I Migrated My 10-Year, 29 GB Knowledge Base With ONE Prompt

Added: 2026-05-10

[00:00:00]I just wanted to show you that the migration works perfectly using an LLM.

[00:00:06]This is insane, guys. Thinking about the past when it came to these migrations, how long this took, and you know, losing all the connections and all this, impossible. And now I'm very confident downloading any of the backups from other PKM tools. I'm now able to migrate everything into one local folder, everything interconnected. By the way, this converted from a database into just markdown files as you perfectly can see here, and everything is still there. In previous videos I already showed you how I just have a local folder that runs my whole business through AI. I don't care about Claude, Gemini, or Codex. All of this works in this folder. We call this PK, personal knowledge assistance, and a knowledge base that we are building there is completely transparent. It's all based on markdown files. In another video I also showed you my AI team helped me to build a whole membership learning platform for members. And in there we now have the course to build this PK system. And this PK folder structure that we explain step-by-step here is also available for free on GitHub, which is pretty crazy. They're all the time things, and the latest version is literally insane because this comes with, I think it's nine agents that I consider as really the foundation of any AI team that you need to build whatever you like. So initially we had only these three to help you with this.

[00:01:35]Larry, he's really the core agent, the orchestrator, the Spock, the single person of contact, then Nolan, the HR person to hire new agents, and Pax to redo the research to hire these agents.

[00:01:47]And then a journal writer who will pick up whatever you share with Larry and organizes everything in WikiLinks and interconnected notes, all right? But now there's also Mac, the automation expert.

[00:01:58]So, because I personally already use all these agents that I added here and even more to connect external tools to my AI team. And then, there's now also Silas.

[00:02:09]He's the database architect who ensures that your local folder will stay organized and everything properly interconnected. Then, we also added now Charter, the infographic designer that we literally use to create any social media images, any slide decks, the magic slides that we provide in a membership and all this. This is all based on Charter here. And you might have seen the hype about HTML versus Markdown.

[00:02:35]This is not something I recommend to switch to HTML for everything. That makes just no sense for AI agents to work and HTML for customers, for user-facing content. That's the easy split. And Charter is creating based on HTML all these graphics. And then, there is also Pixel who is actually generating proper images if you connect him to any services like Nano Banana or Open AI's image generator, too. And Iris, she is the design system architect. So, I added her, too, because I thought that looks like a lot of agents here. But with this, you as a professional can really leverage a local folder. Again, it is just a local folder that we are talking about here. And all of these files are Markdown files. They're all readable.

[00:03:24]You can open them up if you go into the team knowledge and you open any up any of the AIs you see. Build an infographic. That's the SOP for Charter.

[00:03:32]So, you see the fold owner, Charter, reusable by any agent. And then here is perfectly explained how she builds these HTML graphics and converts them into images and so on. And all this, you don't need to know because it works out of the box. So, what is this video about? Well, to show you this update first, available for free now, and the course is updated too with now 30 lessons to explain to you step-by-step how this is built, how every agent is built, why the folders are there they are how they are because it's all based on our AI core methodology, which defines a productivity system end-to-end in the physical world. So, working with real teams, and this is just a replication of real teamwork inside a folder for AI teams to work. And what we have here now is this one, bring your old knowledge in, import from any PKM tool. The thing is, with all these iterations, we had people who started with a previous version. Maybe you watched the previous video on this channel, I think it's now 2 months old, where I show you a simple basic folder structure on a desktop, how you can build this and so on. And now people were worried, they built something and they iterated on it, and now we have this new version and they cannot use this version. The thing is, guys, I have two separate folders. They look completely different than what we have here because they are built on top and iterated for nearly a year. I started working with Claude code in June last year, in 2025. Since then, this has evolved and we had the personal knowledge assistant coming in. Our co-founder Paco Cantero built his own personal knowledge assistant and everything got merged together again to this final version that we have here that really have everything in there.

