How to Reverse Engineer Anything with AI: Complete Guide to REA GitHub Repo

So guys, this is the repo that is going viral right now, and its job is to reverse engineer anything. Alright? So

we will cover this repo. It has over 200,000 stars on GitHub. I don't need to tell you that. At the start of

October, nobody was even paying attention to it. But now, thousands of people are giving it stars every day.

So many are forking it. It is being used so much. There is no limit to it. Alright? So I thought, bro, let's use

it in this video. So I will literally show you. I will also show you the results. This is very different from

other YouTube videos. Like here, I have opened MindNode. This is an Apple tool called MindNode. Alright? We are going

to clone this. And first of all, let me give you a disclaimer, a warning. Please take this video for study

purposes, for research purposes. Do not clone and sell it, and do not make it your business. You must always remember

. If you think you can clone anyone's business and become big, that's not how it works. You should have your own moat

. Moat means you shouldn't just have code. You should have distribution. It could be that your customer support is

good. It could be the way you tackle a problem, bring in new features, or your marketing is top-notch. The way you

present the product, your UI, UX, the way you handle onboarding, the user flow you decide, and how you support

the user—that is a moat. So you have to create that moat, and code will keep coming and going. How many things are

going open source? How many tools are being built? So, we are going to build this tool. And let me tell you, this

task is huge. My code has been running for many hours, and it's still not finished. But whatever is done, I have

shown that in this video. If you stay until the end. You will learn the entire procedure and many concepts of

how it actually works. You will actually find out, because if you watch other videos on YouTube, you won't get

it. This is very unfiltered content. I know it won't get views, but I am making this video for my loyal

subscribers. I mean, I am so excited for this video. I will be very sad if this doesn't work. But I am still

making it in an unfiltered way. Because I know this video is the only video after watching which you will be able

to use this repo. Which is the REA repo , the Reverse Engineer Anything repo. You will only be able to use this when

you have this video. Watch any other video on YouTube, you won't be able to do it. I can give it to you in writing.

Okay? So now look, what is the difference? So the difference is this. We have created our own application

with this. Okay? I tested it a lot and it has actually replicated this mind node. Everything works in this. If I go

to two here and create a node, then here test three four, whatever you want to write, and you can specifically bold

it. You can change the shape. You can change the branch color. You can make it yellow here. You want to make the

branch color something else. You can do more. You can go into the theme and change it. So it has picked it up

exactly and it is still working. Okay? How does this work? I will explain some functioning in the video, okay? If

anywhere you feel like, brother, I don't understand this, Vivek. What am I explaining to myself? I am getting

confused. I have never been this confused in any video before. I saw a post somewhere on X. Because it is very

viral on X and other platforms too. Everyone on X is talking about this. Someone is tweeting that, brother, I

created this, I created that, I created a game. But when it comes to actually using it, it is very different. So, how

the whole mechanism works? What are the best practices? What settings you need to set within Claude, inside Claude

Code, I have explained that too. So, I have explained everything in such a detailed way that you will enjoy it and

you will actually be able to use this thing. Okay? And let me tell you, I used to do this thing before too, but I

used to do it in a different way. So, my personal opinion, I have shared it at the end, what it is. So, I don't

want to impose my opinion on you at the beginning. Everything, what its pros are. What the cons are. You will

understand all the pros and cons by the end of this video. Just don't skip it. And if I go into text, and then go here

to text, you can see I can make it bold . I can make specific parts italic. Then after that, I will go to the mind

map. So here, everything changes into branches like this. And there are also templates here, you can run multiple

pages together. Like this is a template I created. So AI created it, showing you how to see whatever needs to be

done. All these possibilities are right here. If you want to add checkmarks. We have created our own mind note here.

Which has all the features with proper checkmarks. I didn't have to explain what features to add or what not to do.

Claude analyzed it with the help of this git repo. Now, how its MCP server works, how it connects, how it will be

visible, how to give instructions— literally everything is covered in this video. So you don't need to watch any

other video. Now, you don't really need to know the mechanism of how it works. But still, if I tell you in simple

words, whatever you ask the AI. Right? Like, "Hey, create a clone of this" or "Build this for me." So what does Rea

do? It helps. It works like an MCP and a CLI. In which the AI does what? It uses Rea and then uses analysis tools.

What is there for analysis? There are many tools. I keep forgetting the name of yours. But it is something like

Ghidra. There is one called Hopper. It uses these tools. Now for the browser, Puppeteer is used. Playwright is used.

