "I, honestly, have completely lost track of what we are doing." Okay, good. We are not doing anything yet. We have installed code. We have installed code. If it's difficult for you, we will install it again, but in a different place. Why? Because I draw your attention again, this is in the terminal. I will issue the command Q exit. Well, that is, exit from it, exit from code. Let's start from the very beginning. >> I'll clarify. Just in case, when I post the recording of the broadcast, I will also make a short guide, so that it is possible >> not only to watch, but also to read? Uh-huh. >> Okay. Look, so, uh, the core, code for us is just

No, not a model, an agent that knows how to navigate folders, search for files, modify them, assemble something from them, because everything is files. If you are creating a website, the website is assembled from files. If you are creating some kind of bot, the bot is also assembled from files. If you simply want to create a knowledge base for yourself, upload all the files into one folder, launch the Claude and say: "Figure out what's going on inside here." This is a thinking and editing machine. This is an agent. An agent that can do everything itself. To this agent, things, so that it works well. The first thing we understood, what is it, context. These are all the files that we upload. Your prompt can be context, a request, like do this, do that, we are located there, I want to do something else. This is also context. We need to connect to external, uh, external data, for example, to email or, for example, to search or, for example, to something else. For this, MCPs are needed. And we also have something like skills. These three things will enhance your Claude. But for starters, we figured out what context is. Context. We communicated through it, within it through the terminal. Claude himself made a bot for me, through which I communicate. I showed this at the very beginning. This bot is here. If you go to this link, click on this thing, go to this link, then you will get to GitHub. We will talk about it now too. You all need it very much. And here there will be a description in Russian, what it is, with full installation, how, what to do, why. You can read it, but it's even cooler if you just give this link to Claude, and he will do everything for you. And here I want to show why we need an external server. It is very, very necessary, because, again, we have limitations of these companies. They do not allow us to communicate with their models in some countries of ours. And, uh, there is also such a concept as the server always works. And if I installed, I installed it here on my computer, yes, here is my terminal, it is on my computer. I closed the computer, finished my work, closed it, went to eat, that's it, all processes are stopped, nothing works. But the server always works. For example, when you communicate with a bot at any moment, even from your phone, it works on the server, and this is very important. The server is not located in forbidden countries where everything is forbidden, but in normal countries. I recommend Kantaba server. I will explain why now. It is very cheap. It is not always stable, but it is super cheap. And, in principle, it is enough to set up a proxy there, so that you have your own, well, I don't know, VPN practically. Well, not a VPN, but like a place through which you communicate with Claude. I have such a diagram. I will show it to you. We will return to GitHub now. The point is that if you are used to connecting to the internet without a VPN, it will directly connect to Claude, and Claude will tell him: "No, sorry, dude, your IP address, that is, where you are located, we don't provide our services there. But if you set up a server and install a proxy there," then you will send requests through your server, and they will always work on your home computer and on the server. The only thing you need to understand is that the server is needed in another country. And that's why we need a contact VPS. Uh, here I am on it. Choose the cheapest option, the cheapest one. Uh, we don't need better. The cheapest one will suit us. Nothing needs to be changed here. Here, like, choose this, all this is selected. You can take it for 12 months, for 6, or for a month, but it's more expensive, like 5 bucks. Let the server be in Europe. We will have 150 GB of memory. Ubuntu system. Don't touch anything. Don't touch anything at all. We don't need anything here. We don't need anything here. We don't need anything at all. Don't touch anything. Just write the password here, with which you will log into your server. That's it. That is, by default, as they gave it to you, just choose one month, for example, to start and choose a password. Then there will be payment. Payment goes through almost any card. Uh, everything is okay. It goes through with a Belarusian card, for sure. I know that. You will receive an email. I specifically bought a test server, I don't mind showing it. Here is the email I received, and I will delete it later anyway. Here it will be written, here you have, excuse me, you have an IP address, like, which is assigned to you. This is your address on the internet, not unusual letters, but numbers. And this is normal. And your username. Usually it's root, well, like the root user. And then you need to set up Claude code on the server. It's not difficult. Everything is done through the terminal. You can ask ChatGPT how to connect. Send this email and ask them to connect. But the point is that you need to click on the command that looks like this: SSH, then the username, @, and this server. I press Enter, and it will ask for the password. The password that I set here. I enter the password. It's not visible when you type it. This is normal. This is a completely fresh, bare server that I logged into. And it will show you that everything, you are inside. It will also ask you if you want to log in for the first time. Well, everything is okay. So let's go to a bare, completely bare server now. That is, if I press ls la here, like, what do we have inside? Let's display the list of files. Uh, nothing, you see, just empty, absolutely empty, some command files, nothing else. Empty, absolutely empty. Right. Let's enter a command here that installs Claude directly on the server. This is the first command. Don't paste anything else here. Paste this command, press Enter, and Claude will be installed here, and it will live here. This will be its home. And you won't need any VPN for this. You won't need anything. It lives here. It is located in Europe, and the only one who can access it is you, because you have the password, you have the address of this server. Don't give the password to anyone, don't give the address of your server to anyone. This is exclusively your property. It's like a key, a house. A house in Europe for your Claude agent. You could say that. While it's installing, I'll explain what else you need. Absolutely. Why I say absolutely, get yourself a server. It will give you so much benefit that these 5 dollars, these two cups of coffee a month are nothing. While it's installing, I'll tell you one more thing that you will need. You will need a GitHub account. It's