2- How Does Claude Code Work- Installation And Quickstart With IDE

Hello guys, so we are going to continue the discussion with respect to the Cloud ecosystem. And we are going to start with the first product that is nothing but Cloud Code, okay? Now let me tell you about Cloud Code. This is becoming a very popular product nowadays because many companies even in the enterprise level they are basically using the enterprise version of Cloud Code. Specifically to develop the product to build the products from scratch, to maintain the products and even for the shipment purpose, right? So as you see in the future interviews also now, you know, in companies they are definitely going to ask you whether you know how to work with the help of Cloud Code because definitely the productivity level is quite high. Now when we talk about Cloud Code, many people will just think that are we just doing wipe coding with the help of Cloud Code? It is not that guys. There are a lot many things that we really need to learn from Cloud Code, okay? So again this will be a series of videos. This is just the first part of Cloud Code and in this I have designed an agenda. So first of all, since we are going to start completely from basics first of all we'll try to understand how Cloud Code works, okay? Then second we are going to see the installation process and third we are going to go ahead quick start how you can integrate with your ID, your favorite ID whether it is a Visual Studio Code or Cursor or any other IDs as such. And then how you can go ahead and start the session, okay? So everything I will be discussing about step by step we'll go ahead, okay? So the first thing what we need to understand is that how Cloud Code actually works, okay? And this is something really important for you all to understand. Uh So here I've just drawn some diagrams. I'll make you understand what exactly it is. But whenever we talk about Cloud Code, right? What exactly it is? It is nothing but it is an agentic coding tool. Okay? Agentic coding tool. Now this term is super important because whenever we talk about agentic coding tool, right? This will be interacting with your project. That basically means it will be able to read your code base, edit the files, it will be able to run the command, it will be able to integrate with your development tools, whatever development tools you are specifically using, okay? So you need to understand that, okay, fine. We are trying to use this as an form of agentic coding tool. Whichever projects you have given access to, you know, you will be able to give some instructions and Cloud Code with the help of your approvals will be doing some kind of task. It will be editing some of the files, it will be running some of the commands, you know, it will be integrating with your development tool and many more things, right? And the best part about Cloud Code is that this is available everywhere, right? When I say everywhere that basically means it is available in your ID. terminals, right? You can directly use it directly from the terminal, you can also use it from the desktop app, right? Now when I say desktop app that basically means Cloud Desktop, right? And even you can actually directly use it from browser. So this all I will show you, right? As we go ahead in the sessions we will try to understand how we can actually use in Obviously I've shown you with respect to desktop, you know, I'll also give you a quick idea about desktop because once you have downloaded, right? In the previous session I've already told you to download this. So here it is. This is the Cloud Desktop, you know? Directly you can select a folder and you can give the instruction automatically the coding will also happen over here, right? So you can use this also. But as a developer what I would suggest you try to use ID and terminals. That way you will be able to see many more things in terms of how the coding is basically happening, you know, you'll be able to quickly review things, update the unit test cases and many more things itself, right? So — [clears throat] — Now let's go ahead and understand how does Cloud Code actually works, okay? As I said, Cloud Code is an agentic coding tool, okay? So it actually uses an agentic loop, okay? So here let me go ahead and write it down. It actually uses a agentic loop. Now what does this agentic loop basically mean? Let's say you go ahead and give a prompt over here. Let's say the prompt is, "Hey, go ahead and create a mobile app or a UI design for doing this, for tracking your calories, whatever information you have. " That we specify in the form of a prompt, okay? Then what this agentic loop module will do, it has three main parts, okay? The first part is something called as gather context. Okay, so first of all it will try to gather the context from the project base itself, whatever context it has, okay? Once it gather the context, whatever information it has, it will pass to the next step and here it will go ahead and take some kind of action. Now when I say