So those things are used for the browser. So do you need to know those tools? No need. Right? I am not

presenting much either. You will understand everything with the flow of the video. Claude can control your

browser and computer, but I have also explained in this video how to give access. After that, it checks the

evidence—what came with RAG? It reads the evidence. Like code, references. Now, how is the code extracted? There

is some application. Its code cannot be read by us humans. Now the AI reads it. So who handles that task? It does RAG,

but for that, you have to install Hopper or this application, Gather. All right! So how you will set it up and

how it works, I have covered that in this video too. I have explained it later in the video. Now use what you

learned! Okay, so whatever you learned through evidence, you explained it, implemented it, tested it, and followed

up again with RAG, and retrieved the information that was needed, and that loop keeps running. Because of this,

this procedure becomes quite large and token intensive. If you don't have the Max plan, you will face a lot of

trouble and you will have to choose a good model. And how to do it, how to build all the agents, what to do?

Because HQ and Opus have both been used here, I have explained all the things about how we used them in parallel. And

here you can see how much of our usage has hit; we have consumed 19%so far and the work is still ongoing. So I am

letting it run, the work that is currently in progress, because multiple calls are being made, there are

multiple loops, and by running through such loops, it will give us a perfect product, but I would still say to take

it for research and study purposes. Do not clone it. Do not get into these illegal things. All right? Because you

are legally cloning the code. You are not building it yourself. You get the whole code, so you get the whole logic.

All right? You will get to know how the logic comes. What mechanism works. But I will say to avoid it. All right?

Don't get into legal trouble. So, that is basically it. Also, another repo is getting famous. If I go beyond reverse

engineering. Now, they have also done reverse engineering on this. But they have done it with their own

intelligence. This is ArtCraft, about which I made a video just yesterday. If you look, this is the ArtCraft video

that I made yesterday. All right? It has already gotten quite a good number of views. So you can watch this video

if you want to know about ArtCraft. But for now, let's come to this repo, which is of ArtCraft itself. But this one, it

has built everything in Rust. But the difference is that this one has been built from scratch by observing all the

features and so on. It hasn't picked the code directly. It hasn't used the practices that we are going to look at

in this video. But still, it became quite famous. It has built an entire clone of Adobe. Photoshop, Illustrator,

and all those things. So you will get to know everything. You will get answers to all your questions. The

video is long. Please watch it with a little patience. I have exposed everything. I have even contradicted

myself in between, saying let's end this video. But I didn't end the video because the procedure went on, but it

has turned out to be a great and smooth video. Go watch it, you'll enjoy it. Okay? So without any delay, let's start

the actual video. Alright guys, now we are in the RAK repo, so copy this and come to your Claude. Let me open Claude

Code. And after opening Claude Code, we will take a new project here. So I have opened a new session. Let's go to the

folder here. So I need to change the folder here. I will create a new folder . By going to Open Folder. So let me

create one. So here I have created a folder named RAK Testing. Inside this, I will put the first one. Okay? Without

RAK. Okay? We will do one without it and one with it. So what am I going to do? First of all, I will set it up.

Install RAK, okay? This is the one without RAK. So let me change the folder once, Open Folder. So after one

without RAK, let me create one more folder RAK, and after creating the RAK folder, let's open it in a fresh

workspace. And look, remember a few things. Okay? I have kept Bypass Permission on. And we will also keep

Computer Use on. So you will understand its significance in a little while. So here we will provide the git repo link.

And then let's enter it. So I won't use Opus 3.5 right now because it needs to be installed and set up, so Haiku can

also work. So I can go ahead with it, but right now I will use Opus because I have the limit. If you have a low limit

, you can use Haiku or something similar. So I will go ahead with Opus. So look, earlier it didn't install.

Which is good, pretty good. So let's let it install once. Okay? Whoever is installing this. Hopper himself had

installed this. So, I have told him to go ahead and do it. And when you use it , or rather, I am going to show you two

ways. One where we use RA and one where we do not use RA. Okay? Meaning, we are not going to use the complete reverse

engineering concept. So, what will we do? First of all, you need to go to settings, and always remember, like in

Claude Code. In Claude Code, you get many options. So, make sure that things like these browser tools here. Are all

turned on. So that if it needs to interact with anything in the browser, it can. Then, one more thing you need

to do is go to the Claude Code mode, stay in sessions, and go to sessions to find the 'allow bypass permissions'

mode, as it will require many permissions. You can bypass it, but at your own risk. Okay? So, there usually

isn't any problem. I do it that way, but still, do it at your own risk. So, I am giving you a warning beforehand.