absolutely free. Absolutely free. A GitHub account. What is GitHub? GitHub is, in essence, a social network for developers, I would call it, where they brag about what they have done. For example, if I go to my GitHub, well, here, here I am. Create a profile here, honestly. Well, it takes 5 minutes to create a profile. This is my alter ego. If I were an IT person. As you can see, I have 109 repositories here. Repositories are projects, they are folders. So I have 109 project folders stored here. This is cloud storage. This is cloud storage where code is stored. And plus, here, if you go to the main page, if you follow someone, and you can follow someone, it will always be interesting to observe who, what is coming out of whom. Why is it cool? Because GitHub stores what is called open source, open sources, code, just specific code of ready-made applications. Take it, download it, adapt it for yourself, and do it. And I have a friend Ashot, who said a very cool metaphor. I like it very much, I will repeat it. That if when you want to create a program with Claude, you are growing it now, not writing the code yourself, you are growing it with Claude, as if you are watering it with your data, I want it to grow there, I want it to grow there, but you are growing it from a seed, from zero. There is nothing there. Zero. An empty folder. And you will grow it for a long, long, long time. Maybe it will grow into something useless. In the middle, you will have to uproot everything and grow it again, and you will get a small tree. So, on GitHub, there are already ready-made seedlings or small trees that you don't have to grow from scratch. You can always go, for example, to perplexity and write something like this. I will use voice input. Uh, listen, find ready-made Telegram bot projects on GitHub that I can connect ChatGPT to. ChatGPT. That's how it recognized me. And it will tell you that there are already about fifty projects on GitHub. Here are ready-made projects. Here you see, ready-made projects. I'm going directly to the GitHub link. I don't need to grow this stupid little tree from scratch. Everything is ready here. There is already a ready-made project that has already been created. And I will just need to clone it. Clone it or download the ZIP file. Here you see download just the archive, or these are clone commands. And even easier, even easier, just copy the link to this project and give it to Claude. And he will understand what's here himself, and he will install it for you himself, and he will deploy it himself. And the same with my agent. The link to it is on our board, the link to my agent, and everything is described there. And Claude will also understand. Here is the one that can communicate, the one that stores information, organizes everything. And you can take this link, exactly this link that is embedded here, copy it, send it to Claude and say: "Install everything for me and set it up." Let's do that. I just clicked on this link, it will open my repository, agent Second Brain, where there is a huge long description. You'll get tired of reading it. We won't read it, we'll just copy the link. Copied the link. You see, I'm returning to the server. We are on the server now. This is not my work computer, this is a house in Europe. Remember, we logged in here, a house in Europe using the command that looks like this: SSH, then the username, then the address. Yes, I entered the password that I set, and we bought the server, this VPS, here on Kantaba for 5 dollars a month. Here we are in a house in Europe, and we don't need any VPN, nothing at all. Look, Claude was installed with one command, one single command, this one that we also copied earlier. I'm launching it, Claude. Well, look, I'm launching it in the working directory. It will complain now. Uh-oh, working directory. What do you mean? Aha, command not found. And this is normal, because it says the command is needed. I didn't read this. The command needs to be added to your internal path. Copy this command and paste it first. Now let's try Claude. Hooray, it's launching. It offers me which mode. I want to choose dark mode. I like this one. And as always, connect. Connecting. I've already shown this. Only now I'm working from a house in Europe. And it says: "Oh, dude, you don't have a graphical interface here. There's only a terminal, so take this link, go to it." Here I take this entire link. Copy. Copy this link. Open it in a browser on my computer in a browser. Paste this link. It says: "Invalid." Something didn't like it. Let's try again. Maybe it cut it into parts. Cut into parts. Look, it's not difficult, not scary. Open ChatGPT. Somewhere it cut it. There were just line breaks. Say: "Output as a link." And paste this long link broken into pieces. Here is a normal clickable link. Here it outputs it completely. Yes. Click open. Everything is fine. It says: "Log in, authorize." But since it's a local computer, it says: "I can't access the server. Copy and paste this code there yourself." And it writes here: "Paste the response code." Paste the response code, press Enter. Hooray! Loginful, we're in. >> Sergey, I'll interrupt you a little. Friends, I see all your questions. I'll ask them later, Sergey, because, well, it's impossible to interrupt here. >> Let's, let's, we'll get to the point. I, look, we have Claude installed in a house in Europe now. It doesn't need a VPN, nothing at all, and you can communicate with it. Right. We installed it, everything is okay with it. What do we do next? Since you need this agent, this Second Brain, it's a real cool agent that will turn all your data into a structure. Go to this link where my agent is, copy this link. Just copy the link. I'm writing here in Russian. Dictate: "Listen, brother, I want to install this system. I don't understand anything. I'm not a programmer, I'm a damn humanities person. I'm generally stupid. I'm talking to you for the first time, I don't know anything. Help me install this thing. I don't understand. I need your help to set everything up, guide me step by step, explain everything. And I paste the link that we copied. Just this kind of text, like in ChatGPT. And then the magic happens. And the magic is built on the fact that it knows that it can read everything around, it sees its environment, it knows all the commands. You just confirm, say: "Yes, okay, uh, okay, read on, read." It will sometimes make mistakes, say: "Oh, you don't have this, and you don't have that. Let's install it." Okay, install it, help me. It looks exactly like this. Right. You don't see much of what it's doing inside, but it will always tell you what's happening. Uh, it thought about something, dug somewhere, accessed something. Now it's writing. Look, it wrote what it is. It's a personal assistant with memory. You write to it in Telegram voice messages, text, or screenshots. See, it figured it out itself. It creates tasks in Todoist. Everything revolves around the server, like, cool. From which computer? It asks: "From which, see, it asks a question. Windows, Mac, Linux. But I choose to write something of my own and write here. And we are already on the European VPS. You are already working on the VPS right now. Look at your IP