action, it is basically going to do the task based on the prompt that is basically given, okay? Now once the task is basically completed, then we go to the next step wherein we verify the results. Okay? Now this verification results again here human can also do it, even our Cloud Code will also be doing it, okay? So all these three right? And these three will be continuously happening in a loop unless and until we get the satisfied result. Okay, so let me again go ahead and create this loop. So this will continuously happening in a loop, okay? So after the verifying result is basically done, right? If the results are not that good, then again it is going to go ahead and gather the context again, it is going to take the action and again it is going to create the output, right? And once this gets satisfied, it will go to the done stage. That basically means the task is basically completed. As a human being since we are giving instruction to the Cloud Code to actually work, you can interrupt over here, right? here as a human approval, you can add context over here. And you can do multiple task. You can even change some of the instruction, you can modify the prompt if you want, you can add more information related to the prompt and you can basically do that, okay? So this is how the Cloud Code actually works. Now see, this is important, right? Let's say this is your project. Whenever you start Cloud Code in your ID or anywhere, like let's say in your terminal and all, right? Cloud Code. Now this Cloud Code will be having the access of this entire project, right? Now project is basically having the access of some of the context related to this particular project. And there's some more information that you can save it in the form of context. There we have something called as cloud. md file, right? This I will talk about. You can also go ahead and create this. This will basically have some information about the project. And you know, this will be very helpful. Even understand, let's say that you are a new team member who is joining a team. Let's say you are joining a new team, okay? Now you really want to have a KT. So what the team will do, they'll give the entire project repo. Now just imagine you have the Cloud installed in that particular system, like let's say in your system. Now what you can do is that you can just say to the Cloud Code, "Hey, please explain me about the project. " Okay, project and it will be able to give you the entire information about the project itself, what it is. You know, probably tomorrow let's say tomorrow they give you a new use case to work on, you can quickly ask Cloud to probably modify some of the use cases and tell or provide more information how it can be implemented, right? So this is how it works, right? And this agentic loop plays amazing role with respect to this. And the loop actually adapts to what you ask. A question about your code base might be might only need context gathering, okay? A bug fix cycle through all the various stages, you know, repeatedly. Like it can probably run this agentic loop unless and until the bug is not fixed. So let's say when this agentic loop is basically executing, there may be a bug that may have got, right? It will also try to execute, it will try to execute the script in the sandbox and you get some kind of bug, right? And based on this agentic loop it will again recycle or it will probably re-iterate or run this entire three modules unless and until that bug is not fixed, right? So this is how it actually works, okay? And you are also the part of this loop you should need to understand. You can interrupt at any point, you can add context at any point. You know, let's say that you are not satisfied with any specific model, you can change the model also, okay? So everything is actually there. Not only model, you can also change various tools that can be used right over here. If you want some more additional context from web search, you can add that particular tool also. So in short, Cloud Code serves as an agentic harness around Cloud. It provides a tool, context management, execution environment that actually turns a large language model into a capable coding agent. At the end of the day you actually get a very good coding agent. Okay? So this is how the Cloud Code actually works, okay? Now let's continue this particular process and let's see that how we can go ahead and do the installation, okay? The installation process is very simple. Over here you can see that first of all, if you want to directly download it in the desktop or VS Code or anywhere, you can directly click on this, okay? But the first step of the you know, Cloud Code is basically installing it in your system, right? And when I talk about pricing, see there are three different plans, Pro, Max. As I already told you in the previous session that you can take a $20 plan every month and you can start using

that. Through that you'll be able to access the Claude code and all. Okay? So, first thing first, what you really need to do is that I will just go ahead and open the documentation. Okay. So, let me click the documentation over here. Now, with respect to the documentation, you have all the options. Let's say if you are a Mac OS, you have Mac OS or Linux or WSL, you can install from this particular command. If you have Windows PowerShell, you can use this particular command. So, what I will do since I have a Windows, I can probably go ahead and do this. If you have command prompt also, like let's say you have Windows CMD, you can use this particular command, like a curl. So, what you need to do is that just go ahead and open CMD and paste this particular command, right? So, once you do this, the installation will start taking place. Okay? Since I have already done the installation, then it looks absolutely fine that it is you know, I don't need to again reinstall it, but I've already done the installation. But, what you can do is that you can directly go ahead and search for