And what will happen? Claude will control your computer many times because it will need to understand

other apps, copy them, and then validate them. So, in that case, what will you have to do? You have to keep

this on. Okay? Keep this computer awake for remote control whenever it takes remote control, and then I will tell

you the next thing as well. In case, suppose there is a folder that needs access, you can always grant it. Like

there are 12 folders. Claude lists the six folders you use in Claude Code most and updates the list as changes occur.

Okay? So, you can use the ones you mostly use for Claude Code. So, you can go to 'show folders' to see which ones

are mine, like the downloads folder, so it has access to everything, and if it downloads anything, it can access and

see it. So, I am saying this again. If you are having trouble doing anything, you can adjust these settings according

to your needs. Now, look, the next thing after this is that you have to come here to your system. And after

coming to the system, you will find many options here. So, make sure that if you want to control it from your

mobile, you can turn dispatch on. Otherwise, look, it has installed it now, so it has opened it. Okay? License

file, try demo, offline activation. So let's look at this now. Coming over here now. So the browser use is already

on here. So we can do this. We can use our browser. So there will also be an option to use the computer. Let me

check computer use. Alright. You can see here, if you scroll down a bit in the system, you will find computer use

here. So enable computer use, go ahead and enable this. Unhide apps when Claude finishes. So you can turn this

on. So turn these things on, and how Claude uses the apps you allow. So, whether to run in the background or

give full control. So that is up to you . So what will happen with this? It will be able to use that tool on its

own. Suppose you are doing something on your computer and you don't want those things to be affected, then you can

keep it in the background for that. And you can give full control by default, or it can take full control, or ask for

your permission before taking full control; that's up to you. So I have kept it in the background, because I

need to work here; I don't want it to start taking full control. So that is why I have kept it in the background.

And if you feel that it shouldn't touch any apps, you can also list denied apps here. So all your applications will

appear here. And here, if it needs to do screen recording and check everything, you will have to grant

screen recording and also grant accessibility. So make sure of these things so that testing can be done

properly and Claude can control your computer to execute everything. Now look, to make it simple for you, I told

you that you have two things. One is Ghidra, I don't know the pronunciation, whether it is Gidara or Ghidra. So

whatever the pronunciation is, you can figure that out. So we have one Ghidra and we have one Hopper. So it has

installed Hopper, so that has arrived. But what is its role? I will tell you that too now. I must have already told

you at the beginning, but I am telling you again now because I am recording this part first, before the intro. So

look here, what is the system? It is written here. Okay, our step is, what does it do? Example, so you ask a

question. Okay, who does it? You. And what is an example, how does Notes search work? Okay, I asked, I asked "

How does Notes search work?" So what will it do? So it picks the tool, runs it, and collects the evidence. So who

does this, this part of picking the tool, running it, and collecting the evidence, Ria will do that. Meaning the

GH repo we installed to reverse engineer anything finds the Notes binary and sends it to Hopper. So it

will find the binary and send it to Hopper. So in simple words, if you look here, what does it do? They are

disassemblers or decompilers. So what happens is what runs on the machine, like I will even read this out for you.

Okay? Then it will be easier for you. Apps you download are compiled machine codes. Whatever we download is

machine-compiled code. So it is in such a concise format that we cannot even read it. We won't understand it, but

the machine interprets it. Just numbers the CPU runs with no readable source. Okay? So what do Hopper and Ghidra do?

They are disassemblers or decompilers. They turn that machine code back into something close to readable code. So

not exactly, but it provides it to you in the closest form. Ria is the middleman; Claude asks Ria a question,

Ria has Hopper or Ghidra open the app, and Claude reads their output. You can use either of these two. So this acts

as your decompiler. So it decompiles your machine-level code and provides it to you. Simple. Okay? So here it

provided what we asked Ria for, and this Ria contacted one of these two tools. Okay? Ghidra or Hopper. Okay?

Now it opens the binary and rebuilds readable code. What do Hopper or Ghidra do? It turns it. Like this, there is

some number written like this. Okay? It will turn it into some code like this. So we will understand from the code

what needs to be done and AI will also be able to add properly. So these tools perform this complex task. So now

explain it in plain language. So Claude , it indexes words as you type, then matches prefixes. So Claude tells us.