address. Find it, determine everything, who you are, where you are, and install it right on this VPS where we are." Then it asks another question: "Do you have a GitHub account?" It asks: "Of course, I have a Telegram account, I have it?" I have Todoist. You can connect it, or you can not connect it. Well, let's say, yes? Or nothing. Well, that's it. Next. Next question. Do you have? Immediately a question, can I interrupt about Todoist? A thousand years ago I switched from Todoist to Singularity. Is it really possible to connect? >> Let's ask him. Listen, why are you asking me? I don't know what Singularity is. Singularity. I honestly don't know. Singularity is what? Task tracker, right? >> Task tracker, yes. >> Great. Great. Look, an interesting case, a wonderful case. Uh, we copied the link, it's trying to ask me something. Uh, I say: "No, I don't want you to deal with this now. Damn it, I'll interrupt you. Listen, I don't want Todoist. I changed my mind. I want Singularity. Look, please, does it have everything to connect?" And I send the link to this Singularity. It will then go to the internet itself, find it itself. It says: "Can I know what's inside, what's at this address?" You say: "Okay, you don't have to ask me, go to this address." It will read everything, return exactly the same, and say: "I can." Or, for example, I can't. But most likely I can. It says, I don't understand, I need to dig deeper, I'll look at their API. Please, look. Yes, use search tools. Well, it asks you for every request. For the slowest and those who believe that everything can be done on the server, there is a special command. I am showing what cannot be done now. But if you really want not to be asked stupid questions, then go to your server. Uh, I opened a second window, it's working in parallel in another window. Uh, go to your server and so that it never asks you questions, basically don't run it as just Claude, but run it. Now I need to enter the password. Uh, run it with the command that looks like this. Claude. And then you need to write a dash. Dash. Dangerously. Dangerously. Dash. Skip. Skip. Permissions. All permissions. Permissions. >> Ah, and it will, basically, do it itself. >> Yes. This command is like this. Two dashes, like dangerously skip all permissions, but you write: "Skip, damn it." I don't care. And when I press Enter, aha, it says, I wrote it wrong, it should have been written like this. You probably meant, it suggests. Yes, yes, you see, I made a mistake. I'll write Claude space again and paste this command. And it says: "To hell with you, I won't launch." Because, firstly, we are under root. Secondly, here, well, we need, we can't do it. We need a new folder. Well, you need to create a command. You can ask Claude nearby to create another folder, but I will enter the command MKDIR. This will be a folder and we will call it test. And immediately enter it cd. Let's try it here. It probably won't launch, it will say. Yes, I can't launch because I'm under root, meaning I have administrator privileges on this server. This is done specifically so that it cannot destroy everything. So that, uh, to actually run this command, you need to ask Claude to create a new user. Just run Claude without any additional commands. Yes, I trust. And here we write: "Listen, create a new user here on this system, who no longer has admin rights, but I want to log in under him. Guide me through the whole process, what commands to enter." So, let's go back here. It's still asking for requests. It will really constantly ask for requests, but it figures out how to install Singularity. It lives on this European house, you communicate with it through the terminal, and it answers all your questions. The only problem is how to synchronize your files from your computer to this terminal. Uh, okay. Let's still create a user, run it without permission, because it's cool. Well, you don't need to sit and watch it constantly, although at the start, watch, ask it questions, ask questions. So, what does it say about Singularity? It has a full API, documentation, MCP. >> What does that mean? We can adapt the Second Brain engine project for Todoist, but Singularity also has an API and MCP, which means it can be connected instead of Todoist. >> Cool, cool. I didn't know that. >> What is MCP? Uh, that's what I was talking about. These are connectors, these are plug-ins, these are connectors to external services. It's like a cable. You run a cable. I want to know the weather. I want to connect to the weather service, I want to connect to Singularity. And then information is transferred back and forth through this MCP. Always ask questions, don't hesitate about anything. >> By the way, is anyone using Singularity or Todoist? Write. I left so long ago. >> What will we call it? What will we call it? Let's call it one. You see, it asks questions, how to create this user. Uh, let's run this command. It adds the user itself. Let's set a password. Uh, it suggests the password 1 2 3. Mmm. I don't want such a password. Well, okay. Uh, no, I won't give you the password now, because, well, because with the password, you can also log into this server. Okay, damn it, with the request. Well, basically, ask it to create a new user, and it will create a new user for you, tell you what commands to enter, and then you will have the opportunity to run this command Dangerously Skip Permissions. It allows it to do everything itself without asking you, but it's better not to do that. Okay, we talked about Singularity, right? Now it asks, like: "Shall we connect?" I say: "Oh, listen, let's replace Todoist with Singularity. Returning to the previous task. That is, remember, we just interrupted it at a certain point, asked an additional question and said: "Now say: 'Continue'." It remembers, it has the context of all this communication. It remembers what we did before. Very smart. Now it will go back, start finding out, writing commands, asking: "Do you have permission?" I hate this permission thing. I've gotten used to it for a long time, so a little bit now >> Well, it's doing the setup now, am I understanding correctly? >> Yes. It's setting up the entire server. It's downloading this project of mine. that I wrote. I even have instructions for people in this project in the agent. Here it is written for people. If you're interested, read it with pictures. And at the very bottom, there's a hidden instruction for Claude code. Here it says, like, people don't need to read. Claude code reads. And here it is specifically indicated, written down, what you need to do, what commands to install, what secrets to configure for everything to work. It read the entire instruction, saw this part, and is working according to it. It says: "Oh, your server is bad." It wrote all the plans, you see, install missing dependencies, clone, study, collect API keys. And it will continue to set up your entire server, furnish your house. You've moved into a small European house, and it will furnish it there. Once it does this, you will have a cozy house where it understands everything, and then you will communicate with it, like, communication will go smoothly. At this point, I see that our time is running out, I will