Why I treat my vault like a database

After 18 months and 1,318 notes, my Obsidian vault stopped being a “second brain” and started being something else: a flat-file database with foreign keys.

This isn’t a tutorial. It’s the moment the metaphor shifted, and the things I changed once it did.

1 · the problem

Every note-taking system eventually becomes a retrieval problem. You write a note. You write another. You forget the first one exists. The system that was supposed to make you smarter is now a pile you have to grep through.

I hit this wall around note 800.

2 · what the vault is now

A directory of ~/.vault with one file per note, a flat namespace, and frontmatter for every column you’d ever query on.

# note.md frontmatter
id: 202604180921
tags: [k8s, networking, eBPF]
refs: [[202603220812]], [[202602100731]]

Every note has an ID, tags, and references. Those three things, plus a vault-wide search, are the whole “database.”

3 · keys, indexes, joins

The ID is the primary key. Tags are a secondary index. refs are foreign keys — except they point to filenames, not row IDs.

ripgrep is the query engine. A join looks like:

rg --glob "*.md" "refs:.*202603220812"

That finds every note that references the note with ID 202603220812. That’s a JOIN.

4 · queries I actually run

# "what do I know about cilium?"
rg -l "cilium" ~/.vault | wc -l

# "what did I write about k8s networking in March 2026?"
rg --glob "*.md" -l "k8s" ~/.vault | xargs grep -l "networking" | xargs ls -lt | head -5

# "what are my most-referenced notes?"
rg --glob "*.md" -oh "refs:.*" ~/.vault | grep -oP '\[\[\K[^\]]+' | sort | uniq -c | sort -rn | head -10

None of this is fancy. All of it is fast.

5 · what I’d do again

Flat namespace. No folders. Tags are your index.
Numeric IDs based on timestamp. Unique, sortable, grep-able.
Backlinks via frontmatter refs. Obsidian shows them automatically.
One note per concept. Not one note per source.

6 · what I wouldn’t

Rely on Obsidian’s graph view as the primary interface. It looks beautiful. It tells you very little.
Use Dataview for anything you could just grep for. The abstraction cost is real.
Write notes longer than 300 words. Long notes are a smell that you haven’t atomized the concept yet.

The vault isn’t a second brain. It’s a database. Once I accepted that, I started querying it properly.