How Jesse Cunningham Uses LLMWiki with Claude Code

0:00There's tons of money to be made right now with AI, no doubt, but most people are using it wrong and it's actually costing them way more money than they realize. I'm going to give you a prompt in this video based upon what Google just announced recently. A lot of people are using this new method when they operate any AI system. Problem is your AI has amnesia and it's costing you an absolute fortune. Most people use AI and then they start a new session. But how do we get around that? And what's the problem? Well, it's like a tax. Every time you use AI this way, and I use

0:27Claude code mostly. I'm using Claude, I think it's the best. But if you use it the improper way, you are spending thousands of dollars. And let me tell you, as the frontier models come out and become more expensive and the token usage, etc., it's going to be a problem. This is what most people do, the Groundhog Day loop. They explain, so you teach it your business again. You work, you close the tab, and you forget about it, and the AI forgets about it as well, and that's the problem. Now, here's what Google says, AI is only as smart as the context we give it. What

0:55we're going to do here is we're going to give our AI sessions a brain, but it's going to be a brain on the side, which helps with cost and other things. So, introducing the OKF, the Open Knowledge Format, an open specification that formalizes the LLM Wiki Pattern into a portable, interoperable format. Basically, we can be more efficient, cost-effective, and we have way better sessions with our AI. Now, something to address is some people think this is overhyped and I'm going to be an actual problem for others, and we're going to get into that. But, here is what the LLM Wiki Format is and why it matters for

1:29you. So, it's five basic rules. One folder full of text files, one topic each. Each file starts with a little label. One file is the table of contents. One is the diary. Files point to each other. So, what this does is, you have a brain on the side, where you can build it over time. So, if we come back over here, I have a slide that kind of visualizes this. We're going to get into it, but I'm kind of skipping ahead.

1:53So, this is what it's going to yield. So, you have your business in the middle, right? And you have your methodologies, your pricing decisions, markets, clients. Using this method, and I have a copy-paste prompt you can use to build this out, it allows you to build over time instead of restarting every time you fire up your AI. Now, this article right here is super interesting. I want to share it. So, the open knowledge format, what Google really shipped and why the interesting question is not SEO. So, SEO is search engine optimization, basically how to rank a website on Google or on AIs. Some

2:25SEOs are thinking we can use this tactic right here, right, to to rank on Google. Some people may be proposing to put a file, like a brain, on your website so AI can crawl it, but, right, that's probably not the case. So, this concept will not rank your website directly, I have something else for you. So, the tech world built this for coders, right, techies. You know what I mean? Like, this is like very technical stuff from the outside looking in. And nobody's really brought it to the money yet. So, that's the goal of this channel, let's find the intersection of tech and money.

2:57How do we make money with this? Let's talk about it. Starting off, if you can make a folder, you can build this thing. Do not be overwhelmed. You make three folders, you give it a job description, more or less, and you can just have Claude ask you questions. Say, "Interview me to help me build this thing." Right? And then you're going to feed it one client or one project. Now, you might be asking, "Isn't this just like a GitHub repo?" And GitHub has its use case here. So, GitHub alone stores files, right? It's version control. The wiki is different. It has the rules for

3:25the AI. It has settled pages instead of raw piles, an index, a log, etc. It's like a librarian, whereas GitHub is version control. But, when you combine both of them, then you're really cooking with fire. Okay, meat and potatoes stuff. One prompt, copy it, paste it, done. So, let's go over this line by line. You are the librarian of my business knowledge base, right? That's how we pose it as a librarian. We're building an LLM wiki, wiki, a folder of plain markdown files that that your permanent memory of my business. Set it up now by following these steps exactly.

3:55So, step one is we're going to build the vault. Create this structure in the current folder. So, you have to pick a folder for where this lives and usually that's locally. Right? Locally on your desktop, but you can also have a private repo and you can push it to the cloud and do all these things if you so choose. So, we're going to build the raw, my original sources. You may read these, but never edit, move, or delete them. The wiki, this is your territory.

4:19You write and own every page in here. Then we have the wiki index, which is a catalog of every wiki page with one-line summary of each. Then you have the log, which is append-only history. So, one line per change, newest at the bottom, and optional page skeletons for repeated formats, which is templates. When you start to combine these things, then you'll see the power of it, right? So, step two, you're going to interview me.