the documentation and over here, based on your laptop or desktop or whatever things you are actually using, you can use that specific command. So, the installation is very simple, okay? And then you can go ahead and do the installation. So, I have already done it, so I'm not actually going to do it. Okay? Now, what I'm actually going to do, I'll just close my uh Cursor ID. I'm actually using the Cursor ID. You can use VS Code also, but I'll just show you. Let's say that you're working in some of the projects beforehand, right? So, what you'll do, I will provide my Cursor with the access of that particular project, okay? So, I will try to open that project with the help of Cursor ID. You can also use VS Code. It is up to you, whatever uh ID you want to specifically use, okay? So, I will just go ahead and this is my Cursor, okay? Quickly, it is basically opening it over here. And let's say I go ahead over here. I will open the folder which I really want to work. Okay? Uh so, let's say I want to work with my project, some of the projects that I've been working. There are many projects which I'm actually creating, uh which is actually just for my sake, you know, I keep on seeing multiple things and all. So, let's say that I've selected this particular folder, okay? So, this is what I'm actually going to use. Let's say I'll close this, okay? And then I will also close the terminal. Okay? So, let me close it. Now, I have opened the terminal over here. Just to check whether everything is working absolutely fine or not, what you can actually do, I'll just clear the screen, okay? I will just go ahead and probably see after the installation, I will go ahead and see whether my Claude command is basically using. You can also directly open it from the command prompt also, right? Over here only you can just write Claude, okay? And you can check. But, my working as I said that I want to give the Claude access to my project. So, let's say this is one of the project that I had actually done related to LangChain. And this is just like my updated LangChain with respect to the recent version, all the notebook files that I've actually recently created, okay? So, I will just go ahead and write if the installation is done completely, I'll just go ahead and type Claude. Once I type Claude, you can see that it is giving me all the specific information. Okay? So, it is saying me that choose the text style that looks best on your terminal. I'll just go ahead and choose dark mode because I like dark mode, okay? I don't like light mode or color blind friendly, this all things. I can adjust with dark mode. So, I'll just go ahead and select this. Now, it is telling me select login method. Now, you need to have either of this Pro Max, Team or Enterprise. I since I use this Claude code in my company, so I have the Max version. So, what I will do, I will just go ahead and click on first and it will go ahead and authorize or do the authentication from the website. Okay? So, I've authorized I've given the authorization, okay? Build something great. You're all set for Claude code. You can now close this window. Now, I will just go ahead and close this. Now, you will be able to see that logged in as krish naik06@gmail. com. Please enter to continue. So, here I will click on and I'll press on enter, okay? Now, here you can see that there some security notes which says that Claude can make mistakes. You should always review Claude responses, especially when running code. Due to prompt injection risk, only use it with code you trust, okay? This is also there, so I'll go ahead and create this. Then for the optimal coding experience, enable the recommended settings for your terminal. Shift plus enter for new lines. Yes, use recommended settings. I'll just press enter, okay? And this is how it has basically started, okay? Now, you know that since I have opened my project over here and you can see that it is basically having the access of the specific project. Okay? Now, here is what you provide prompt. So, the first step that I told you, right? Over here. When we talk about this is the prompt, right? And this prompt you can basically provide in the specific terminal now. Okay? And here you can see tips for getting started. Run {slash} init to create a Claude. md file with

instructions, something is there. {slash} model picker now list models and all. So, let's see some of the commands that is given. So, first of all, I'll see, okay, which models I'm actually using. So, I'll go ahead and see this. Now, here you can see that there are multiple models that is provided by Claude. One is Sonnet, then you have Sonnet with 1 million context. Then you have Haiku, right? The default recommended is Opus 4. 7 with 1 million context. This is most capable for complex work. When we talk about Sonnet, it is 4. 6, best for everyday task. Let's say you are doing some kind of task um which is uh like a normal generic task, you can basically go ahead and use Sonnet. Then you also have Sonnet with 1 million token context, right? So, here you can context basically means that much uh conversation history it can basically store. So, here you have Sonnet 4. 6 with 1 million context, builds as extra usage, so and so. And there is also something called as Haiku. Haiku uh 4. 