i built a second-brain because mine wasn't enough

The way I'm thinking about this is a stove. I have different things going on and somehow I want all four burners of the stove at the same power. The problem is that the fuel I have in me is basically the same every day.

Or maybe the more direct way of saying this is: i have a limited number of brain cells that i can spend every day

I don't mean that as an actual number. It's just how I think about my attention. If I'm looking at two things, both of them are going to get a limited amount of it. They don't even need to be happening at the exact same time. I still spend something when I stop one thing, then try to load everything about the other thing in my head again.

context switching itself takes a lot of energy and brain cells and stuff.

Then I started thinking about engines instead. The fuel stays the same, except some of it is already gone before the engine starts because I had to get all the context back. So I want each engine to do more with whatever reaches it.

in aerospace this doesn't make sense but here it does

I know I'm mixing a stove, engines, power, and fuel now. This isn't the physics of a rocket. I don't think the analogy survives if I keep pulling on it. But for attention, this is the bit I'm trying to say: what this does is make the engine itself so efficient that with the same amount of fuel it can push through more.

One of those engines for me is search engine optimization. Sometimes I'm completely inside that. Then sometimes I go completely into reading and researching space stuff. I can understand both of them, but moving between them means finding the material again, remembering why I saved it, and remembering what I had already understood.

So I built a second brain. Actually I built two versions of it, second-brain and second-brain-lite. They're based on Andrej Karpathy's LLM wiki idea. Both are GitHub templates, so you can use the template, put in your own sources, and initialize your own brain in Claude Code or Codex.

I started this wanting to dump my knowledge into one simple box. But there's a restriction in how this works which I think is actually good: a brain works best when that box is about one field. A Mars habitat can be one brain. SEO can be another one. Product management can be another one.

I know that's already not one box anymore. It's a few boxes. But I can jump into the one I need instead of trying to keep every field active inside my own head, so for me that's closer to the thing I wanted anyway.

second-brain-lite is the easier one to explain. It has the actual sources, a wiki, and the LLM going between them. The sources could be PDFs, docs, and stuff. As long as they're readable as text, it can use them.

The wiki part is where I always end up jumping to the Mars example. On Wikipedia you click one of those blue words and it takes you somewhere else. Our brain also works a little like that. You think of one word and that word opens a whole pandora's box of other things you already know.

Let's say I'm asking about ECLSS, the environmental control and life support system for a Mars habitat. There are so many things around it. One source might have something about carbon dioxide removal, another might explain the oxygenator, and another might have water reclamation. It can make a wiki page for water reclamation, another one for carbon dioxide, and connect them back to ECLSS.

The LLM first tries to answer from that wiki. If the answer isn't there, the lite version searches the source files using grep and ripgrep. It finds the text related to the question and answers from that. Then it creates a wiki page from what it found, which means it doesn't have to begin from the same place when I ask something related later.

the more you ask about something, any kind of thing, and the more sources you have, it writes more and more wiki pages.

And then there's another scenario. Maybe what I'm asking isn't in the wiki and it isn't in the sources either. Then it asks me if it should go to the internet. If I say yes, it does the research, brings what it finds into the sources, and can use it the next time without going to the internet again.

This is why I compare it to how a child learns. Every time I ask something, something stays. Maybe one page, maybe one more connection, maybe just something it can find faster the next time.

The other second-brain is heavier. It does semantic searching as well, so it creates text embeddings and uses a local open-source model on the computer to search by meaning. The lite one is searching the words in the source files. This one is doing more work around what those words mean.

I gave both of them almost the same 300-something PDFs and asked the same question: if I were going to build a Mars habitat, what are the things I would need to know? Some of the PDFs came from NASA, some from ESA, and some from other people.

The answers were almost the same. second-brain got me more information. second-brain-lite missed some things, but it still got most of what was there and took roughly 10% of the time. For whatever it missed, I could've just asked a follow-up question. That's completely fine for me.

The first question on second-brain took about two hours. That's also understandable because it was the first question and it had to do all of that work across the sources. After that, the next questions are basically instant.

Still, I'd want actual GPU compute for the heavier version. A gaming PC or another computer that can handle it. If the project had tens of thousands of sources and it was a major thing, I'd choose second-brain. That's when getting the extra information would matter to me more than waiting for that first search.

for most of the normal user use case i would definitely go with second brain lite.

And then I get back to the one-box thing. I said I wanted one simple box for my knowledge. But I also said a brain works best when it's about one field. So I'm saying one box and a few different brains at the same time.

When I'm working on SEO, I can be completely inside SEO. Then I can leave that brain there and go read about space. When I come back, I'm basically talking to my own self which already knows the field I'm returning to. One self is an expert in Mars habitat, one is an expert in search engine optimization, whatever it would be.