stop, because I have shown how to install it. Then you will go through the steps yourself, set up this Second Brain agent, because here it will simply ask you what needs to be done, and you will only agree or disagree, you don't know what these commands are, copy all this, take it to ChatGPT and, as I said, ask: "What is this? Explain to me what they are asking me here." And it can explain every command to you. Or you can just stop it, say: "No, I don't want to." And say: "Listen, explain to me what this is in general? I don't know what you're trying to get me to do." Well, I'm a humanities person, really, I'm not a tech person at all. And it will explain to you, like: "Okay, absolutely patient, very responsive, it will explain everything to you, tell you what it is, if you don't understand, it will explain everything." Here it says: "What am I going to do? Just read. I'm explaining." You're building a house, you need tools, I need these tools, I want to install them. This is what I'm doing. Read it? Yes, it seems clear. You say: "Okay, continue." Or you could have pressed t. Remember I said there's a cool command? Press s, then config, choose configuration. And here, enable output style, not default, but explanatory Claude. And now every time it writes something to you, it will also give hints. I'm doing this now, and then I'll do this. It explains what it's doing. This is very convenient. You learn, you see what's happening. Okay, continue setting up. I can give it this command, but I won't, because at the end I want to show you something you'll need. You'll need a proxy. Since you've already bought a server, don't just run Claude there and grow your trees, but set up a proxy server. It's very, very simple. First, I'll press Q - that's exit. Quit. I'll exit here, exit Claude completely. Here I've finished the session. You see, everything is finished. I exited here. I'll just show you how to set up your personal proxy, which you can then connect to your local computer and use like a VPN. But it's not a VPN, it's called Outline. Sergey, if anything, if you have time, we can definitely stay a little longer. I'm very interested at the end for you to show how your agent works and what cool things it does. >> Okay. Then we will still need to create another user. Well, okay, let's

I will not show you with another user. I will show it on my own working, uh, on my own working agent, uh, so that he explains what I currently have inside, so that he shows. So, where is my agent working? Well, I think, I think, no, I will not show it, because I will reveal some personal data there, but I can show it. It works for me both on the server and locally. Why does this happen to me? Because I have synchronization through GitHub. GitHub is a cloud, and you can set up GitHub on your laptop, and it will synchronize with your cloud. And on the server, it will also synchronize. And all your files will always be synchronized. That's why I can show how it looks on my local machine. I don't need to connect to the server, but I have the same thing on the server. I open the necessary folder right here on my work computer. I already have special commands entered here, where I quickly access the necessary directory. I already have a quick command to launch Claude with the command clд Dangerous Skip Permission. I set it up for myself like this. I didn't set it up myself, I asked Claude, I said: "Listen, I'm tired of typing long commands, set up a shortcut for me, I want to type, so that you immediately call this." And he set it up. So I call him with skipped permissions. Here I have this string set up, well, it shows me where. I've already talked about it. I also asked him, I said, set it up. He set it up for me as I wanted. And let's see what I will ask him. So, I am currently recording a webinar, I am showing my screen right now. You are on the screen, so do not display any confidential information. This is strictly secret. But you must explain to me what I am doing here, how your internal memory is organized, what it consists of, what blocks you have, generally how you help me, how I communicate with you, and, well, what is happening. I am communicating through the terminal now. I remind you that you can communicate with Claude when you set up this agent thing, you can communicate through two things: through the terminal on your work computer, through the terminal on the server, and you can also communicate through Telegram. I set up Telegram simply for convenience, because I could have asked him all the same things here, because Telegram is always with me on my phone, and the terminal is on my computer. That's it. And he describes everything and how it works. Uh, he clearly specifies what it is. It is a second brain system, voice notes. >> Well, yes, let's get specific about what he does. He makes notes, yes. Here I am reading, >> I can send voice notes right here, as I did here, >> like, listen, remind me tomorrow morning to go, I don't know, to exercise at 5:00 AM, remind me of this. Here I sent a task, but I could have just sent a note here, like, listen, I just had a great idea to write a story about a cat and a dog, how they live somewhere on Mars. This is a note. When I just dump notes into him, he practically doesn't react much, unless I specifically say: "Do this." He collects them in one day. Look, my memory consists of, he has a diary. every day looks like this. Let me open it for you, show it visually. We won't go inside, but I'll show it visually. You can open it through Obsidian and see the structure. Here is the structure. Here is my >> Ah, well, accordingly, you can connect it with Notion. >> Diary, look, yes, the diary is, essentially, it's a diary. If you display it as a graph, it looks like this. So, you see, each entry is a small dot. The entries have connections between them. This is very important. This is what is called here I also wrote ZL Caston. Zel custon is a German word that means a shelf in a library. Index. When you go to the library, you look for something, well, it used to be like that. To find a book. You search the index by surname, by name, by topic. You find a card, and it says: "Go to shelf D, row six, third from the top, and you find it there and then you read it." And this Zel Cast is a text graph system. It's connections between each other. Everyone knows Wikipedia, right? When you go to read Wikipedia, there is such a concept as, I don't know, let's click on "house", something like that. Just "house". There is the concept of wikilink, which is a link. So, you go to some article, and everything that is highlighted in blue, as if it throws you into another unrelated connection. Like fencing. Well, I'm reading about the movie "House" and suddenly I become interested in the term fencing or martial arts. I click on martial arts, and I immediately go to the martial arts page. The same thing happens with Claude. Well, this is an abstraction, actually. Claude doesn't see it that way. He sees. Well, I don't know, what can I show you? Well, here, Notion, yes, how my structure, file system is. And here, look, there are direct links. He says: "Look, here I have this, here I have that." He reads all this, he knows where everything