[00:05:21]There's the whole MyLife concept, everything gets organized by the AI team, and you can just plug and play any AI into it and it will work out of the box. So, the thing is, I need this more than you. You could just start from scratch and move a few files over if you just got started a few weeks ago, but for me, we're talking about thousands and thousands of of entries of different work streams and really complex interconnected knowledge that I have already built over the years, but the thing is this version of the folder is just much more efficient and properly structured than the things that I currently use. And that's why in this video, I will be the use case to show you how we migrate this into this folder structure because I cannot wait to start using this instead of the version that I currently use. So the thing is it's so easy to give you a library full of agents and try to sell it to you and so on. Here we are fully transparent and we are using the things for our own businesses. That's why we know it's working because Paco as a multi-business owner uses exactly the same setup as I'm using here to build the My Icro application. And what you guys have access to now is a refined version of it. And I want to show you now the current state of what I used split into two different folders and we will merge this together now in this video into this new folder scaffold. So here's my PKA. I call it already legacy and here is what I have. It's not as refined. We have scattered things. We have here paperless movement, our invoices. We have PKM and there is a lot of things, you know, random. There's no organization when it comes to using the My Life concept. And here that I even built an app to visualize it. If you follow me on this channel, you might have seen it already that I have a whole app built to visualize all this and so on. And this is now in this folder. And if we go outside, let's see. If I open this up in get info, this is a 29 GB folder with 27,000 items in there. So if you have a knowledge base and you worry about migrating, I should worry a lot more than you do probably if you just got started a week ago, two weeks ago. So I will do this live, okay? I will download the latest version. This literally will download it from GitHub directly. Here is now the zip file. I open this up. It It opened on a separate window, and I bring it in here, and there we are. If you look into this, it's an empty folder, okay? It is just really the setup that gets you started. And here's the PKM, the CRM, people, everything perfectly organized, but it's all empty.

[00:08:05]The journal, there's uh just These are just previews, and then here everything empty. All these things are empty. In a team, here is the set of agents that I just showed you on GitHub that we have out of the box. We have the team inbox to share things with the team. We have the team knowledge. Here is now the SOPs. These are the SOPs to get you going. And now the thing is that people say, "But this is not the same as using Claude skills and Claude plugins and so on." And in the end, yes, this is. The thing is, this is much more raw and accessible than if you use Claude skills or plugins, because there you download something, and it goes into this dot Claude folder here. So, the thing is, standard operating procedures is the same as a skill for Claude. So, why do we have it this way? Because it's much more flexible. First of all, I can point any AI to this, and it will understand it, as this is all based on natural language, which by the way, skills and plugins are, too. Talking about plugins, the work streams might be the closest to plugins, because here is a combination of different agents working together on something, and this is using different SOPs to give your final product in the end. And then we have the guidelines, which is the static information, like the design system, for example, so they know your brand design or things like this. By having it this way, it's just highly efficient. The same thing applies to team, okay? If I go to team, you see now there's the agents MD file, which replaces the Claude MD file and makes it just accessible for any other agent.

[00:09:40]But, I show you in this video, as I'm still using Claude as my primary AI, that this perfectly works. But, in the future, I could point just any other LLM to the same folder and would just keep working. And that makes me completely independent. First of all, I have a local folder that I have fully access to readable content that I can open and just read, and I'm completely independent of the AI. And I could even run a local LLM if I really want to have everything private. I'm really excited to start using this instead of the scattered folders that I showed you. And this is how we will do it. So, all I do now, I will point Claude to this folder.

[00:10:19]I do this by right-clicking on the folder, going to services, and I start a new terminal the folder. And you can do this, you can also open up Claude co-work or Claude code inside your desktop app and point them to this folder, and it will also work. I showed you this in the previous video that this works, too. I just prefer using the terminal in order to do this. To me, it's much more transparent and also more reliable. There are always glitches in the desktop version, so here I can be sure I open up now Claude. What I also do is I go to {slash} memory, and here you can switch off auto memory. I don't care about auto memory. In fact, it's just messing things up. So, I always switch this off, also on the desktop version, wherever it's possible, because the knowledge, the persistent memory that we are building up lives inside this team knowledge, okay? We talked about its work streams, guidelines. This is already knowledge that we build up, so the team knows about me and how we work and what we work on. But, on top of this, we have the session log. And in here, the team keeps logging whatever we discussed and the insights. That's what they provide you with the auto memory, but it is hidden in a folder on your drive and so on. Here, you have full transparency and it makes it easy to migrate to any other LLM again. If you use this, you are stuck in there, especially if you use their servers to store this in the Claude desktop. And that's why I want to have it here local and that's the reason why this is switched off. Now, let's get started.

[00:11:49]All I want to do now, so this Claude is now working in this T folder, which is now the scaffold that I just downloaded.

[00:11:56]And all I say now is something like, you know, like "Launch yourself in this folder." or something. I always prefer to say "Initialize yourself inside this folder." And now you will see Claude loads the skill in it. This is just an internal skill that Claude always have.

[00:12:11]If you would use Gemini or Codex, it would also use initialization, which simply just means it goes through the folder here and looks for anything relevant and it immediately found the adapter prompt. It just showed it up here and and already loaded it. See, it says here adapter prompt and if we look into this, this is the instructions to initialize Larry. Larry is the orchestrator agent working in here and see, we have now an onboarding where it asks you for your first name the moment you use it for the first time. So, he used the email that I'm logged in with, okay? That's the reason. And now it updates already in here. Let me go here.

