Comment je me suis construit un « deuxième cerveau » IA (Obsidian + Claude Code)

Hello everyone, in this video we will create a second brain. So this second brain will be useful for a lot of things in your life. I use it for several things. Already for my IA, as soon as I have something, it comes to add it to my brain. Now it can also be super useful if you are a developer, you work on a project and you want to have as much context as possible on this project. Here this video will be super useful for you.

We will create a knowledge brain. We will learn to add things to this knowledge brain. This knowledge brain will automatically manage everything itself. It will entertain itself, improve automatically. And you will see, it's not even that complicated actually. Stay until the end of the video. Carpathie, the person who co-founded OpenAI and who also directed the IA at Tesla, posted something on Twitter that absolutely exploded. We see that his tweet has made 20 million views.

And in a week, this project received 5000 stars on Github. And the concept is so simple that you will wonder why no one had thought about it before. And he found out how to turn Claude into a second brain that grows by itself. A personal wiki like Wikipedia but on you, on your projects, on your day-to-day life, on everything you want, on everything you love.

And in this video, we will try to understand what it is about and above all, to recreate it. If you use LLM every day, you may have noticed a little problem. It's that your IA has the memory of a red fish. For example, you ask Claude a question about something you just did. Great, he manages to answer. You come back the next day, he has already forgotten everything. There you will tell me: "But yes, but Ben, there are the projects, there is the memory, Notebook LM, the custom JPTs. " Okay, yes.

But all of this is rag. And rag has a fundamental problem. At each question, the LLM comes to pick up small pieces, chunks, and reconstitutes an answer. Every time since the beginning. And really, every time, it's since the beginning. And that's kind of the problem, by the way. Nothing accumulates.

.

Compiler, that's the word here. And here, I encourage you to go to the link in description. Super interesting link. And here, the idea is the following. Instead of the LLM reading your sources to each question, it builds and maintains a persistent wiki. And here, we're going to have a three-layer architecture.

Au niveau de la première couche, on va mettre ici tout ce qui est sources brutes. PDF, articles, transcripts. Et le LLM va les lire, mais jamais y toucher. La deuxième couche, ça sera le wiki. Là, on aura des fichiers .md que le LLM va écrire et maintenir. On aura des pages pour chaque concept.

for each model. And if, for example, my goal is to make a tool that will allow me to remember all my IA day, the principle here will be to have pages for each concept, each model, each IA box, each benchmark.

And we will also have links between them, as on Wikipedia. And finally, we will have the third layer, the diagram. And there it will be simple, it will be a file "claude.md". It will come to explain to Claude how to structure and what conventions to follow. It's somewhere your contract with him. And the principle here is that Obsidian will become the IDE, the LLM will be the programmer and the wiki will be the codebase.

The principle is that your knowledge becomes code. And here, as we add knowledge, we will enrich the existing pages. We will really have something that compiles itself, that maintains itself and that grows with each added source. So already here, why did we choose Obsidian? Obsidian is a local and free Markdown editor. It has three superpowers.

The first is local + markdown. So we're not on proprietary database here. We're not on mandatory cloud. And if Obsidian disappears tomorrow, well, your folder in which you will put your knowledge is .md files. It can open with any text editor.

So even if Obsidian disappears, it doesn't matter. Because here the goal is that your second brain really belongs to you. Second superpower is the Graph View. We'll see it a little later, but it's absolutely great to visualize things. And that, it literally allows you to see the shape of your knowledge. And finally, Claude speaks very well the .md.

No problem with that. Because .md is the most present format in his training. For him, writing a proper wiki page is super easy. Same, if you want to ask him to synthesize 50 pages, he does it extremely well. Let's go, we're going to put all this in place. So already, I really encourage you to download Obsidian here. Whether you are on Windows or Mac,

This tutorial will work very well, no problem at this level. So come on, I'm going to download Obsidian on my side. Let's go. Then we're going to install it. Come on, very well. And that's it, it's already finished. And there I'm going to launch it. Very well, and you arrive a little on this interface there. And listen, we're going to create a strong case, a vault in which I'm going to, for example, put my IA veil. I click on it.

and I will create a new Vault. Let's go. So here I can call it "Veille IA" for example. I can also choose exactly where I want to come and store this. So I can put this for example here.