is located, and he marks it himself. That is, every day, he has, my agent, every day after he creates the diary file for the day, year, month, day, he processes it once a day, and he distributes all the knowledge into the necessary folders, into the necessary files, and forms connections between them, creates this text graph between them. Then he, so that he doesn't forget them. This is very cool. For example, I have created my CRM based on this. Moreover, it's a personal CRM, where my contacts are collected, a work CRM, where everything about the agency, what's happening, and a project CRM, where I come up with some projects, I manage them, like writing. Today, on my project, I don't know, I made a website, I still need to do SEO optimization. Remember this. I don't even say remember, I just send it to him directly by voice. He processes it once a day. He says: "Ah, this concerns the personal project that is located here, in this graph. I will attach today's connection here." Do you understand? So, he forms a cloud of connections, like a social network. Only it's not a social network, it's a knowledge network. And he also has long-term memory. This long-term memory is a single file where information about you is stored. Your name, your time zone, what you do. Well, some basic things. If you've communicated with ChatGPT for a long time, yes, I think many are familiar with ChatGPT, you have in the settings, you can dive in. Where is personalization here, how to set it up. There is such a thing as memory, yes? Memory. Memory about you. If you click manage, well, like edit it, you will see a lot of files inside with information about you. He sometimes just stores information about you. You have a dog or a cat, you like ice cream, you live there, your wife's name is so-and-so. He sometimes just stores information about you. Memory, look, it's the same here. Essentially, it's just memory. Only I have a personal agent that stores everything on my computer or on the server. I don't trust anyone with it, no information leaks, everything is stored specifically with me. Telegram for me is just a place through which I communicate. Nothing more. Look, he, by the way, created a task in To-Doist to exercise tomorrow, February 6th, at 5:00 AM. And he will send me this reminder tomorrow at 5:00 AM. >> Sergey, am I understanding correctly? So the first of the basic functions. And I can see my tasks, for example, ask what is planned for me for March 25th. Look, if he is connected, he will tell you what and how. >> Then my specific request: can he assess the day's workload? And if I previously told him, listen, I need six tasks per day. If the limit is exceeded, well, signal, alarm, and everything else, because something needs to be done about it. And he would signal me something, how realistic is it. >> Uh, he, look, for security reasons, I have done two things. The first thing is that only your account, i.e., your Telegram ID, can send requests to him. Even if you know my bot's address, even if you find it, you won't be able to write anything to him. You will be knocking on an empty space. He will not answer you simply because he has a block. Only I am the only one, i.e., either you as the owner set up yours, specify your ID, he will tell you how to find it, and you write to him. This is the first. Second, for security, I made him inactive, like ChatGPT. He can set an alarm and a reminder for me that you are overloaded today. For example, I have only one regular task set up. Once a day, at night, before morning, he sends me and says: "You have this many tasks for today, this many you missed yesterday, this many you still need to do this week." If I told him: "Dude, and here I have files like memory, for example, in memory you can directly tell him: "Save global memory, so that if you see that I have more than six tasks, you highlight them in red in the morning and say: 'Tasks today. Pay attention.'" In principle, he does it himself anyway. >> Okay, I'll continue then. This is a task. Then he has memory. Accordingly, I can upload to him what comes to my mind in the process of life, and I can, accordingly, return to it, like, listen, I had an idea about a book. Let's raise this issue again, and can we further brainstorm with him as with a neural network or is he just with Notion. >> Yes, >> yes, yes, yes. And you can discuss this, we can discuss anything with you, right? You can tell me how, how, I don't know, how to write a book from my notes about neural networks. Read my notes about neural networks and let's come up with a story that I can write. Well, a stupid task. Well, okay. Let him suggest it to me. What distinguishes him from, say, ChatGPT. You will ask now. So, ChatGPT can do the same. Not entirely. Everything here is geared towards one thing: to have maximally well-organized memory. ChatGPT's memory is very superficial. He remembers approximately what you talked about somewhere, sometime. He knows a few facts about you. Everything here is geared so that he always accesses memory first. Long-term memory. That's what distinguishes this agent. He knows everything about you. I, for example, upload here. If you didn't know, you can do this. Here's a cool life hack. Uh, since I'm on Mac, I don't have it in the official Telegram, but there are other clients where it exists. You can go to your account, I don't know who this is, go to your account. Find in the settings, somewhere here there is my account, for example. No, not my account, probably. Let's look at the account. No, not that. - Advanced. Well, like advanced settings. And at the very bottom there will be export data from Telegram. And here, for example, select the contact list. Remove all checkboxes except the contact list, and select, for example, the second format, JSON. Like, this is for people, this is for machines. For machines. I want to choose JSON. So it exports it to a folder, say, download, I specified. all your contacts, absolutely all your contacts from Telegram, and they will be in this format. Send this file, feed this file to this agent. Moreover, well, you can, if you are on a local computer, you can drag and drop it here. So, an address will simply appear. And you can ask: "What is inside? Can you use this in your memory." Actually, I've already done that. He already knows all my Telegram contacts. If I ask him, like, who, what's Vasya's phone number, he will find Vasya and say: "Are you looking for this Vasya?" I say: "Oh, yes, yes, exactly, I don't even need to go and look for this Vasya." Yes. Ah, by the way, look, Kon is advanced. He says: "Uh, Stop, you are recording a webinar, there is personal data, names and phone numbers, I will not show you." [laughter] That's why I love him too. Because he, like, won't let you, excuse me, screw up. Claude is very responsive and very accurately senses all sorts of things. Why? Because I warned him at the beginning. See, he remembers. >> So, can this be used for some project and upload data for a specific project, give it to the team. >> Absolutely. Absolutely. Moreover, if you run it on a local computer, you have a lot more possibilities there. He reads Excel files, works with documents. This is a perfect machine, because