[00:12:49]We have the user YAML and in here is now our first name. So, this is just so he knows who he is talking to. He keeps initializing. See, he goes through the things here. And what he creates now is a Claude.md file. But the thing how this folder works is that he will create just a very basic Claude.md file that just tells him to look into these other folders and files instead of creating these Claude.md files. It's just redirecting Claude to these folders to look them up instead of having everything in this Claude-only format.

[00:13:26]So, here as you see, now it's writing tool pointer Claude.md. But, the project adapted template perfect. I'm really excited because this is a pivotal moment also for me because I'm now literally migrating all my knowledge into the thing that we built for the community and that it became even better than what I was using all the time. Really pumped.

[00:13:47]Here we go. He says already, "I'm Larry, your team orchestrator." He says now what he's able to do. That's the team role say he he knows now about these guys. And here is the Claude.md. Let's have a look what is in there. See, it's very short now. This means the moment you launch Claude, it's not loading in a lot of tokens. It first needs to know what you want and then he looks up what he needs in order to work on it, which saves tons of tokens. And no, this is not HTML files. I will stay on MD files.

[00:14:18]It's much more efficient for the agents to read. And HTML files, as I mentioned in the beginning of the video, is for the user-facing content. But here, I can perfectly read it. It's all natural language. And this is all this is. And here, he doubles down the hard rule that he always points to these agent.md files that he finds also in the team sub folders. And this works perfectly because if he starts now working with the team, he launches agents and these agents are based on these agent.md files. It works. So, now I want to show you that this actually works. I exit now the session. I clear the terminal. And now we're on a blank slate, okay? I launch again Claude as I did before.

[00:14:56]That's a new session. Nothing is here.

[00:14:59]And And now if I say, "Who are you?" as I launched him inside this folder, he now goes into the Claude and you see how fast immediately says, "I'm Larry, your team orchestrator at my PKA." So, now is the moment of truth. I want to migrate now I will start with my actual PKA. I just right click on it. I hold down option and I get the path name this way to this folder. Click on it and I just pass on this folder by I can also just drag and drop this in here, hold down shift, and this just brings in the paste name. And now I can say, "Can you help me importing this knowledge base?" So, the thing is what you need to know, this basic folder structure that we provide here is markdown first. This means that you can open this up perfectly inside Obsidian and I will just open it up in Obsidian because this will be a nice visual effect when we do this. Because Obsidian is based on markdown, okay? I can open this up in here. I go to desktop. Here is the folder that we are currently working in. I open this up.

[00:16:02]Boom. Here is Obsidian and here is all the things that you just saw, okay? That we just downloaded. The thing is, everybody loves this graph view and here you see already that there are things interconnected. The A's are connected with the agents and so on. So, if you just open up the scaffold inside Obsidian, you already see all these interconnections. The starting point is but now what happens if I start migrating? So, as I said, the basic folder structure is based on markdown format with wiki links. These are just wiki links connected. So, if we go to this for example, so you see here the connection. I can go here and I can perfectly read all these files inside Obsidian. But the thing that I'm importing now was based on actually SQLite database. So, let's see what happens how he will migrate this in.

[00:16:49]Because this folder can get upgraded to SQLite on the fly. It's always easy to grade from a markdown folder to a SQLite. For those who of you who get confused, forget it, right? It's just another level where you get a database structure instead of just having markdowns. The thing is, everything will stay the same. You will just get an additional database that makes it even easier to access your information and cross connect information if you want to. So, let's just see what happens if we say, "Can you help me importing this knowledge base?" Now, he goes in. Let's see. Expand this. This is what he's doing. He goes to work streams. So, because in work streams, we have here one work stream, import external knowledge base. So, he's likely this reading this right now. See, he's routing this to Silas. As I showed you in the beginning, he's the database expert. Now, he is just checking to get an idea what this is all about, and so then he will hand this over to Silas.

[00:17:45]There's no cloud skills. So, this is just based on this basic folder structure, and that's the crazy thing about the whole thing. There we go. It recognized now everything. Okay, it found the inventory. What is there?

[00:17:57]1,700 markdown files, guys. 47 team, 12 knowledge, 12 SOP, 17,380 attachments, photos, even videos, journal media, um then also the PDF files, I'm sure he finds. So, this is not a small knowledge base, and it is my real content that he's getting in here.

[00:18:19]He gets now the scope, entity, and so on. So, now I he's asking, "What should he do?" Okay, these are the questions you just need to answer in order to get it going. Amazing, just these questions that recognized these things already, and now he's normalizing this into here, and this makes sense. The thing is now I can go to all my other Evernote app the base and download the backups, and I will do this in a follow-up video where I import even my legacy content from 10 years ago. This will be very exciting.

[00:18:49]And that was it. So, now let's see what happens. Now, he's switching to Silas.

