I see many people recommending #obsidian as THE solution for your “AI brain” / “storage for AI data” / “AI OS” etc. It’s a great note-taking, don’t get me wrong, I’ve been using it for a while now (9 months and counting).

Meanwhile. I’ve tried the popular MCPs, CLIs… to make my Claude cooperate with it more smoothly. Am I still using it? Partially yes… What does that mean?

• Yes, they work together.
• Yes, it can save you precious tokens.
• Yes, it can help you build a graph-like knowledge structure that can be queried.
• Yes, it can save you from repeating definitions of all your projects,…

As always “CAN” doesn’t mean that it “WILL”. So let’s continue with the story.

What everyone keeps posting

Maybe it’s just my social bubble but I keep seeing these ideas in my feed across all my social media:

Drop everything into your Obsidian, so #claude can use it.

AI OS <insert any name of fictional AI> build with Fable 5 but usable with any model.

ETC…

Mostly I found empty promises ending with “comment <whatever> and I’ll send you the GitHub repo”. No, thanks. Only once did I find a direct reference to at least a structure that might be worth having in your Obsidian. A “folder” structure. So it basically doesn’t help, but it at least gave me an idea of how someone else thinks about it.

Four flaws I found in a few weeks

I like experimenting with ideas, so I started building my own Obsidian Vault with informations collected from my work, my social life and my ides. It took just a few weeks to uncover main flaws in what initially inspired me:

1. No / Poor metadata model. This is one of the keys, how to make it work properly. You need to have right metadata, so you can find your starting point as soon as possible.
2. Distant nodes. If a node is far away (meaning a lot of nodes must be loaded in between to get the full connection), it can easily cost you many more tokens and lead to more errors, because the chance that Claude takes a different path along the way keeps rising.
3. Poor graph structure and imbalance. If it’s built like multi-star / multi-snowflake schema, it won’t give you an effective way, to query it. Same if everything is connected to everything.
4. Claude will find the way. Of course it will find a way, how to answer to your question. Will it be deterministic and stable every time? Probably not… usually it’s not.

Ok, for all the AI fans who now want to tell me that I got the whole idea wrong and that I need to build my agents with proper development patterns to make them more deterministic: you are right! I’m sharing my observations, not the state where I am now.

Rebuilding it as a weighted graph

I had to dig up some memories from university, where we discussed graph design, shortest paths, node centrality and yes, even Hamiltonian paths. I remember precisely why my brain hurt so much back then. But after some tests and calculations, I found out that my Obsidian design was simply… terrible. So I had to start from scratch and rebuild the whole structure into a functional, weighted graph with a few internal helpers that navigate my algorithm (yes, I’m not saying Claude anymore) in the right direction.

Privacy is another layer

Privacy is another layer that I had to add. If Obsidian is supposed to be my main note-taking tool, then I also want to keep notes there that are a bit more private than the rest. And I need to make sure that AI will NOT read those or at least make it really hard. Do I create a separate vault for them, or are there ways to prevent access? In many tools like Claude Code you can define allowed and excluded paths, but once your request differs slightly from the standard one, there is a chance that Claude will find a way in. Read / grep / osascript / a Python script… and you need to make sure you’ve blocked all of it properly. That’s where another round of fun begins. It’s also one of the reasons why I had to rebuild the whole reading layer of my algorithm around a heuristic decision tree that decides when Claude actually gets involved, and instead of raw access feeds it already extracted information.

So, is it worth it?

• Am I still using Obsidian? Yes!
• Am I still using Claude with Obsidian? Yes!
• Do I need to rebalance my graph from time to time to keep it effective? Oh yes.

Is Obsidian the self-solving answer to the whole “AI brain” problem? NO! It’s a tool like any other. It serves its purpose, and it’s on you how you will or won’t use it. The same goes for Claude, or any AI.

How does your Obsidian Vault handle it? Or are you still in the “drop everything in and pray” phase?