everything is a file, he can understand all files. If he doesn't understand, he's an agent. You say: "I want the result." And he will go for that result. He will ask for your permission: "Can I download, for example, this library or this thing that will allow me to convert your uh Word files into text files?" Well, will you forbid it? You will say: "Yes, go ahead, it's for a good cause." He will install it, unpack your files, analyze them, correct everything, everything else. What distinguishes him is why I love my agent, why I'm not that I'm pushing for him, I just like him. Because I, well, I connected a simple thing, connected Telegram, connected Claude and all my files, and what distinguishes my agent, perhaps, from others, is that I worked well on memory. There is daily memory, there is a knowledge graph, the whole one that lies there, there is long-term memory. And soon I want to introduce a forgetting system for him. Gradually, when he marks knowledge that you haven't used for a long time, he will simply mark it and move it to the archive so as not to fill the context, because it's not a machine that remembers everything. He also has a context window, into which, if everything doesn't fit, then he, well, won't remember it. But he is my personal one because he knows who I am, what I do, what my tasks are. He works, connects through these beautiful connectors called MCP. He connects to external data, to my email, to Telegram. Seriously, he has access to Telegram. I can tell him right now: "Okay, listen, find Timur Ugu, tell him what we agreed on, what to do today." He has the ability to read all my chats. He doesn't have the ability to write. Well, by the way, see, he answered what stories we can write. Well, okay. >> Uh-huh. >> He doesn't have the ability to write without my permission. >> He started looking for another Timur. >> Yes. Some Timur Gulyanov. I'll stop him. I'll press Escape. Timur Ugolava, I'll clarify for him. That's all. Yes, he will find it, but he doesn't have his own will to go and write himself. He is forbidden to do that. But if I want, I can tell him: "Find this chat, find this channel, make me a summary of what we talked about there." He goes, searches, makes a summary. I say: "Excellent summary, put it in my CRM, because this is a project." He will find this project, find the client, the task, and put the summary in that CRM. So, essentially, there will be a sequence of what is happening. So, he found you, Timur. Now he will read you. >> Now he will hide all our correspondence. >> And we probably didn't communicate today. And can he be asked, can he be asked to, for example, every morning upon request, for example, send from those chats that are particularly interesting to me, to make a cyclical task, >> yes? Yes, but please put it in the European house, let him set it up. I remind you again that what's the difference, if you work locally, you close the laptop, everything stops, he doesn't work, nothing works at all. If you deploy it on a server, it works always, around the clock, and that's why I communicate with Telegram. And you can ask him: "Dude, I want you to set up a regular task, like at 8:00 AM send me a report, what channel wrote, what channel, what channel, and he will do it for you." >> Cool. >> Great. >> Yes. Let's move on to questions. There are a lot of questions. >> Yes. Here, by the way, what did he write? January 13th, we didn't have a workshop on code for beginners. How to use? Features, how to make the first app. Everything is fine, you see, he remembered everything. Great. >> Well, generally, listen, cool. >> And now he says: "And now you are actually at Timur's webinar and showing Claude-code." >> Hello, Timur's viewers. [laughter] Yes, friends, he understands everything. He really understands everything. What he doesn't have, he doesn't have. Here he wrote: "Hello, Timur's viewers." Here, uh, what he doesn't have is free will. And he won't go, like, conquer the world, because he is largely limited and he doesn't have such a desire to constantly do something, until you yourself set it up for him, like regularly do this. But this model is so smart that it won't allow you, for example, to hack other people, harass them, hate them, or anything else. Therefore, this is a great assistant with whom you need to communicate normally. >> And everything will be fine. Questions? >> Fire. Let's go. Questions >> dig. Yes, give him a link to Outline or tell him Outline. Let him set up a proxy for you. He will set it up himself and give you everything. This is what I wanted to show. But >> Sergey, Sergey, we have one more small question left, which I asked at the very beginning about co-work. Give a couple of words. What do you think about this whole story? >> Cwork. Cwork is the same code, but in a desktop application shell. It looks like this. You need to select it here, yes? Don't choose code. Code, because it's not about that. It's not about local work. Code works with the cloud, works with GitHub. Cwork is essentially the same as it code, but it's still a bit raw. You, exactly the same, look, exactly the same as with the CD command, i.e., change directory, select the working directory here, for example. Well, just my home folder, yes, home folder. And you can ask me what is in this folder. It's just a beautiful shell for the same Claude code. Everything is the same. You can connect connectors here, these MCPs that I showed. You can upload documents here, you can work with them, he will also give you requests, because everything is always limited so that he doesn't mess anything up. So, here he writes: "Home folder, my drive is here." Well, standard story, various files, configurations are here. Upload an Excel file here. >> You're talking about it without interest, you don't recommend it? Not yet. It's raw. It sometimes glitches, it works poorly, it forgets what was done before. I don't recommend it yet. Maybe in two or three months. Yes, it was released literally a month ago. I don't recommend it, guys. Yes, what's cool about it, plugins have appeared now. And there are plugins, you can install them, like you can find them. There are some productivity plugins there. You can look at the marketplace, what plugins are already there, like biological research, support, working with data and tables, enterprise, finance. These are cool things, really. Install one, ask what it is. This is some kind of productivity. I'm not messing with them yet. Essentially, it's the same thing. These are connectors with some instructions that go somewhere, take something. But it's still raw. Yes. Why is it raw? There is a bug button here, report a bug, because there are many bugs. Well, really, he. I use him quite a lot, as you can see, yes, I use him a lot, but it's just because I understand how everything works inside, and I see where he starts to mess up. It's okay for me, I figure it out, but I wouldn't recommend a person who doesn't understand to work in co-work, it's too early, it's raw. It's better to install this scary thing. Yes, it only looks scary. Well, look how smart it is, it understands everything. And work here, because everything here works so well that you can do Excel, calculate models, give