[00:18:52]And if you have a smaller database, you know, as I said, it's a very complex folder that I'm migrating over. That's literally worst case that could happen to you. So, maybe this is helpful for you. But if you do this step, and you just got started with the knowledge base, and you just had a few things in there, it will be much more straightforward, no doubt. But here I want to show you with this Opus 4.7 1 million context window. By the way, here for these cases, I would recommend to use a more much more stronger model like Opus or at least Sonnet with a 1 million context window, so it gets the full picture of what is going on for the migration. Once you work in here, you can use a much cheaper model to work for you. All the LLM needs to do now is to write a proper script, an importing script based on the information it has and then should be pretty straightforward to do these things.

[00:19:43]After initialization, you see there's the dot Claude MD file, there are agents, and there are these agents in there. Claude needs these in order to properly launch different agents in parallel, but when you look into these MD files, you see it's just referring to these folders. It's just redirecting Claude to the right folders, and the same will happen for Gemini and Codex.

[00:20:05]In the end, the actual knowledge for these agents live in these agents MD files, and that's what you will get when you initialize. So, even if you delete this Claude folder here, it could re-initialize it because the knowledge of all these agents remain still here.

[00:20:21]The good thing is that even if you hire new agents, Noel and knows to also update any folders in here for these LLMs, no matter what you use. As you can see here, he has instructions for the commonly known versions in order to make sure that this gets updated and synchronized in parallel. All right, it's starting to import now, and the thing is that I just want to mention here, there have been, for example, quotes in there, and there was not a quote table, so it asked me if I want to combine everything into one new file, but I said, "No, make a new quote entity." And it actually did it, okay?

[00:20:58]Now we have a quote.md template, it created this index file, now we have a PKM quotes index, and there, moving forward, the agents will understand that if I share a quote, where it should go to, and it perfectly will integrate this into my database. I just started the actual import, and you see this live here inside Obsidian. It already expanded very quickly. There are no cross-connections yet, but it's building them out, as you can see. There are a few popping up here, connections. And now see how it's connecting the things.

[00:21:30]You can make this bigger, and this is the famous animation that people see on social media. But this is what's happening actually, okay? So, I'm migrating right now my data into this folder, and the folder is just open in Obsidian, showing now the wiki links getting created by the LLM. That's how it starts connecting everything together. See how the clusters are building. Let's see what this is.

[00:21:52]Blueprints, health system. And as you have seen, it is a lot of content that gets now migrated here. As you can see, it's now still indexing here. It keeps creating the connections for Obsidian, but everything is already there in the folder. We just complete the visualization, see how everything gets connected. I love to see this, but you we can already see here there are all the files. I can open the file. Well, it opens up the journal. It's perfectly imported with all the properties. This has nothing to do with Obsidian. With Obsidian's the visualizer, because we didn't do anything with Obsidian. We just opened up Obsidian to show you this in this graph view here that everybody loves, but also this. But in the end, all this is based now in my folder structure here. So, I can use this with any markdown visualizer. You see here, there's the image reference and so on.

[00:22:45]And in another video, I show you how I can use my own interface to show this.

[00:22:49]But if you don't want to build your own, you can use obviously Obsidian to visualize this. I just wanted to show you that the migration works perfectly using an LLM. This is insane, guys.

[00:23:01]Thinking about the past when it came to these migrations, how long this took, and you know, losing all the connections and all this, impossible. And now I'm very confident downloading any of the backups from other PKM tools, I'm now able to migrate everything into one local folder, everything interconnected.

[00:23:22]By the way, this converted from a database into just markdown files as you perfectly can see here, and everything is still there. We have here now the new quote section. I can open this up. This is actually German quotes that it saved here, but we I can perfectly move from here. We have here now the team. Here's the team. I can read each team's agent file. Everything works. So, I'm really happy with the results. Look at this.

[00:23:48]This goes all the way back to journal entry 2017. I had the first entry on day one that I actually imported previously already to a local folder where I launched my YouTube channel, guys. This is how persistent this knowledge management system was since 2017, and now it ended up in a local folder in this beautiful representation. This is insane, guys, when I think about it, what we had back then with the note-taking apps and what we have now.

[00:24:17]If you want to have the system and you want to get direct support, join us in my iCore. My co-founder Paco Cantero and I are personally there. We share all the details. We dive a lot deeper into this, and we will support you along the way so you get this running. In another video, I show you when I migrated the Larry's team that was working in there, so we have it also in here, and how to get actual business work done inside this folder. More and more will come on this channel, but you can do it already now just implementing the as I said, free on GitHub or inside the membership in the course if you want to have step-by-step instructions how to use this. But we've been never so independent than today having a local folder, and I just point any LLM on it, and it perfectly works out of the box.

[00:25:05]Stay tuned, and I'll catch you up in the next one.