4:43So, before writing any rules, interview me one question at a time, eight to 12 questions in total, and find out these things. Now, you can customize these to your liking, but for this purpose, because on this channel, right, we are SEOs, typically AEOs, and we're making websites to make money for clients and ourselves. So, this is kind of how I have it structured. What my business does and who my customers are, what is this wiki for? Push me to be specific, right? What kinds of sources will I drop in the raw?

5:11Uh what I always wanted to remember, such as pricing rules, past decisions, etc., etc. Now, you can build this type of concept for an individual client. You can build it for your agency. You can build it for a specific project. I do like segmenting it out for different use cases. Now, let's talk about this, the jargon translator, because if you are watching this video, you're like, man, this is too intense for me. I'm telling you it's not. Jargon, I hate jargon.

5:35Jargon sucks. So, let's just go over how to break this down simply. Markdown and .md files are just text files, which open in notepad. Don't be overwhelmed by that. Claude.md is the AI's job description. Index.md is the literally the table of contents, right? Log.md is the diary, so what changed and when. Now, I want to get to the slide. Here we go. Let's come over here. So, we're going to talk about Claude.md next.

6:01Very, very powerful, very, very important. It's the AI's job description. So, it gets read first every single session. You cannot forget this. Every time, right? Every time the AI is going to do a thing, it's going to read Claude.md. So, it's like hyper important, right? And it turns a generic AI into your employee. So, you can have either a librarian or a junk drawer. And that's the difference when you do this properly. Now, let's take our time with this one because Claude.md is intense.

6:31It is important. So, using my answers, write Claude.md in the root as your permanent rule book, and it must cover the purpose, folders, page formats, ingest workflow. This is a cool one. So, when I say, "I added a new source, read it, and update the wiki." You will read the source from raw, create or update the relevant wiki pages, add links between related pages, etc. So, the whole point is we're building a system that can grow over time. This is not like a one-and-done.

6:59The whole goal is to have a conversation with an AI in a specific like conversation bubble that can grow, and it knows what we did yesterday, last week, last month, right? Conduct. We're going to cite the source file for every factual claim, and maintenance. So, when I say run a lint, sweep the wiki for contradictions, stale claims, broken links, and orphan pages. Report findings before fixing anything. Now, this next part a lot of people skip, but I don't think you should. I think this is really important. So, a repo, a GitHub repo, private repo is a time machine. It's version control for your folder. This is

7:30what it looks like in real time. So, day one, wiki is born, right? Day four, you're added client, you have your data, you're like ready to rock. Day nine, we do a pricing update, we do any update. And then day 12, there's a problem. The AI made a mistake. It made a mess because as we all know, AI can mess up. What do you do if the AI in your conversation screws all this up and you don't have version control? You're screwed. And this makes sure you're never screwed, to put it in simple terms, right? What you can do is go back

8:00in time and say, "All right, let's go back to day nine snapshot and let's just restart the darn thing." I've had a lot of people talk about when AI goes into a folder, screws things up, and they're like, "I have no idea what to do now." When you're building a website, when you're building anything really, right? That's a problem and this solves it. So, what we're going to do, we're going to come down here in step four, we're going to say, "Make it a repo. We're going to run this thing, then make the first commit with the message Wiki Board. If I

8:23give you a private GitHub remote, add it and push." You can do a lot with this and I don't want to talk about it in the video, but basically version control, version control, is where it's at. Now, on this channel, we build what are called microsites. It's a way that we make money online building websites for local businesses. And quite literally, this is how we do it, right? We have a Claude.md. We have an Wiki LLM that can build these things. And the crazy part is, if you do it correctly, the speed at which you are going to work is way

8:52faster. Now, we've been talking about this for weeks in the community and people are building with it. So, set up an LLM Wiki directly in my GitHub repo with Claude.md schema at the root paired with Superbase for the data layer. Every Claude code session now auto-loads my framework rules and market knowledge before touching any code. That is where it's at. So, if you want to be part of the community, right? Where we talk about these things. Come check it out.

9:13Right Exp Academy, the classroom, we talk about a lot of things. Talk about a lot of things. I mean, probably too much to talk about on YouTube. And we have a community call. We have a lot going on, but basically, thanks for watching this video. Let me know what you think, right? A lot of people do it different than this. Do you do it different? Is this how we're going to do it from now on? Let me know in the comments and I'll catch you on the next one.