5 is fastest for quick answers. For simple things and all, you can basically use this, okay? Use {slash} fast to turn on fast mode. So, by default, I will be using Opus 4. It is more capable for any complex work, okay? So, I'll press enter, okay? This is basically my model. Now, understand one very thing one important thing, okay? In the first stage, you know, whenever you see some kind of um some kind of projects, you know, it is very much important that you go ahead and create this Claude. md file, okay? Now, what is this Claude. md file, I'll tell you, okay? So, let's say — [clears throat] — I will just go ahead and ask a question to the Claude code. I'll say that, "Hey, uh please tell me what this project is all about, okay? " So, this is my question. Now, when I'm sending this question, what Claude code is going to do it? It is going to analyze this entire project repository. check each and every file to find out what this project is all about and it is going to give me a summary, okay? is all about and provide me a summary. Okay, I'll just ask a question. Okay? Now, once I'm asking this specific question, what as I said, Claude code is going to analyze all this particular project. See, it is basically listing all the directories, readme file, md file, everything. And it is going to gain that specific context, okay? Now, see over here you can see bash command. It is also asking me to execute some of the bash command to see some of the information over here because there are files inside this particular folder. So, I'll just go ahead and press one, okay? So, here you can see it is going to go ahead and test, check every file, everything. If there is an error, it is going to reiterate. So, it is basically running an agentic loop over here now. This entire agentic loop is basically running based on the prompt that I've actually given, okay? So, here you can see, right? So, here I have got the entire information. See, a hands-on tutorial report demonstration, the new LangChain 1. 1 or APIs notably create agent and the middle system that replaces older agent patterns. These are all the files. You can see intro to this, model provide integration and all, okay? Now, I'll say, "Can you create me a uh Claude. md file? " Claude. md file. Now, see why I'm creating this Claude. md file. See, in the previous question when I said, "Hey, tell me about this particular project," then Claude what it did, it basically went through all the specific files. So, it had no context about the project, right? No information about the project. And it started scanning each and every files and it probably, you know, generated all the information and it took some amount of time. Now, instead what you can do is that if you see on the top, right? Here what it says, "Run {slash} init to create a Claude. md file with instructions, something is given, okay? " This Claude. md file plays a very important role, okay? It will have some information about the project, you know, so that it can provide a context to the Claude code. Right? Initially to provide some kind of context, okay? You can actually use the Claude. md file, okay? In the Claude. md file, we'll have some information. So, let me just go ahead and ask a question over here. "What is the use of Claude. md file? " I'll just ask question over here. Okay, this is also very good way of learning things. You can directly go ahead and ask questions directly to the Claude code, okay? And it will probably trying to uh know, give you some of the information over here. Okay? "What is the Claude. md file? " Okay, sorry. This I think it did not uh Okay, there is no Claude. md file. It is basically checking uh in my folder even though it is not. But, here it has given some information. Claude. md is a special instruction file that Claude code automatically loads into context at the start of every conversation in a project. It's where you put persistent guidance you want Claude to follow without having to repeat it session, right? So this is really important. Okay, now what the code base is what it

is for right typical content all this information. So I will take all these things and I'll say and remember this Claude dot MD can be created in locations also. First, you can also load this in a project level over here within this project. You can also create this particular file with the name dot Claude dot local dot MD, right? And this is only for your personal purpose, right? You want to give more context to the Claude code, you know? So I will go ahead and create Claude dot local dot MD file, you know, with more information. I don't want to share it that with the team. If you create Claude dot MD in the project level, this you can actually share with the entire team, you know? So that team will also be, you know, when they are using Claude code, some basic context information will be available. And then you can also save this in a global dot Claude dot MD file, which applies to every project that you create in this particular system. So this location, where exactly it is, where you install it, okay? So I will just show you where that particular file maybe, you know? So I'll go to my C drive. I'll say users. I'll say Win 10. And here you can see dot Claude. go ahead and create a Claude dot MD file. So this is where the installation is basically placed. So this basically applies globally to any projects that you create every Right? So with respect to any ID that you use, if you are creating that Claude dot MD file over here, it applies to all the projects. You can also create project level and for your personal level also you can create it. And remember this personal one, you will not even commit in the GitHub, right? This dot Claude dot MD, since it is in a project