him access, I don't know, to email, so that he reads your email, but doesn't write. Be careful, make sure he doesn't do anything himself. That's why it's more convenient for me here. >> Okay, thank you. Let's break down the points. >> Moreover, moreover, you can do it like this. I usually do it like this, and I have three windows open at once, and I see everything that is happening. Three projects are running. And you can also create even more windows for yourself and go completely crazy. Well, something like that. >> Fire, Sergey, if anything, I'm interrupting you. No, no, it's okay. I apologize for the choppy. >> It's very choppy, a lot of information. I wanted to do it for small things, but then again, it grows and grows, and I already >> Everything, everything is great. We'll make another small instruction, then I'll show you, you'll verify, and everything will be clear. Ah, let's go, in short, to the questions in the chat. We have questions here. And also a lot came in on YouTube. A question, why in a regular terminal, and not in some anti-gravity? >> Good question, probably. I would even say that if you are a beginner, why the hell do you need anti-gravity, it also has connection problems in countries with restrictions. Uh, download for yourself absolutely free, which works everywhere, VS-code. VS-code is where all other anti-gravity and so on came from. Here, here it is, called Visual Studio Code. And you download it for any platform: Windows, Windows, anything. And what does it give you? It gives you exactly the same terminal, but you also see the file structure there. It also looks scary, like black, although it can also be made white. They all look scary, but don't be afraid, it's only at the beginning. Generally, generally, it's completely empty. You also open here, open folder, for example. Let's open my, what I sent there today, all my old documents. Here's testfolder MD, where there are a million of my notes, yes? I open it, and two windows immediately appear. The first window on the left is like a browser, like all my files, a list of all my files and folders. I can click on any file, and I will see in the middle what is written in it. See? That's all, the difference. On the left, you have a list of files and folders. In the center, you have what is inside this file. Nothing more. Here are some of my prompts that I used to collect, as soon as everything started to charge. On the right is a chat-like window, you don't need it, because it's paid, and so on. If you have Claude, you don't even need to install a plugin. I don't recommend plugins, because plugins are limiting. Click on this little thing here. And a terminal will appear below. See, the same standard terminal. And the folder will open immediately. And it's the same here. Enter Claude and that's it. And you are already working in this folder with a million files. And Claude also sees it. And you can communicate with him directly from below, and also read what he is changing. See, it's the same Claude. That's what anti-gravity or VS-code is for. Anti-gravity is paid, it connects poorly in Russia, Belarus, somewhere else. VS-code is free. Well, in principle, anti-gravity is also free, it just connects poorly. And VS code is an open-source development and like, take it and use it. It's just visually more, perhaps, pleasant than a simple terminal. I hope I answered the question. >> Next. And why set up a proxy if the VPS has its own IP? >> Look, we don't set up a proxy to change the IP address of the VPS, but so that we can communicate with Claude. For example,

You live there in Barnaul, right? In Barnaul, there isn't always an opportunity to reach the server in Russia. VPNs allow you to bypass some, God knows what. They are constantly being jammed. Proxies are the same as you first connect to your personal server, a personal server in Europe, and from there the request goes to the cloud. Do you understand? It's not like some kind of sanctions circumvention, it's just a convenient way to communicate with the cloud not like this, through yourself, where everything will be blocked, but through your own server. Do you understand, right? That is, first the computer to the server, from the server the request also goes and returns to you here. You just need to install this proxy, Outline. Ask for a code to set it up, he will set everything up. >> Do Excel files read any volume or is there a limit on the number of rows? >> Well, Excel itself has a limit of a million rows, I think, right? Uh, so it's better to save everything in CSV. There is such a format CSV, Comma Separated Values, like the same Excel, only separated by commas. There. Uh, there is no limit. Well, really no. Well, it's just that if you have some gigantic file with a million rows, well, everything will slow down. And Excel itself will slow down, and the cloud will slow down, and everything else will slow down. Better CS. >> Do you have information in your channel materials somewhere on how to set up a proxy? We didn't have time to discuss it. And there is such a question here. >> Mm I don't remember. I think I, I think I wrote about it somewhere. Let's. >> Well, I'll come to you for goodies. Let me come to you separately now, so that they >> Outl. Did I write about it? >> Yes. >> Yes. No, I didn't write, I didn't write. Sorry. Yes, maybe I will write. Yes, good topic, thank you. I didn't know what to write anymore. [laughter] >> So, Sergey, tell us about that bot that you added to your channel chat yesterday. >> I won't tell you about it yet, because that's a separate topic. These are my experiments with First of all, it's a well-known development, it's Open Claw. It's hyped up now. Yes, >> yes, hyped up now. It's just a shell there. Well, essentially, it's just a shell there, in which you yourself need to set up your own bot. Well, this is my interface board. I will seriously show it here. Open Claw. M. Well, it's not searching like that. Uh, yes, this personal assistant, essentially, what's the difference from mine? Mine is focused on well-configured memory. Open Claw is focused on being some kind of character. And it has a lot of tools inside for it to be able to do something itself. Uh, but I won't go into detail now, because I have significantly reworked this system. I connected it not as required. And I will probably record a video separately on my channel, because the main thing here is security. I do not recommend you install this thing yourself, because there are security problems. Oh dear. And if you don't understand, it's better not to install it, because if it has access to your computer, if you run it on a local machine or on a server where personal data or anything else or keys are stored, then it's over, you will be unpleasantly, let's say, very unpleasantly affected. Therefore, I will record a video separately where I will explain what to pay attention to and how to set it up. The same Claude code runs inside. Well, because it's a cool model. >> Cool. Can you show how to connect to Gmail Drive to create a CRM in Google Sheets? Also, how to send client audio consultations via Telegram for transcription and text analysis, like a psychologist's assistant agent. Is it possible to