There you go, and here we go. Very good, and I get to it. So on my side, I prefer the dark mode anyway, so we're just going to come change that. There you go, it's a little better like that, at least I think. And next to it, we're also going to open our vault. So I'm going to open the good one, it's the VIA here. I'm going to open it directly with VS Code.

And here, we can already see that we have some files in there. It's quite normal. And we're going to launch Cloud. Let's go. And then we're going to take back all the text that was on the link in the description.

We make a copy. We put that directly into Cloud. You have to know that Cloud here is triggered really at the level of the folder in which I have my IA, all linked with Obsidian. And so there, I've already done a lot of tests. I advise you, whatever happens, to put everything in English. I've already tried to put things a little bit in French. Well, it's better to type in any case the commands in English. I warn you. And there, I can tell him something like this.

You should know that all the texts will be available on my GitHub, the link will be in the description. But what did I tell him here? We had taken Carpathian's structure. And I told him that from this structure, from this information, he is now my LLM Wiki agent. He is working on this file of ideas as it is, so that it serves as a second complete brain.

Guide me step by step, create the schema file "cloud.md" with all the rules, configure index.md and log.md, define the naming conventions of the files and show me the first example of ingestion. We will come and place different things on it and it will have to ingest, come and process this data. You will see that a little later.

And from now on, each interaction will follow this pattern. Let's go, we're going to launch this. So there, he's going to create files and folders for me. So I tell him, "Go ahead, it's cool, let's go." So, on the other hand, I got the wrong folder, but it doesn't matter, I'm going to move everything, it will also work, it doesn't matter.

Very good, it just finished. So just one thing, it's that row, wiki, cloud.md, index.md, log.md, it goes here directly in vei.ia. That's very good. And row, it goes in vei.ia too. By the way, assets, it was in row. Okay, so we have this structure there. Let's take a look at what's at the cloud.md file level. Very good, very good.

So there you can read that, it's quite interesting. All of that was generated according to Carpathian instructions that we copied and pasted. There's no problem at this level. And at the level of the wiki, I'm going to have concepts, entities, sources and synths.

I find that the structure of the files is good for the VIA so there is no problem at this level. And besides we will see it a little later, it will be interesting, well you will especially see it a little later but it may be relevant here to come and modify the file "cloud.md" over time to adapt it really to what you want to do here. On the other hand, in this folder here we will create a ".cloud" and

and we will then create commands, /commands, customs, commands. We will have here several different /commands. We will have ingest.md. Then we will have lint.md. I will explain everything to you, really the interest of each command here. Query.md.

And finally, we will have save.md. And then, at the level of the content of these commands, once again, you will make copies and pastes of everything that is available directly on the GitHub link. All these files here will be available. You can make copies and pastes or download them. There is no problem at this level. Now, it would be nice to come and test all this. But first, we will download a browser extension called Obsidian Web Clipper.

which is really not bad and which will really allow when you will be on a page

You are on the page of an IA blog and you want to directly convert the content of the page into .md files. To do this, it will be absolutely great and it will allow you to add this page automatically directly to Obsidian. This extension is available on Firefox or Google Chrome. So here we go. I'm going to add it here to my Firefox.

And we can see very well that now I have the extension here. So now let's suppose that I'm on a blog that is super good and I see an article that interests me a lot. And it's really a great article, I want to integrate it into my second brain. What can I do? I can go directly here on the extension and we can see that it has well shaped the article and that I can directly add it to Obsidian

Let's go. And we can see my article here very well. It is in the Clippings. And I remind you that first, we want to put the data in the row folder. Really, that's where we're going to put everything in a mess. And anyway, it's Claude who's going to do the sorting. So there, we're not bad. What I'm going to do is I'm going to put it directly in row and assets.