So the MCP server is Ria's, which we are using. Now Hopper versus Ghidra. So what is the difference between Hopper

and Ghidra? It is made by a small independent company based in France. Ghidra is what the NSA considers open

source. This is open source. What is the cost? This is free. Okay? It has a demo, but a full license costs $ 99,

while this is free. And it has no limits. So mostly you can use Ghidra here rather than Hopper. So after

installing Ria, will everything just work? It will not work. You have to [ use] after Ria. Okay? This feels a bit

strange. Ria, okay. After installing Ria, you will need to set this up. Either one of Hopper or Ghidra. Now,

its demo limit is 30 minutes. Okay? It is a 30-minute session. You can't save; you cannot save. And there is no

demo-related scene in this one. Install size is 100 MB, while this is 500 MB; it needs Java, and you have it. Okay? I

have it, that's why I'm saying it according to the machine. If you look at the speed, it is fast to open files,

while the other is slower. It is heavier on memory. So Hopper has its own separate advantages. So we will try

the 30-minute demo once. Look, it is best for Mac and iPhone apps. So it is best for Mac and iPhone apps. Your

objective could be C or it could be Swift. It can understand and write that . Everything including Windows, Linux,

firmware, and odd CPUs; it can understand all those things. Decompiled code quality: its quality is good,

while the other's is usually better and more detailed. Okay? It stays very detailed. Analogy-wise, if we

understand it as a quick handy pocket knife versus a full workshop toolbox, this means Ghidra is being described as

the best. Okay? So we will try it out and see what needs to be done. Now, before versus after, what Ria can do.

Okay? I will explain to you right now what Ria can do. Without Hopper and Ghidra, and with them, what are the

possibilities. So, what is our target? Electron apps like Slack, VS Code, Notion, which are made in a way like a

wrapper. They are using a browser. These are browser-based things. So, wrappers and such are mostly used. I

don't have that much knowledge yet. But Electron apps work since they are JavaScript. So basically, Electron is

used at the JavaScript level. Because the Electron apps I have built are mostly the JavaScript ones. But we use

it like a wrapper. Because I built it browser-based and I wanted to run it as a desktop application. So I used

Electron for that. Now native apps, many Windows tools work in it, and it's the same with this. Websites work, and

it's the same. Now with native Mac apps , when does the problem arise? When you go to native Mac apps, if you open a

normal website and want to clone it or take some of its features, that's a different story, but there is some

differentiation here. Now, with Windows .exe and Linux programs, you hit a limit. Okay? The same limit applies

here; you can only see file names and text strings, and you face the same limit here too. So it cannot read Mac

apps. It cannot read Windows apps that well. It cannot do things like game engines, firmware, or CLI tools. But

this is solved with these tools. This is readable code. One can read the actual logic. One can understand the

logic, like what logic has been used. What are the frames in the animation, what are the transitions. How the

button is moving. What is performing inside it. So you can extract all that logic. But remember, this is for study

purposes only. And it does not suggest that you clone it. That is very wrong. If you clone it for study purposes,

it's okay. Okay? So you only have to study from this. Please remember that. Now Ghidra is especially strong here.

Be it game engines, firmware, or CLI tools. So Ghidra is much stronger. Example: How does Apple Notes work here

? Now, what was it before? There is a function called IC Search Indexer. You get the names but not how it works.

Okay? This used to happen before. What will it be like after? Once you install Ghidra or something like that. So here,

it splits text into words, stores them in a local index, and ranks results by match type; here is the decompile

function as evidence. So this is how it is going to work. So I have explained the logic to you on how things will

work. I am explaining it in a calm and detailed way. Okay? I am explaining it slowly because all the videos on

YouTube are quite fast-forwarded. And I don't want you to fail to understand. Understand these things concept-wise.

Now, what happens next? Now, help me to run Ria once. Okay? So let me write it and let's see how this works. So we

don't use Hopper. I'm thinking of just deleting it. Okay? We can delete it or even run Hopper. Let's see. Now look,

when we ran it normally, it worked with Obsidian, which I had installed. So it generated some report for Obsidian.

What were the background workers? Heavy work done off the main screen, blah blah, what are the findings? In plain

words, what is it trying to tell me? So how many function calls are there? What are all these things? It told me

everything. But now I want to use it properly. So it's telling me to include a new session. So first of all, I will

say. So I have a question. Would I be able to see all the MCP tools of this Ria under my connectors? Or will it be

under my desktop MCP config? Explain this to me or whether I need to quit Claude or not, because I don't even

know the answer to this yet. I guess I'll have to quit. So let me see, my main purpose is that I want to copy an

app. I will show that to you. Look, that application is this MindNode, where we can create anything inside

