as a programming tool. As a programming tool, it’s really amazing and does an incredible job, and has probably made me a multiple times over more efficient engineer.
However, Claude Code can also be used for a lot of other super-useful application areas. You can use it for tasks such as creating presentations, which is very efficiently done by writing code, for example Python or LaTeX. Furthermore, you can apply Claude Code to use cases that are completely not related to coding, such as organizing your sales outreach or creating a knowledge base, which is what I’ll cover in this article.
An LLM-powered knowledge base is an incredibly powerful concept that can make you far more efficient simply because you can fetch relevant information faster. I’ll cover how to set this up, why you should set it up, and how you can get the most out of your knowledge base.
Why set up an LLM-powered knowledge base
First of all, I’ll cover why you should set up an LLM-powered knowledge base. The simple reason is that LLMs work better the more context they have, and the more context you can provide to your language models, the better they will be able to solve different problems you throw at them.
Thus, storing as much information as possible into a knowledge base is incredibly powerful because the LLM can access it when relevant.
I try to store absolutely everything I do into a centralized knowledge base. This could be things such as:
- the meetings I’m a part of
- notes that I think about
- mistakes that my agents make and how to avoid them
and so on.
Before LLMs, you would have thought that there’s no point in setting up such a large knowledge base with so much information, because you simply couldn’t fetch the relevant information when you needed it. For example, if you needed a note from a specific meeting, it would be a lot of work to first find that specific meeting transcript, then, in that transcript, find the specific point you’re looking for and thus get the information you’re after.
However, this completely changed when large language models became a thing, because suddenly you could easily access a vast amount of information within just a few seconds. The availability of information vastly increased overnight.
LLM’s vastly increased availability of information, which made knowledge bases a lot more valuable
To answer shortly, the reason you should build an LLM-powered knowledge base is that it allows you to fetch relevant information when you or your coding agent need it the most. The fact that you get a powerful knowledge base with LLMs makes the information super easily accessible, meaning you can store a vast amount of information.
How to set up an LLM-powered knowledge base
Now I’ll get into how you can set up your own LLM-powered knowledge base. In reality, it’s quite simple. You just need to store all the information in one folder on your computer.
In practice, however, it is a bit harder than you think because you need to make it a practice to store all the information in that location. However, I’ll take you step by step through how I started setting this up, and by focusing on it over time I became better and better at updating my knowledge base and adding all relevant information into it.
You can start by having a meeting note taker and ensuring that it always stores the notes with all relevant information, such as the attendees of the meeting, the time of the meeting, the name of the meeting, and the context of the meeting. This is information that you can fetch from your calendar, for example
Simply adding all of your meeting notes and transcripts into a, a central folder does a lot of the job.
Note also that the knowledge base doesn’t have to be a local folder. It can also be cloud-based applications, such as Notion, where you can store text files. A knowledge base doesn’t really need to be much more than that. It simply needs to store text.
Continuing on this, I also have reminders several times a week to store things in the knowledge base that I either think about, learnings from my work, or other useful information that I thought about. I try not to be or not to overthink what I add to the knowledge base. The most important is to add the information, which I do by simply prompting Claude Code:
Add <piece of knowledge> to my knowledge base
Claude Code then figures out which file or subfolder is the most relevant to contain this piece of information and stores the information for me.
Another thing I do to update my knowledge base is that I ask Claude Code to do a daily walkthrough of all of my agent interactions throughout the day. This can be both interactions that I’ve had with my personal Claude Code and that I’ve had with my OpenClaw bots or other people I’ve had with my OpenClaw bots. We discuss what went well, what didn’t go well, and it automatically stores relevant generalizable knowledge into my general knowledge folder so it’s accessible for later. This is done by simply setting up a cron job, which runs daily.
You will naturally have other workflows than I have, and thus store different pieces of information and have access to different pieces of information as well. I thus urge you to think about different knowledge that you wish you had stored, and try to store that knowledge in a knowledge base. The general rule of thumb is that you should store as much context as possible in the knowledge base; it doesn’t really matter if the context knowledge base becomes bloated, and you should try to make storing information in the knowledge base as automatic as possible
What I mean by automatic is that you don’t, for example, have to manually copy over your meeting notes into the knowledge base every time you have a meeting. That’s, of course, a lot of manual work, which for one becomes annoying to do over time, and secondly, you might forget to do it after a meeting. Instead, you should set up an automatic script or flow that stores the meeting notes in a knowledge base for you.
How to use the knowledge base
In the last section, I covered how you could create the knowledge base and store information on it. The second half of the knowledge base is, of course, how you utilize the knowledge base in practice. For this, I have two high-level topics that I’ll cover.
- Search for pieces of information whenever you personally need it
- Provide Claude Code or your other agents access to the information that you can fetch when it’s relevant for them to complete a task.
In a lot of situations, I find myself searching for a piece of information that I know I discussed in a meeting or that I thought about at a previous point in time. In these situations, it is super frustrating if you can’t access the information. I ask Claude Code to look through my knowledge base and find the answer to my question. Sometimes it doesn’t find a direct answer, but finds plausible other answers or pieces of information that are connected to my question, which is also very useful in many situations.
The second application area of the knowledge base is to give Claude Code access to it, so it can access it whenever it thinks it might be relevant. For example, if you ask it to complete a coding task, it might be useful information in the knowledge base for it. Or whenever you are making presentations, for example, you can look for previous presentations.
If the knowledge base is a centralized folder, you should ensure that you have, first of all, given Claude Code or your coding agent access to that folder.
Secondly, you need to have a user-level skill file or claude.md file so that the coding agent is aware of the knowledge base and how to access it at any point in time. It is important that your coding agent is aware of the knowledge base at any time, no matter where you prompt the model from.
Mistakes to avoid
I also want to cover some mistakes that are easy to make when making a knowledge base. The first mistake I want to cover is what happens when the knowledge base becomes outdated. Of course, there is drift in information. Your opinions on topics might change, or pieces of information might become outdated. It’s thus important to do regular check-ups on your knowledge base to look for outdated information.
This check-up can simply be a cron job that you run on a weekly basis, where you have Claude Code go through all of the interactions that you’ve had with it over the last week and compare it to the knowledge base to look for outdated information.
Another common mistake is that you don’t make the agent aware of the knowledge base when it’s running in specific folders. For example, if you only inform the coding agent of the knowledge base inside a project-level Claude.md file, Claude will not be aware of the knowledge base if it runs in any other folder. This is, of course, very problematic, since when Claude is running in that other folder, it won’t have access to all of the same information.
This is where a user-level Claude or skill.md files are useful, since they are always loaded into your coding agent, no matter which folder you run them from.
Conclusion
In this article, I covered how to build a Claude Code-powered knowledge base. This is basically a centralized storage for all of the information that you interact with on a daily basis, and it gives you a powerful advantage because you can access information more quickly, and you can search through the knowledge base using LLMs. Coding agents are incredibly efficient at searching large pieces of information, such as folders, and storing/accessing relevant information will thus be quick, even if the knowledge base is large. I urge you to try to set up a personal knowledge base. Try to store all of the information that you interact with on a daily basis into that knowledge base, and try to actively use this knowledge base both to search for information yourself and make it available to your coding agents so they can become more efficient. I believe such personal knowledge bases on your preferences and specific information will become more and more important in the future when coding agents become more powerful.
👉 My free eBook and Webinar:
🚀 10x Your Engineering with LLMs (Free 3-Day Email Course)
📚 Get my free Vision Language Models ebook
💻 My webinar on Vision Language Models
👉 Find me on socials:
💌 Substack