# 163 I Built the ULTIMATE AI Second Brain (Karpathy's LLM Wiki Setup Guide)

Show Notes

🔗 Deploy your own Hermes agent on Hostinger -10% off any plan with code COREY10: http://hostinger.com/corey10 (use code COREY10)

Grab the free step-by-step Hermes Second Brain setup guide here: https://corey-ganim.kit.com/60fc0fe6d9

In this solo episode, I walk through exactly how I built the ultimate AI second brain using Hermes and its built-in LLM Wiki skill — inspired by Karpathy's wiki setup. I break down the three core operations of the Hermes agent (ingest, query, lint), the three layers of the knowledge base (raw sources, the wiki, schema/tags), and why I chose Hermes over Claude Code, OpenClaw, or Obsidian for this build. From there, I show the full one-click deploy on a Hostinger VPS, the OpenAI model setup, the Telegram bot integration via BotFather, and the Markdownload Chrome extension trick I use to feed any tweet, article, or webpage straight into the wiki. By the end of this episode, you'll have a clear, step-by-step path to deploy your own self-improving second brain that gets smarter the more you use it.

Key Points

Hermes ships with a built-in LLM Wiki skill that handles the second brain function out of the box — your job is to curate sources, the agent's job is to summarize, tag, and file them away.

The knowledge base has exactly three layers: raw sources you give it (read-only), the agent-owned wiki of markdown files, and the schema/tags Hermes builds to make querying low-friction.

Run the Lint command roughly once a month — it audits your wiki for contradictions, sources older than 90 days, and oversized files that should be split for more accurate retrieval.

Deploy on a Hostinger VPS instead of locally — it's a one-click install with no terminal, cheaper than a Mac mini, always-on, and your data stays private.

The simplest workflow for feeding the wiki: use the Markdownload Chrome extension to clip any tweet, article, or webpage to markdown, drag it into your Telegram chat with Hermes, and tell it to add it to the wiki.

This is a compounding asset — on day 1 the knowledge base is the dumbest it'll ever be, but feed it consistently and by day 90 it becomes incredibly useful for retrieval.

Hermes agent VPS deployment on Hostinger (use code COREY10 for 10% off) - https://www.hostinger.com/vps/hermes-agent-hosting

Markdownload Chrome extension - https://chromewebstore.google.com/detail/markdownload-markdown-web/pcmpcfapbekmbjjkdalcgopdkipoggdi

Telegram BotFather - https://t.me/BotFather

OpenAI Codex - https://openai.com/codex

FIND ME ON SOCIAL X/Twitter: https://x.com/coreyganim Instagram: https://www.instagram.com/coreyganim/ LinkedIn: https://www.linkedin.com/in/coreyganim/ YouTube: https://www.youtube.com/@coreyganim

What is Build With AI?

Most AI podcasts talk about what's possible. Build With AI shows you how it's done, live. Each episode, host Corey Ganim brings on entrepreneurs and operators who share their screen and build real AI automations, workflows, and tool setups right in front of you. No boring slides. Nothing that hasn't been battle-tested. You'll watch actual implementations get built from scratch so you can follow along and do the same in your business. If you're a non-technical entrepreneur who wants to put AI to work without becoming a developer, hit play and build along with us.