And then, for example, I can potentially fall on another article, Open AI Blog. Let's take a look at what's on the Open AI blog. That's it. It's great. It's a bit in correlation with the article we just saw. It's a bit on the security here of everything that is IA, but also of everything that is NPM. There, for once. In any case, I'm going to come...

Get that and put it on Obsidian directly. Here we go.

We look a little bit anyway, what it gives us, very well, and we come to put that directly in Raw Assets. Now I want my Cloud to deal with all this data, that it creates a kind of second brain, that it deals with this data really as best as possible and then we can see this data directly here in Obsidian, you will see it, it's really super nice. So for that,

What I can do is trigger the command "ingest". "Ingest" which means "to ingest". It will come and take the articles that we gave it and it will come and treat it in its own way, in the right way, so that it is then as clear as possible, so that I can find myself well. So there I will be able to launch the "ingest" command and it will be on

So here just for information I renamed my articles to article1.md, article2.md because before there were a lot of special characters and also spaces and he doesn't like that too much visibly. So here we have article2 but I would also like to have article1.md, very good. And we're going to tell him to ingest that and we're going to see the result a little bit, it's going to be super interesting.

Very good, so we really see the important information of the article and it corresponds with what was said in the article. Here I authorize him to create the corresponding .md files, there is no problem at this level. And I would need several, he creates an md file with the Glasswing project, he creates a md file that comes to explain what is SONNET 4.6, OPUS 4.6.

We'll see in a moment what it will look like on Obsidian. Super interesting. I think we can already see the result. Yes, it's being set up. So in fact, in my article, I had information about FreeBSD. Indeed, a zero-day fault has been found.

And that's it. And that's also linked to the Cloud Mythos preview, another article. We see that it is linked with Opus 4.6, that Opus 4.6 is linked with Anthropic, and we can really see how all this communicates together directly there. So when you add knowledge, it will try to see if there is not already something that exists, update it.

improve it and everything will communicate with the existing one. I find that really great. So for example, for the Glasswing project, which was created following the preview of Clone Mythos, you know Clone Mythos is too powerful.

We have potential security issues that can appear at the level of large applications, large companies. So Anthropic decided to create the Glasswing project. They really put Facebook and Windows together to try to correct the security flaws that Claude Mitos finds before giving it to the general public. So here we see that everything is linked.

And we can go back to it, we see that things are added. Let's look at what we have here at the cloud level. Very good. Now I'm going to tell him, well, it might be time here to come and ingest the second article. Article 2.md. Let's go.

Once again, he really understands what it is about and he is creating the pages now. So I like it when he's creating the pages. I like to go to Obsidian and look a little bit what's going on. So indeed, we really had an attack on Axios here. We see that everything is linked with OpenAI, with the article, with Codex. In short, he is creating things for me for about everything.

So here I can see that Axio is linked to a lot of things. We had an attack of supply chain, we can go on it. And we see that it is also linked to other things, it's super interesting. And everything starts here to mix, let's see a little what we have at the level of Claude, very good. Now, okay, I'm adding knowledge here. But what is also super interesting, is once you have this knowledge, it's typing requests.

Tell me what happened to the supply chain Axios attack. So here I ask him to question my second brain. So here he tells me the answer, he tells me exactly what happened and above all, here he worked on something. So he asks me: "Do you want me to classify this answer as a synthesis in the wiki?" Yes.

class this answer as "synthèse"

Ok, very good, so we can see here that the synthesis is now classified and this synthesis will be found directly here on my Obsidian. So every time you will also interact to ask for things, it will also allow him to see, to understand new things and to update the second brain. Now it could also be useful to have a lint

which will allow you to correct the errors, to take a little tour of the second brain to see if there are any contradictions, things that do not go, etc. Very well, and there we have the report, so there are no contradictions. Well, he did some things here, probably relevant. I should look a little more in detail, but we also have a command to come

inspect the health of our second brain a little bit. And now, as soon as I have a new thing at the level of the VEIA, I'm going to put it in, I'm going to put it in there and I will therefore really have my little visual brain here to see how everything is connected. I find this concept really great. If you liked this video, feel free to like, comment and share and we will meet again very soon in the next video. Ciao!