MindNode. Like the landing page system. I have created this. So I want us to build something like this. I want to

build exactly the MindNode application that I have. We can also do AI integration and so on. But let's see if

it can build it or not, that's what we need to check. Now you can see that its code is at this folder level, the

folder we are in, I guess it did it at the .claude.json level here, yes this is where it's set up, the local MCP

server for the code tab. Okay? Why don't you say it? I just checked this session server list. It shows your

connectors, Gmail, Sag, blah blah, and desktop order API. Okay? I have done all this too. This session started

before Ria was installed and servers only load. What to do? You want Ria to do this. Okay? The code tab, start a

new code session. Ria will show in the session's MCP server list with about 170 tools. Okay? Quit Claude? No, there

is no need. The chat tab as well, run this setup for Claude desktop below. Then reopen it. Yes, the docs say to

quit Claude. Fully first, because a running app can overwrite the new config. Alright? Let's see. What I will

do is, for once? I will open this folder here as well. I went to a new session, and in the new session, I

wrote, "Can you access Rhea?" So it's saying, "Yes, I am able to do it." So it's okay. So let's open it once. Let's

suppose we open it in Cursor. Okay. Okay, it's asking for permission. Let's say yes. Now look, whatever output it

saves, it will save here. So we can also see where it is saving the output. But for now, let's talk about something

here. So you can see where it is located; we have found that out. Meaning, it is there. I mean, in case

you don't see it, there's no issue. You can also transfer it, by the way. No problem. If you want to see it visually

, you can do it. But in the connectors, we won't see it, man. This concept of Rhea, and you can go here and check.

Here too, like the developer one, it appears here as well. You will have to transfer it. Like my previous video was

on Vectorcraft, which is Artcraft's, the Vectorcraft that comes inside it, just a copy of Adobe Illustrator, we

did that. So that thing, so let's see, everything is done here. I want to use Girdhara. Alright? I don't know the

spelling, man, what it was. But let's use Girdhara. So look, it's not able to understand, so I will say, "Understand

Rhea first and let's use Hopper." Alright? Let's use Hopper. We are using Hopper now. Let's go to Hopper. Where

did Hopper go? Alright? Hopper, Hopper has arrived. So, try this demo. I clicked on "Try this demo." So our demo

can run now. So we are going to use Hopper. So let me ask Claude. Brother, use Hopper. Understand Rhea first and

let's use Hopper for now and let's recreate MindNode. Also, select the stack accordingly. But first of all,

analyze it very deeply and let's replicate it completely from top to bottom. I should be able to actually

create documents and import the real MindNode format app files, as well as be able to create more nodes, control

colors, fonts, and do everything from it. I want a complete recreation of it. Okay, so we will do this, and after

that, I will say you can create up to, I'll say up to eight sub-agents. Use Haiku 3.5. Or use Opus 3.5 according to

the complexity of the work. Try to use Haiku to save tokens without compromising, without compromising the

result. Okay? Okay? So, look, don't compromise the result. So now, let's switch to Opus 3.5 here, let's switch

the model and keep it at high instead of medium. Okay? And let's go. So now it will run. It will read the MindNode.

Let's see how it works. So we are working in a new session. Remember this . If you want to perform this thing.

You have to do it in a new session. So I am showing you everything practically . The video is getting long. So guys,

please watch the video. Like it, build the hype. Because this is very important so that it reaches more

people. Now it's asking for a permission, so let's allow it for Hopper. So now you will see how it is

performing. I am just showing you how everything is going to work. Loading Mach-O file. Okay? So now you can see

how the work is being done. So I am showing you everything in real time. How it actually works. Whatever happens

on the screen now. I will keep showing you. I don't know, my demo period has expired in this. You can see. So the

demo period has been expired. So now we will have to use another tool. So guys, I told it to use Ghidra. That should

save the work because Hopper isn't working. I know it won't work for you either. So you have to say it like this

. So alright, our Ghidra has started installing. So it is verifying everything here now. So we are going to

run this. So you understood the differentiation, right? I have kept this video completely unfiltered so

that we non-coders can actually implement this thing properly. So that is why I am keeping this video

completely raw. I know the video won't get views. It's my loss, but I am making this for those who actually need

good content that they can actually rebuild. Okay. So now I guess it is installed and it is figuring out all

these things by itself. Alright. So it is going to create sub-agents as well now. It has written sub-agents here. So

now the whole thing will be created. You can see the MindNode clone here. It has created the scaffold folder, and

this is where our entire application is going to be built. So let's see how it works. So it will be quite interesting

to watch. And yes, this is for study purposes; our main motive is not to clone anyone. Because I know many

people might think we are cloning it. But it's totally illegal. If you are doing it, please don't. For study

purposes, for research, to understand the code logic, you can do those things . So it's only for study and research

purposes. The rest you are wise enough, I know you won't do it. Okay? So the concept of reverse engineering

everything and anything is that it will analyze all these things. You can implement everything with Claude Code.