Corey Ganim: You're in the right place if you're finally ready to build your AI powered second brain because in this episode I'm going to be showing you how to build a compounding knowledge base that actually gets smarter over time. Specifically, I'm going to be showing you how to use an AI agent called Hermes, which is an open source AI agent, which is kind of like open call, but a lot smarter and a So why are we using Hermes versus a Claude code or an obsidian? Those are the two most popular tools that everybody talks about when it comes to building a second brain. And there's a couple of reasons. So first and foremost, Hermes has a built-in skill that comes off the shelf called the LLM wiki skill, which is literally the second brain function built into Hermes. So let's talk about what that looks like. So first and foremost, with this skill, there's a division of labor between you as the human and the Hermes agent. easier to set up. So in this video, we are going to be breaking down the why behind using Hermes and not other popular tools like Clog Code or Obsidian, how to quickly and securely set up Hermes with zero code required, how to set up your second brain with Hermes, and then how to test it to make sure it works. So by the end of this episode, you are going to have a fully functional AI powered second brain giving you and your business a massive advantage against your competitors. So let's get started. So you as the human, your job is to curate the sources. Your job is to go out and find the data and the references that you want inside of your second brain. Now the Hermes agent's job is to summarize and file that away for future use. So what does it look like ⁓ as far as the core operations of the Hermes agent and filing those away? So the first thing it's going to do when you give it a new reference file, let's say we give it a tweet or an article or a website. it's going to ingest that document and add it as a source to your Wiki. So it does that automatically based on the built-in skill. The next thing that it's gonna do is when you ask the Hermes agent a specific question, it is going to query your Wiki. It's going to look at anything that you've given it as a source, anything that it's ingested and give you your answer based on the query that it did to your knowledge base. And then lastly, The third core operation is what's called Lint, which is an audit slash health check of the Wiki. So you can run the Lint command. ⁓ best practices to do this once a month. And what that's going to do is the Hermes agent is going to look at your entire knowledge base, your entire second brain, and it is going to determine, do we have any sources that are contradictory? Do we have any sources that are more than 90 days old that are likely out of date? Do we have any. sources that are too long if there are sources over I believe it's 200 lines or maybe it's 200 pages Hermes will flag that automatically as a source that can potentially be split into multiple individual sources which will make the querying function of those sources a lot more accurate and then As far as the three layers of the knowledge base ⁓ When it looks if you were to look at it under the hood, there's only three layers, right? So there's layer one which is the raw sources. This is what you as the human are giving it to ingest. So these files are read only. The Hermes agent does not touch the raw sources. It's simply ingest them, right? The second layer is the actual Wiki itself. This is the agent owned essentially web of different markdown files. So the Hermes agent is going to take the files that it ingests that are sitting in layer one, and then it's going to organize them into a Wiki. So then layer three is the actual schema, the structure and the tags that the Hermes agent builds out in order to make the querying operation easy and low friction. So it's going to apply certain tags. It's going to create an internal structure to make the querying a lot easier. Now, again, why are we using Hermes versus Claude code or something like OpenClaw or Obsidian? Now, the only reason you would really need to use Obsidian for your second brain or for your knowledge base is if you as the user plan to look at the files or edit those files yourself, right? Me personally, I have no desire to ever edit my knowledge base files by hand or even read them by hand. If I want to ask a question or I want an answer, I'm going to just ask my Hermes agent. So the only reason you would use Obsidian is if you want to look at those files yourself, but it's a lot easier just to go through your agent. Now, Hermes as an agent has a self-improving built-in learning loop installed in the agent out of the box. So, open clause notorious for forgetting things and for going off track and losing context, but Hermes has a built-in learning loop where it will literally proactively create skills that it thinks you need and log learnings over time to where it actually gets smarter the more you use it. It's got a built-in scheduler for recurring tasks that actually works. Whereas OpenClaw is notorious for the recurring tasks, otherwise known as cron jobs, to forget to fire or not work properly. Just like OpenClaw, Hermes has multi-platform support, so you can set it up on Discord, Slack, Telegram, WhatsApp, Signal, email, et cetera. Just like OpenClaw, it will deploy sub-agents for complex tasks. It has ⁓ multi-model support. You can set it up via OpenRouter. or via Anthropic or OpenAI, and we're going to go into exactly how to do that here shortly. And it has persistent memory and it can search through past conversations to know exactly what you talked about. The biggest differentiator is the self-improving built-in learning loop where it creates and refines its skills over time. So next, we're going to go into exactly how we set Hermes up and deploy it on a virtual private server inside Hostinger. We're going to learn how to quickly and securely deploy the Hermes agent on a host in your virtual private server. But before we jump in, you can grab the full Hermes deployment checklist that step step by step, super easy to follow in the description or the show notes. So that way you don't need to go back and rewatch the whole video. Now, why do we choose a VPS