How My Obsidian Vault Became a Feedback Loop

About a year ago I started seriously using AI, and I ran into a problem almost immediately: I would reach a conclusion, a solution, a good idea in a chat, and then it would all just... disappear the moment I started the next one.

Every new chat meant restating myself. Re-explaining the context. Rebuilding the same mental scaffolding over and over again. It felt like having an endless stream of good ideas, but each one lived inside a temporary chat window with amnesia built in.

I'm a DevOps engineer. We automate things for a living. So the obvious next question became: how do we automate memory?

Enter Obsidian.

Obsidian is, at its core, just a folder full of Markdown files, and that is exactly why it works so well for this. Markdown is easy for humans to write, easy for machines to parse, and future-proof enough that I don't feel like I'm locking my brain into some SaaS startup that will pivot to "AI for pets" in eighteen months.

So the first version of the system was very simple. I'd have a good conversation with an AI, and once we'd reached something useful, I'd ask it for a checkpoint:

> Checkpoint this conversation in Obsidian markdown format so I can continue in a new chat.
>
> Keep it compact but preserve:
> - current context / situation
> - key decisions made
> - open questions / next steps
> - any important preferences or constraints
>
> Format it cleanly for reuse as a prompt.

Then I'd copy that into a note in my vault, and that was that. The knowledge was no longer trapped inside one model, one tab, one session. Next time I wanted to continue, with the same AI or a different one, I could just paste the checkpoint back in and keep going.

That alone was a huge unlock.

But then real life kicked in. Sometimes I'm doomscrolling ChatGPT late at night on my phone, or I'm on the couch, or away from my laptop, and my vault lives on the laptop. Which means the note-taking system only works when I'm sitting at the correct machine like a civilized desktop user from 2007.

So, naturally: automate harder.

Enter Syncthing.

Now the vault syncs between my always-on server, my laptop, and my phone. I can write notes anywhere and have them show up everywhere. Suddenly the whole thing stops being "my notes on one machine" and starts becoming infrastructure.

And once it becomes infrastructure, the next step is inevitable: maintenance.

So I vibecoded a daily maintenance script. For new notes, or notes touched in the last day, it cleans up formatting, adds or updates frontmatter, and uses a local self-hosted model to suggest tags. No more "I'll organize this later" pretending. Later now actually happens.

At that point the system already felt useful, but also slightly annoying in the way every useful system is annoying: once it proves itself, you immediately start seeing the next manual step that shouldn't exist anymore.

So we added weekly maintenance too.

Every Sunday, the system looks at the last week's daily notes and turns them into a weekly review. What's the overarching theme? What went well? What needs improvement? What still needs to roll over into next week? It's the sort of reflection habit I always liked in theory and rarely did in practice, because sitting down to review a week sounds great right until Sunday evening actually arrives.

Now the machine nudges me into doing the useful version of it.

And then there was the inbox problem.

Because once you get used to capturing everything, you start capturing everything. Article links. Random tasks. Half-baked thoughts. "Look this up later." "Buy this thing." "Maybe blog about this." If the capture system has any friction at all, I just won't use it, so the answer was to make it stupidly easy.

So in the root of the vault I have a quick dump note. That's the scratchpad. I throw anything in there, in whatever shape it arrives. Then the daily automation picks it up and processes it into something more structured: quick notes, categorized links, fleshed-out tasks, tags, the whole thing.

This is where the whole setup started feeling less like note-taking and more like composting. You dump messy stuff in, and a day later it comes back as something useful.

Of course once you have a big enough pile of notes, a DevOps engineer's brain starts itching in a very specific way. You stop seeing "documents" and start seeing a codebase.

So then came the structural improvements. Folder hierarchies. Dashboard notes. Better links between related topics. Clearer homes for things. Not because I worship organization for its own sake, but because once the vault grows large enough, information architecture starts mattering. Claude Code, Codex, Copilot, whatever flavor of AI pair-programmer you prefer, they can all help reshape the system if you point them at the vault and tell them what problem you're trying to solve.

And once you've accepted that AI can help manage the system, it becomes pretty obvious that it can help capture into the system too.

So now voice becomes part of the flow as well. Dictate a rough thought, let AI turn it into a clean note, drop it into the right place, and move on. That lowered the barrier even more. There are so many things in life that are worth tracking but not worth opening a blank file and performing "I Am Now Writing A Proper Note" as a ritual.

Voice notes fix that.

And then, after enough months of feeding the machine, you get to the really fun part: the analysis.

Because now it's not just "I wrote some notes." Now it's: I have a longitudinal dataset of my own life, work, hobbies, spending, moods, standups, projects, and random obsessions.

I track personal expenses. I keep daily logs. I write work standup notes. I dump thoughts into the system. And after a while you can start asking interesting questions.

What have I been working on for the past six months?

Easy. I have years of standup notes.

What patterns show up in my spending?

Apparently one of them is: when I'm bored, I buy stupid games on the Nintendo eShop.

What do I use AI for, actually?

I tracked that too, mostly by dictating quick notes whenever I remembered to, and without really noticing it I ended up with two months of data without having to type a single word.

This is the part that keeps surprising me. I used to think "I should track this" and then not do it, because the barrier to start was too high. Now it's more like: "Hey, let's track this." One note becomes ten. Ten become a process. The process becomes a habit. The habit becomes a dataset. And then the dataset starts giving something back.

It's a feedback loop.

The more you put in, the more useful the system becomes. And the more useful it becomes, the easier it is to justify putting more in.

At the moment, my live Obsidian vault contains 499 Markdown notes. Across those notes I have about 219,431 words total. That's roughly 878 pages at 250 words per page, or about 731 pages at 300 words per page. Average note length is around 440 words.

In practical terms, that puts the vault at about 1.17 times the length of *Dune*, 1.05 times *Moby-Dick*, 2.47 times *1984*, and 4.67 times *The Great Gatsby*. The biggest chunks live in my work notes and project notes, which makes sense: that's where a lot of the day-to-day thinking happens.

And no, before somebody says it, I don't think this is just "AI hype" or me becoming dependent on AI.

To me this feels much closer to journaling than outsourcing my brain.

I'm still thinking the thoughts. I'm still living the life. I'm still deciding what matters. I'm just removing friction between having a thought and preserving it, and then again between preserving it and reflecting on it later.

If anything, this system makes me more reflective, not less. It lets me revisit old ideas, spot patterns, question my own assumptions, and build continuity between moments that would otherwise be lost in the scroll.

So no, I don't think AI replaced anything essential here.

It just helped me finally build the kind of personal system I always wanted: one that starts with a single note and grows, slowly, into something that can think back with me.

Anyway, that's where I'm at now. What started as "please summarize this chat so I can paste it into Obsidian" has quietly turned into one of the most useful systems I've ever built for myself, and I have a feeling I'm still only at the beginning.