It will control the rest of the computer. It will use the computer. So now let's see what things it does in

this. Alright. So the phase has arrived where Claude wants to take control. So, let's allow it for this session. So

that it can fully control MindNode. So look, Ghidra is still installing. It's a bit of a large application and I

don't know about the internet, maybe it's slower. So fine, it is still installing. So it wants to study, so

let's let it study. So you can see, for studying. Now it is controlling it. Okay? Everything is happening in the

background because, as I told you in the beginning about the settings we configured, it is controlling it in the

background and capturing the screen automatically. So just keep those settings in mind, and whenever it needs

to take full control, it will take full control. So alright, quite a lot is being built here, like the layout,

MindNode features, UX research; it has created sub-agents for all these things . So it's using Haiku 3.5, for some

it's using Opus 3.5, so in this way all its agents are set, and I guess it's also asking us for some permissions. So

import completed, so I guess it tested it by importing and all. So even now, look, this thing Ghidra hasn't been

used yet, but it's still running. So let it be, let it keep running. Let's see what kind of output we get. Alright

, what we were waiting for, Ghidra is now installed. Download is installed at 458 MB. So I am restarting it. Okay? So

now everything is going to work through this. So we have the entire logic and the procedure to decompile the code is

about to start. So let's see, now you can see these are our parallel tasks. Now I asked to do this so that our work

can be done quickly because I have to make the tutorial and it's 6:59 right now. Don't know when I will upload it.

Man, I thought I would show you how to build it without RAG as well. But I don't think I'll be able to because

there's a bit of a time crunch and I don't think I'll get that kind of output. So I've tried other things in

the past, but the output doesn't come out like that just using the application, so we are going to do

everything with RAG. So six tasks are currently running in the background. Everyone has different things,

somewhere Haiku is being used, somewhere Opus is being used. It has given me everything along with the

reasoning. And our Ghidra is also running. Okay? The binary to arm 64 and run Ghidra analysis. So this is also

running in parallel. Now look here, I also feel that it is timing out. This is a problem. I mean, it is a problem.

It's not that easy to copy. So rather than the entire application, as people on the internet are also saying, just

copy one feature. So I am showing you the complete tutorial. It's not gonna be that easy. But yeah, if you are

doing it, then do one feature. You are building an entire application. If you are not able to understand one feature.

If you can't understand its logic, then use it for study. Don't copy it. Okay? Directly. But yes, for study purposes,

one feature or some specific things, that is going to be easier for you. Rather than looking at the entire code,

it will be very large. It will be very difficult to understand. The limit will also be reached. So, there is also an

issue with the limit. But let's see how it performs. So here, it is running again with a 2-hour limit. So look, the

limit is quite large, at 2 hours. So let's see what happens. I am pulling MindNode's official Mac shortcut list,

the binary including the key bindings, and the editor agent needs them. Okay? So, the feature inventory is finished.

I have added MindNode's official Mac shortcut table from blah blah blah. Okay? And the seven built-in themes

extracted from the apps, and passed both to the editor and rendering agents . So, it has sent them to whichever

agents of ours are running here. The communication happening between the agents is also being tracked. But don't

make this mistake. You can tell the AI at the very beginning that its limit shouldn't be reached. You can set it to

2 or 3 hours so that a timeout doesn't occur. Actually, to tell you the truth, it took quite a lot of time. That's why

all the people on YouTube haven't actually shown this properly, and what is happening is that it's taking a lot

of time. You can see it has been 37 minutes. This has been running for 31 minutes. And it says it could take over

an hour. Try asking a question too. Although these tasks are running in the background. So I will let it run. I

will put the results on YouTube Shorts or a link below this video to show what the result looks like. I will make a

small unlisted video for it. So I will definitely show this. But this was the procedure. So that you can do it. I

promised I would upload it today, so I am going to release this video now. So this is the entire procedure. I have

shown you the entire setup. What will happen? How it will be set up. It controls the entire screen and

everything. If you allow it to take screenshots. It actually does all these things. Okay? So I hope you have gained

a clear understanding of how you can use it and what the procedure will be. So thank you so much guys for watching

this video until the end. I hope you will now be able to recreate it; you will find the GitHub repo below. It has

already received more than 200,000 stars. It is going quite viral. You should also try it out. If you want to

take a specific feature or clone the entire platform for study or research purposes, you can do that. Otherwise,

if you are launching a product commercially, this is very wrong. You should not do that. Okay? Now, it is

taking a lot of time to compile this. But if you use Hopper or something, or get a license for $ 99, that will be

better for you. It will work much more effectively. Rather than this, because this is slow. Okay? You will find that

the GitHub one is slow. So, that was the reality of what it takes, plus your tokens will also be consumed. When you

provide an entire application. You can see that I am on the Max plan. If I go to usage on the Max plan. So, it has

reached up to 13%. If your normal plan gives 5x more limit, then divide this by five, so how much would that be? If

it is covering 13%here. Suppose if it goes up to 20%, then it will consume 100%of the other one somewhere. If it

hits 20%, then by that time the session will already be over. So you will have to try again and wait to do it. So, it

consumes a lot of tokens, man. So what happens with this? I mean, logically speaking, it is more beneficial.