over hosting it locally on say our Mac or our windows PC? And the reason for that is when you host it on a VPS, it's very easy. It's a one click. install and set up. You don't need to jump into the terminal. You don't need to know how to code. And it's cheaper at the end of the day than hosting it on a Mac mini, which is going to cost you six or $700. Your data is private and the availability is always on. You don't need to worry about your computer turning off and your agent turning off with it. So what we do is we come to this link here and we're going to put this link in the description. It takes us to the Hermes agent deployment page on Hostinger. Now, as far as which plan we choose, The KVM2 plan is going to be what you need here. It gives you enough bandwidth to where you can deploy multiple agents if you so choose. And you can see here that they also have a one-click open-claw deployment. So we could theoretically deploy multiple Hermes agents and multiple open-claw agents on the same VPS with this plan here. It gives us enough bandwidth. So what we're going to do is we're going to click Deploy. Now it's going to give us a couple of different pricing options here. So obviously the best bang for your buck is going to be if you do the 24 month plan, which brings it down to 8.99 a month. You could use my coupon code, which in this case is Corey 10. And that's going to give you 10 % off, bringing it down to less than essentially less than, well, $8 and nine cents a month, right? You also get a free domain with it. So I think that's a pretty good deal overall. Click continue. And then what we'll do is we will jump into the portal and show you how to set it up once you buy your ⁓ virtual private server. All right. So once you've signed up for your server, what you're going to do is you're going to be brought to the ⁓ dashboard here in Hostinger. So you're going to come over here to VPS. You're going to find your private server here. It might take a couple of minutes to actually spin up and be fully running. So make sure the status is the green check mark and running first. Once it's running, you're going to click manage. Then you're going to come over here on the left and click Docker manager. And then we're going to click compose right here. And then we're going to click on one click deploy. Now there are so 388 different ⁓ tools and agents and software products that you can deploy on your virtual private server at the time of recording. But all we're going to do is we're going to search Hermes agent. We're going to select. And this is where you're going to want to set an admin username. So this is essentially how you log in to your Hermes agent. to make changes in the future. So in this case, you can just leave the username as is. I'm going to set a password that I know is secure and that I won't forget. And then we're going to click deploy. So once we click deploy, you see it's showing created, status is created. We're going to give it about 30 to 90 seconds for that to change to a green check mark and show running. So I'll come back once it's running and then we'll dive into the actual setup. All right, so about two minutes later, you're going to see it's now running. says product project deployed there at the bottom. So to set it up, we're going to click the open here. Now that's going to bring us to the sign in. So this is where we're going to enter the credentials that we set earlier when we deployed our server. And then we're going to sign in that way. And first thing it's going to ask us is how would we like to set up Hermes? So we want to do the quick setup. We're just going to click the space bar. And in this case, I'm going to choose to set it up with the OpenAI models because most people watching this probably have a chat GPT subscription. It's probably the easiest to set up. So we're going to do OpenAI codex. And in this case, it's telling us, hey, to set this up, you go to this URL in your browser, you enter this code. And as soon as you do that, it will have connected your OpenAI account. So I'll be back as soon as that's done. All right, so all I did was enter this code in the tab that popped up when I clicked this link and it said login successful. Now it's going to ask me what default model that I want to use. ⁓ 5.5 is the newest model. It's also the most expensive. So in this case, let's do 5.4. And now it's going to ask us if we want to set up messaging, which I highly encourage you do during this phase. So we're going to set up Telegram, which is going to be the easiest way to communicate with your Hermes agent. So let's do set up now recommended. And in this case, let's do Telegram. And now it's going to ask us for our Telegram bot token. So I'll show you exactly how to get that. So what you're going to do is you're going to pull Telegram up and then you're going to search for at bot father. And then he's going to show up right here. We click open and it's going to ask us ⁓ if we want to create a new bot. So we're going to click create a new bot. In this case, the name, we're just going to call it Corey Hermes. The address we give it, let's just give it Corey Hermes. And it's got to end in bots. We'll do Corey Hermes bot, create bot. Right there, it is giving us our ⁓ bot token. So we're going to copy that. paste it into Hermes, click enter, and now it says Telegram token saved. So in this case, it's asking if we want to give a specific user ID in order to lock it down to where only one person at a time can talk to our bot. In this case, we're just gonna leave it empty for open access because this is just a test. Now it's saying what is the channel ID, so we're gonna go and get that. Now we don't actually have to set a home channel ID for now. If we leave that empty, we can set it later just by doing slash command set home in the DM with our bot, which essentially sets your DM conversation with your bot as the home channel. And that's probably the easiest way to go about it. So we simply click enter and our setup is complete. So now let's pull up our Telegram bot and chat with our Hermes agent there. All right, so now that we've got our Hermes agent set up in Telegram, let's ask it. It should have the LLM wiki skill. installed out of the box. So let's just ask