level, you will be able to commit it, okay? Now I'll tell you, okay, so for your boot camp repo, a Claude dot MD here could be useful. Noting this is LangChain. This is, please create agent all example targeted. So want me to generate one using the slash init skill. So I will just go ahead and write slash init, okay? And then it is just going to initialize Claude dot MD file. And this will basically happen in my local project level, right? So automatically now you'll be able to see based on all the information in this specific project, it is going to go ahead and create the Claude dot MD. And why are we creating it? It is just to make sure that it automatically loads in the context at the start of every conversation in a project. So if I go ahead and close this particular entire prompt and again I open it, right? Again I open my command prompt and start the Claude conversation, it should be able to load that particular context, okay? So now you'll be able to see I have enough context now writing the Claude dot MD file. Automatically you'll be able to see in some time the Claude dot MD file will be available over here. That's amazing, right? So here I'm not teaching you how to do white coding over here. I'm teaching you how productively like in a very good productive way how you can basically manage a project, get more and more information out of it. So now it says do you want to make this edit to Claude dot MD file? I'll say yes, okay? And automatically Claude dot MD file is created and here you can see that. Okay. And now I will reveal in the file explorer. Sorry. I will preview this. So I'll open preview and you can see how beautifully this provide this file provide guidance to Claude code when working with Claude code in this repository teaching material so and so. All these commands are there you will see in this this. Now [snorts] just imagine whenever I start any conversation, the first thing is that this Claude dot MD file will be loaded in the context. And for me, it will be very easy for me to run, right? Now this is one of the thing of getting started over here. Now you can ask any questions, you can create any files, okay? I'll say I'll go ahead and ask hey what does what does uh main dot py file do? Or what is the code in or explain the code in main dot py, okay? Now if I ask this specific question, it is just going to directly read this main dot py. It is basically going to give me a summary like what exactly we do, right? And quickly you'll be able to see that how beautifully all the information will be coming in front of you, right? So now you have started the session. So here you can see, okay, there are two pieces worth knowing the main functions so and so. Nothing is there exactly. It's just a simple main definition of main function. You can see hello from LangChain updated so and so. All the information is over here. Right? So this is how you can actually get started, okay? Now let me talk about more things over here and some of the popular commands that you can directly use. So first of all I'll just go ahead and write slash model as I told you. These all things are basically there. You can click on escape. Now one more important see these all conversation we had, right? If I just go ahead and write slash context. Okay. You can visualize the entire context how it is saving. So here you can see this much context is filled, okay? How many tokens has been filled from 1 million? System prompt 98 tokens 0. 9% system 25. 5 15. 5k tokens.

Memory 63 tokens. Skills 1. 1 tokens. Free space is so and so. Auto impact buffer tokens. This is there. Now you may be thinking Krish, what will happen if this entire token gets filled, okay? And that also I will talking about as we said that there are a lot of videos that are going to come in this specific series. There we'll try to understand more and more. Don't worry about the skills, how did I create the skills? Everything I will create and show it to you and how things are basically done in this, okay? Now let me show one more thing, okay? Now what I will do I will delete this chat, okay? I will again open it. I will open my command prompt, okay? Now you can see that when we are quickly starting something over here, right? And right now you know that we have done a lot of conversation over here, right? A lot of conversation. I want to reload all the conversation. The previous context I want to probably reload it. You can just go ahead and write Claude slash R, okay? So once you basically go ahead and write it, you will be able to see that quickly, you know, it will say you project summary overview request, okay? LangChain updated project overview 1 minute ago, 1 day ago, all the information. So I will just go ahead and press on this. And you can see that all my previous information is basically loaded. Now it is being able to load with respect to that specific context also, okay? That's the most amazing part over here, right? So we have learned about multiple things. slash init. What is the purpose of the Claude dot MD file, you know? Which all location you can basically go ahead and save. You can also go ahead and make this as Claude dot local dot MD so that you can add more instruction just for your project basis, right? And this is how you can basically work with it, okay? So this is just a idea that I'm actually going to give you, okay? And still we are going to see a lot of commands. There is something called as slash agents. You'll be able to see that you can also go ahead and