Otherwise, my opinion is that you should screen record whatever specific thing it is. What I mostly do is screen

record whichever platform I am working with. Okay? Let's suppose I have an application here, like this one. Now I

want a breakdown of this. So what will I say? I will tell Claude to use sub-agents and such for it. To use H

5.5. Which costs us 40x less. So it will create sub-agents. Using the approach I mentioned, what can you do?

Show it by opening it up. Like, if you want to say that this feature is inside it. Open it. Show it intact, that all

these things are there. And add to it as well. Okay? Whatever you say, first of all, it will extract your voice

separately. It will install Whisper and then transcribe the audio to understand what it is. Then, it will match it with

the time and look at the screenshots accordingly. So, you can ask it to analyze at least 15-20 screenshots

every minute. Then it will be able to understand. I have performed video editing in the exact same way. I do

video editing. Right? I had posted a video before this as well. I have built a whole system here to show how it

works. So, the entire export happened right here. You can see how much time it took. This is how the workers were

working here. Creating a clean caption file and making these videos, I have done all of these things. So, the

export was done here. If you have ever done video editing or graphic design >> Where you can see the entire video.

This is the edited video. It is edited so beautifully. >> Created it instantly. Watching it do

all these things right in front of your eyes, man, it was amazing. So, in this video, I am going to tell you the

entire procedure of how to connect everything, plus I will also give you the prompt that I used. I faced a lot

of problems with this. You can see below, this is how I generated it. But I built an entire system for this as

well. I also provided a lot of video references. How it works, how everything is done, and I trained it.

So, what does it do with this? It looks at it frame by frame. I am just explaining the frame-by-frame concept

to you to show how it sees these things . Frame by frame, like this. It is looking at these frames to see what the

style is. How it has been edited. So, it notices frames like this. So, what does it do? It does not analyze just

one frame. It takes multiple frames, gathers them, and uses them in this way . So, what you need to do is: Clearly

tell the Claude that when it analyzes, it should analyze at least 10 frames at once. So, it will extract 10 frames for

every second and notice things accordingly. So, 10 frames will be enough to understand. So, perform

things slowly, do not go too fast in the screen recording. So, I use this approach. This approach is more

efficient. Otherwise, the procedures become way too complex. If there is something that you are unable to solve

even after giving screen recordings or screenshots to the AI. Then you can go for this method. You can also see here

how the work is being done. How it is being analyzed. Now it is checking here if the square it is making is in the

right place or not. Now it made it here , then took it here after analyzing. So , for multiple images like this, it

analyzed 16 images together here. So, 16 frames have been analyzed here basically. So, this is how this system

is working. So, analyzing here, recreating here. Just now it downloaded the logo. For the video editing I get

done, I have created a whole system for it. Okay? So this is the system that you should use. It's not like we are

doing this just now. We have been doing this for a long time. Okay? I am also doing it and many other people are

doing it too. So, what is it? REA, okay ? REA, meaning it just makes things a bit easier. Things become much more

organized framework-wise and systematically. That is all, nothing else. Besides that, if you want to

control the browser or do anything. It is a systematic format. Someone who doesn't have much knowledge, who is a

beginner, can do this. But they will still be confused, like why is it taking so much time? Where are things

happening? So, it's kind of that. It's not like the 2 lakh stars hype is justified. But it doesn't seem like

that much hype either. I mean, if you ask me personally if I am going to use it a lot, I won't. You can watch a

separate video of mine. If you want to recreate something that I had forgotten . I thought I had already uploaded the

video. But I am going to bring a separate video on how to actually recreate and how these things work. An

actual tutorial will come on this that I personally follow. Okay? I am going to bring a video on the things that I

personally follow. Okay? So don't worry . I will definitely bring a video on how to actually do recreation in an

easy and quick way. Otherwise, you can check out my AI SaaS playlist. So if you go to my channel, you will find the

playlist there. So here is the AI-related playlist as well. If you go to View Full Playlist, you will find

the entire playlist on how to build AIs . In this, I explain front-end and back-end stuff at a very high level.