it, do you have the LLM wiki skill currently installed? All right, so now it's saying, yes, it's currently installed and available as the LLM Wiki skill under the research category. I'm gonna say, great, let's set up our Wiki. What do you need from me to start? Now I'm assuming it's gonna tell us it needs a data source or a raw file so that it can start to essentially create it and file that away. So I will show you guys exactly what we do to quickly and easily. ⁓ Pull files from the internet whether those are tweets or articles or websites or whatever and easily hand them directly to our Hermes agent So it's saying that it doesn't find an existing Hermes config or wiki directory, which is perfect That's true because this is a brand new agent and it shouldn't have one All right. So I asked that I said if I can hand you a file Can you begin creating our wiki and it said if you send me a file I can use it as the first source and begin creating the wiki from it but it still needs the wiki's domain and purpose and the storage path unless you want it to use the default, which in this case, we'll just have it use the default. So I'm going to show you guys the best way to find, or basically put files into a Markdown format so that you can easily give them to your bot. So you're going to come to the Chrome Web Store and we're going to search for Markdownload. Now it's this one here at the top. This is the logo. I've already got it installed. But what this does is it acts as a web clipper. So let's just say, for example, we go to X and let's say we want to pull ⁓ one of my articles and give that as a Wiki source. So in this case, we come to the article, we just click the Mark Download Chrome extension, and then we can literally click Download Markdown. And what that's going to do is that's going to turn the content on the page into a Markdown file. So now let's pull our Telegram back up and let's give it... that markdown file. So I'm going to do that real quick. Pull it in here. So I gave it the markdown file. That's going to be the first source inside of our Wiki. And let's see what Hermes has to say about that. Most likely next it's going to ask the Wiki's domain and purpose. And then we're going to clarify that the storage path is the default that we want to use. So after about a minute, it says that it has seeded the Wiki structure. Now it's going to ingest our file as the first raw source and then create the first knowledge pages from it. So once it has done that, we are going to test it by asking it something along the lines of like, hey, let's query the knowledge base and give me a decision matrix, any sort of decision matrix in the knowledge base. so theoretically, it should pull this specific prompt here that we gave it from the raw file. And if it can query that and give us just that answer, then that's how we know it's working properly. All right, so what I did is I asked that I said, hey, can I query the Knowledge Base now based on that initial file? And the answer that it gave me, this is something important to keep in mind. It says, yes, we can query it now, but right now the Knowledge Base literally only contains that one file. So the answers are gonna be limited to what was in that file. And it looks like it... created seven summary slash concept pages based on that initial file. So this is important to keep in mind that this knowledge base is going to be the dumbest it's ever going to be on day one. But if you continue feeding it, if you continue giving it, ⁓ you know, knowledge base files and reference files and information over time, it's going to get smarter. This is a compounding asset that on day one is useful, but on day 90 is incredibly useful. So let's just say query. the knowledge base to give me a decision matrix that it contains. Right, so again, the goal here is to see if it's gonna pull this specific decision matrix that we know was in the reference file, and if it can do that correctly, then it is set up properly, and we know that we can just start feeding it all of our, any information that we come across, and it's going to start to build it out into a Wiki-style knowledge base. It's going to automatically synthesize what we give it. tag it properly, organize it properly, and retrieve it as needed. So in the future, if we have a question about anything, we can ask the knowledge base and the Hermes agent is going to know exactly what we're looking for. And it's going to be able to go find us an answer based on any references or any information we've given it up to that point. All right, so it looks like it passed the test. So I asked it to give me a decision matrix that it contains. I purposely kept it little vague to see what it would give us. And it looks like it pulled the exact prompt from the reference file correctly. So it says, the knowledge base contains this decision matrix prompt from the raw source, it reflected it in this Wiki page that it created here, right? And then it gives us the exact prompt that it pulled from the source. And then it takes it a step further and says, well, hey, if you want, I can also turn this into a filled in matrix for a real decision you're facing. I can extract all 35 prompts from the reference file into individual queryable pages, or I can create a cleaner prompt library section in the Wiki. So it's doing exactly what we set it up to do. The only thing we need to do from here on out is continue to feed it those reference files using the Mark download Chrome extension that I showed you guys earlier. Literally all you do is you download a source as a Markdown, drag and drop it into your Hermes chat in Telegram. tell it to add it to the Wiki. That is all you have to do. So based on what we learned here, if you guys want to quickly and securely deploy your own Hermes agent, be sure to use my link. We're gonna put that in the description and in the show notes, but you must use the code, COREY10, which will give you 10 % off any of the Hermes agent VPS plans. So hope you guys enjoyed it and also be sure to grab the LLM Wiki setup cheat sheet also in the description or the show notes, which will walk you through. very quickly, step by step, how to do everything that we did in this video. Thanks for watching and I'll be back soon.