Build An AI Knowledge Base From Scratch

What if the AI already knew what you were working on? What if it remembered what you decided last week, what your board presentation covers, what your team structure looks like, because it had read your notes?

That is what happens when you put Claude Code in a folder. You do not need to know how to code, and you do not need a plan for what goes in the folder. You just need something on your mind, and about ten minutes.

Step 1: Install Claude Code

The technical bit: open your terminal. On a Mac, search for “Terminal” in Spotlight; on Windows, search for “PowerShell”. This is the text-based interface where you type commands. Run one line.¹ You will need a Claude subscription or Anthropic API account.²

curl -fsSL https://claude.ai/install.sh | bash

This downloads and runs the Claude Code installer. It will not touch anything else on your machine.

If this already feels like too much and you have access to a Claude Max plan, use Cowork³ instead, which is in research preview at the moment. It will let you work in a local folder of files without touching the terminal. Claude Code has some powerful features that Cowork does not have yet, and they could well be useful as your knowledge base grows, but Cowork will continue to improve and is a perfectly good starting point.

Step 2: Create a folder and give Claude some instructions

Paste these commands into the terminal. This will create a folder in your home directory called “notes”. You can also just create a folder called “notes” in your home directory using Finder on a Mac or Explorer on Windows, if you prefer.

mkdir ~/notes
cd ~/notes

Now create a file called CLAUDE.md in the folder. This is a set of standing instructions that Claude reads at the start of every session. Open it in TextEdit on a Mac or Notepad on Windows and paste this in:

# My Notes

This is my personal notes folder. I use it to think things through
and keep track of what I'm working on.

When I share thoughts, you must save them as a note with a descriptive
filename, one thought at a time.  Use everything in this folder as context when
I ask questions or add more thoughts.  Lean towards asking me thoughtful
questions one at a time to help me think.  Pay attention to how I like my notes
organised and update these rules to match.

You can adjust this slightly if you want, and it is not strictly needed, but will help Claude get what you intend to do slightly faster.

Now run claude and hit enter. Claude reads the CLAUDE.md and knows what this folder is for from the very first conversation. You’ll get a prompt which you can use just like ChatGPT, except with much more power.

Step 3: Talk about something real

You do not need to set up a big system to get going. Start with whatever is actually on your mind. Say you have a meeting tomorrow you have not prepared for, for which you are mulling over a decision about team structure. You need to reply to an email and you are not sure what to say.

Tell Claude about it. Something like:

“I have got a board meeting next week about AI investment. We have spent about fifty thousand on tools and training this year and I need to justify the spend. Here is what I know so far…”

One tip: Claude Code is a text interface, so you type a lot. A dictation tool like Wispr Flow⁴ makes this feel much more like a conversation. Talk through your thinking and let it transcribe for you.

Claude will help you think it through, ask questions, and suggest angles you had not considered. When you are done, ask it to save the conversation as a note in your folder.

Step 4: Come back tomorrow

Next time you run Claude Code from the same folder, it reads everything in there. Just open up your terminal program again, type cd ~/notes and claude again. That note from yesterday is now context for today’s conversation. You do not need to re-explain who you are, what you are working on, or what you decided last time. It picks up where you left off.

Each note you add is context for every future conversation. As the folder grows, Claude reads what it needs rather than loading everything at once, so you do not need to worry about it slowing down.

Step 5: Browse your notes with Obsidian

If you want a visual companion for your notes folder, download Obsidian⁵ and point it at the same ~/notes directory. Obsidian renders markdown files with a clean interface, and since Claude Code works in plain text, everything stays in sync. Write rough notes in Obsidian, switch to the terminal and ask Claude to help you make sense of them.

Why this works

After a few weeks you can ask things like “what have I been working on this month?” or “what did I decide about the hiring plan?” and get a real answer drawn from your own notes. The questions that get interesting are the ones that connect things across weeks: “I had a conversation about team structure last month and a similar one yesterday, what changed in my thinking?” or “given everything I have noted about the Q1 figures, what should I actually say to the board?” Every note makes the whole thing richer.

Andrej Karpathy, former head of AI at Tesla, described this as the “LLM wiki” pattern⁶: instead of the AI re-deriving knowledge from scratch every time you ask a question, it works from a persistent, compounding set of notes. The knowledge is compiled once and then kept current.

Everything is plain text in a folder you own. You can open any file, read it, move it, back it up. Nothing is locked inside a proprietary format.⁷

What comes next

The folder starts with notes, but it does not stop there. In the future I will cover how to drop meeting transcripts into the folder and extract something useful from them. After that, articles, links, and other sources that flow through your working life.

For now, just start with the empty folder and whatever is on your mind today.