Similarly, you can build desktop applications. You can use Electron and such. A wrapper gets built. If you tell

Claude to do the same thing, it will do it for you. So that is also very simple . And I will definitely bring a video

on cloning, on how to do cloning. But it is a bit complex through this. Okay? Whatever you want to do, yes it works,

but it will be quite time-consuming for you. So I will share the results with you in the description. Maybe a day or

two after the video is released. So alright guys, by the way, they are also testing it in the browser right now.

Okay? You will see. So what is it testing in the browser? Let's see once. It has run in the browser. So let's

open it on the browser once. Let's paste it. And okay? New document. Look, it made it exactly right. It made it by

analyzing with the skills of AI. So if I go here and type "Mind Node Test 1". Okay? It's actually working, but it's

not fully built yet. Do I guess I have to draw the connections myself here? No , these connections are different.

Siblings add node, so it happens by clicking add node. It's not the normal behavior. So it is working by entering

Test 1. Test 2, Test 3. Okay? If I have to do more, I have to do it through this only. Add node one. Two, three,

four, five, yeah it's working, I mean kind of the text mind map, everything is working. Okay? There are many

options in this, man; there's even a theme, like which theme I want. Okay? It's pretty good. Actually, I can see

the outline here too. I mean, it cloned it, however it does the cloning. Alright, outline mind map, and if we go

further, add node, what else is it giving here? There's font too. Testing 2.3, if I bold it, is it bolding? It is

bolding too. Italic, underline, strike. Okay? What will be the text color? If I want to highlight the text, brother,

it's highlighting too. So it's pretty good. I mean, it's not even fully built yet. Okay? It's not even fully built

yet. Still, it's pretty good. Sir, a lot of things have turned out like that . Okay? Let's analyze. I mean, I didn't

expect it to do this well. But brother, we are also using o1-preview 5.5. So that's also a factor. So, waiting for

you on whether you will enable my notes MCP server. So I can check our exported files in the real app. So, it's asking

to connect the MCP server to do it in the real app. So, we can avoid that for now. But I will point out some changes

that are needed here. The changes that I feel are needed here. So, let's go to this platform. And if we refresh here,

will this file stay or disappear? So, it's here, Creativity Dinner Party. Man , it has created so many files. I mean,

it's pretty good. Actually, I need rounded, I need pills. So, it has created the system exactly like that.

If I want to share it with someone, I have to share it in PNG. So, if I click on PNG, it is also downloading in PNG.

So, this is how you can do it. As for the rest, logic-wise, it will handle that. So, it's pretty good, man. I mean

, I'm happy. It has created it quite well. And I guess we can also create multiple boards here. There is

Creativity One. Need to open one more document. So, let's go here and open one more document. So, this panel opens

on the side. Okay? Yeah, this is also crazy. So, Test 2.3, I mean, if I have to create anything, I won't need a

MindNode subscription. I'll just use this. That's the kind of system it feels like. But our stuff is saved here

too. I guess everything we created is saved. It's appearing even after a refresh. Going into shortcuts, all the

shortcuts are visible too. So, it's dope. So alright guys, the Ghidra analysis is complete here. It has over

200,000 functions. I was about to end the video, man. It finally gave the result, it has finally given the result

. So now, brother, I have to show what it is. CLI doesn't persist analysis between calls; each decompile would

re-trigger the 43-minute scan. I am setting up a long-lived Ghidra MCP server with a local bridge instead. So

one analysis can serve all decompiles and queries; the persistent Ghidra session is starting. Its first analysis

will take about 45 minutes, and after that, queries such as 'find the code that needs shape type and decompile it'

will answer quickly. I will use it for the remaining open questions, the legacy shape dash and layout codes, the

theme shape mapping, and the layout constants. The layout agent is still running its second pass, so that's fine

, I will let it run. Right, it is working in a pretty good way. So, this analysis is done, it is working on its

own, so let's let it be. Actually, it was asking me to connect the MCP server , but I am not doing it. It might be

connecting internally. So, usage is at 19%, so I will let it run. There is no issue because we don't have a shortage

of this right now. Then what will it do after that? It will wrap it into a desktop application. So that's it, guys

. I hope you understood how to work with this. It will take a lot of time, but it will create something great for

you. So that's it, guys. Vivek Mishra signing off. Bye-bye, take care, and keep learning, and always keep that

curiosity alive because you guys demanded it; I announced it yesterday, so it's here. I thought I wouldn't

upload it, but I did anyway, all for you. This is official, unfiltered content from my side. Bye-bye and take

care.