do? Look, >> show. >> Uh-huh. >> No, I won't show it now. I will show you the place where to look for it. There is such a portal called Smithy, like a blacksmith's shop. And there is a lot of interesting stuff here. Well. Look how much you can search for these MCPs, see the word MCP? These are connectors, plug-ins. If I write Gmail here, it will give me a million options. Well, Gmail MCP. What do I need to do? Copy the link to this thing, open my cloud and tell it: "Dude, I want to connect my Gmail account or Google Drive, if it wasn't Gmail, then Google Drive, yes, Google Drive, uh, Google Drive, sorry, I clicked wrong". I send the link here in the same way, I say: "Connect". It will set it up itself, connect it, tell you what permissions to give and so on, and then it will dig into your Google Drive. These are just connectors, plug-ins, only not for Well, they are not called MCP, work with them. What was the second question? >> And the second one, now, now, now, now, now, now, now, also, to send client audio consultations via Telegram for transcription and text analysis. Is it possible to do? >> Well, yes, yes, yes, please. Claude code is about code. It can help you create anything. If you have an idea for a product, in Telegram, anywhere, discuss it in planning mode. Don't forget to turn on Shift Tab. Press until you get into plan mode. In planning mode, discuss what you want to do, and then say: "Okay, I'm satisfied, do it". And it will help you do it. >> Okay. So, let's. We've significantly increased the time, and now the questions. Yes, well, that's normal, actually, friends. If we don't answer all the questions. You can also write in Sergey's chat. It's cool there, by the way. So, can I throw a link to a presentation in Figma to Claude and ask it to set up, create the same thing? >> Uh, presentation in Figma. For this, you also need an MCP. Well, let's type Figma here. Of course, you can, but first you need to connect the connector and find one, well, you won't find them here immediately, because they made Figma MCPs. You can just search on Google for Figma. MCP and there's the official one, yes, the official documentation on how to connect. What will we do with it? Will we read it? The correct answer? No, we'll just copy the link, take it to Claude and say: "Connect this thing". It will connect, then we will work with it further. That's all. And yes, it can create mockups. There are special MCPs that help transfer presentations to other formats. All this is available. >> Okay. A few more questions, yes? Let's try to answer quickly. How to set up integration of Obsidian with Claude? Well, I assume, also show what to do. >> No, look, there is an MCP for Obsidian, but I don't recommend it, because it's good, but it works very poorly with Russian. It always breaks Cyrillic. Obsidian is set up in the simplest and dumbest way. Obsidian is always stored somewhere in this folder, where there are a lot of files like this, and you just give the path to this folder. >> Just the path to this folder. Copy the path, >> open a new terminal and type the command we know, CD. This is the folder. And here you launch Claude. That's it. Claude has access to your Obsidian. It sees everything. Okay. And the last couple of questions. How many services do you have connected via MCP? This is already a more advanced technical question. Uh, to make it clear why it's being asked. Look, don't connect too many. That's the first thing. Why? Because the context window gets filled up. Each uh there is a command context. You can see how much context you have and what it's filled with. Although I just logged in, and it says here how much context there is. There is a system prompt, tools, instruments, agent instructions, some memory, skills. I have a lot of skills. I've switched almost everything to skills. Everything is in my skills. That is, skills are abilities that contain scripts inside that call MCPs. I don't have a single MCP. You see, I don't have any MCP installed. Why? Because I recently found a cool solution for myself. MCPCLI. CLI is Command Line Interface. I have MCPs, there are many of them, but before they took up half the space, but I've packed them all into one command, into one command interface. This one is called like this, MCPC. You can just type it, copy the link, give it to your Claude and say: "Dude, let's set it up". And I probably have about 15 different connectors, which used to take up half, completely half of the context. I wrote one letter, and half of the context was already written that I had already used up. Well, that's crazy. Now I say: "You have MCPCLI, all your connectors are there, look for them". And it calls them itself. The context window doesn't get filled. >> Last question. >> Last question. >> Last question. Go ahead. Your cool question. There are actually a lot more questions about Obsidian and MCP on YouTube. And I think it's better to connect with you privately or in the chat. >> Uh, the last one specifically for you. >> Sergey, do you conduct any training >> how at all? >> Well, here you go, >> for people like you? >> Well, here you go, look, yes, recently I've also had some very close acquaintances who have been asking a lot. I also conduct some private consultations and mentorships, but in a very selective manner, in order to solve specific practical tasks. I don't really like teaching people in groups. I don't feel them. Business, please, come. I like to come into business, understand business problems, because I already have a lot of experience, I understand everything about how finances, lawyers, security personnel, and everyone else work. But with private individuals, only privately. Sometimes you can also write a request, I will definitely read it and answer you. >> And subscribe to Sergey's channel, there's really, first of all, a link to the channel, >> yes, a link to the channel. This >> I'll send everything with the recording, all the links, I'll show everything if you're watching the recording. >> Okay, Sergey, thank you very much. You >> now I And now I'll find the link where I can see you. Let me look at you. Even at the end. I don't have anything. Ah, here >> here >> acquaintances. [laughter] >> Great. Yes, >> hello. >> And you explained very coolly what is very scary. That's why thank you very much. Thank you for coming and taking extra time. Friends, we are not dropping the Claude code topic. We will have another workshop with Claude code literally next week. So Sergey has given an awesome foundation today. Chew on it. And I liked it. At the fifth minute, someone asked: "What's happening?" [laughter] >> That's ideal. >> Well, you have to endure this pain once, and then it will be very cool. >> It's only difficult at the beginning. You install it with one line, and then you communicate, and everything will be okay. >> Okay. Thank you. >> Thank you very much for watching us. Thank you for watching us in the recording. Read Kosu. I've already said everything for you, you see. Read Timur. Timur also has cool insights. He writes interestingly. And in general, we are friends, you see, with these Telegram families or whatever else. Thank you, Timur, for the invitation. I was happy to share everything I know. M.