create sub agents. And what is the use of sub agents? Sub agents main use is that you will be able to have different memories for the sub agents, different context for the sub agents and you can basically go ahead and create those kind of agents for different task, okay? Now let me go ahead and ask more information as over here, you know? I'll say what all tools are there in Claude code? See, at any point of times just keep on seeing the documentation. You find any problems, you find you you're not able to understand, ask that question over here. And it will be able to give you suggestions, you know? Suggestions with respect to everything that you go ahead and create, right? So let's say tools is one keyword that is specifically using Claude code, right? And then we'll try to understand what are tools. You can just directly go ahead and ask the question over here. So anything that you specifically see, right? You should be able to see it in a very easy way, okay? And automatically this Claude code will be giving you the answer. So see here over here you can see what are tools and what all tools are there in Claude code. Claude code tools falls into few categories. Here's a full inventory available in this. You'll be able to read any file, write any file, edit, notebook edit, glob, grep, shell execution. You can actually directly work with bash, power cell, web fetch, right? You can actually fetch from a specific URL. You can do a web search. You can actually create sub agents and then again launch specialized sub agents, general purpose, explore plan, task create, plan work modes, you know? We'll try to also understand about the work mode. Then here you have MCP server. Like what all MCP server you can actually go ahead and integrate with it and all, okay? And one more amazing thing is that, you know, there is also something called as uh uh There are three different modes, okay? So if I press shift tab, there's there is something called as accept edit. There is also called as plan mode on, okay? So let's say that I want to probably go ahead and work on a new topic over here, okay? So let's say I will say hey I want to create some some new Jupyter notebook on vectorless rag using um using Python and page index library, okay? Page index library. Okay? Now I am actually doing this in plan mode. You just have to press shift tab. So once I give this new requirement, you know, first of all the Claude code is going to make a detailed plan out of it, okay? So, I'll say, "Please make a detailed plan and save it in plan. md file so that so that

we can see it, okay? Now, I'll just go ahead and press enter. Now, see the best part is that I've told that, "Okay, regard related to vector less rag, I want to go ahead and create a Jupiter notebook, okay? And using Python and page index library is there. Please make a detailed plan and save it in the plan. md file, okay? " So, now here, what it is going to do is that you can see that it'll note on plan final action plan mode extreme restricts me to writing the plan at so and so. After you approve, I can also drop a copy at plan plan. md in the repo if you want. Yes. So, I'll plan this out. Let me research this this all the information, you know, and by default, you'll be able to see in this MD file will be there and here also will be able to get it, okay? So, I will just go ahead and do that. It is See, over here it has done the web search everything, right? Web search, page index, vectify, vector less rag, web search with this. Tool usage is basically happening. Web search is basically happening over here. So, I'll press enter enter. It'll do multiple web searches and it'll probably go ahead and create a detailed plan. Right? So, this is how you'll be able to see that it can call multiple tools also and that's the power of cloud code over here, right? See, it is being able to fetch information for vectify, page index, all the information. So, this is the GitHub it is basically exploring now. I'll just press enter. Okay, enter enter. Automatically, you'll be able to see that you are able to go into get the information. That's the beautiful part. Uh it'll keep on asking. Please make sure to give every permission you see it. That is where human in the loop approval is basically required. Okay? I'm just trying to show you like over here how that MD file is basically getting created and automatically this file will get created automatically and you'll be able to see that, okay? Just let's wait for some more time. Uh page index collab, page index, okay? So, it is doing all that see all these things are basically happening with respect to the engine tick loop. Right? And automatically, it should be able to give me the plan, okay? Bash command see it is going to go ahead and create a file over there inside my updated LangChain. Okay? And uh you'll be able to also see things are happening over here. Okay? So, learn run cloud continue or cloud resume to con- resume a conversation. It is basically exploring these things and all, okay? So, uh once this file gets created, I will be able to show you this plan. md file over here basically getting created and in the next video uh in the next series of videos, we are going to understand the further things. Uh right now, let the code keep on happening and in the next video, we will continue with this, okay? So, I hope you like this particular video. Uh this was it from my side. I will see you all in the next video. Have a great day. Thank you one and all. Take care. Bye-bye.