the vault · a living knowledge base

Your knowledge,in plain text.

A schema-disciplined Obsidian vault an AI keeps tidy — every note structured, linked, and findable. A second brain that grows itself.

A personal system I run — not a product.

01the problem

A pile of notesisn't a brain.

A second brain isn't a tooling problem — it's a discipline problem. Structure is what makes notes compound, and structure is exactly what people stop maintaining.

the idea

Plain text in.Structure, links,and memory — out.

Just Markdown files in folders — nothing proprietary, yours forever. What makes it a brain is the contract every note obeys — frontmatter, the right folder, links to its neighbours — enforced on every edit. Structure never decays; it accrues.

what the discipline buys
up to80%

fewer tokens per session — an assistant reads a few index files, not the whole vault.

links itself

it wires notes together, surfacing connections you'd never have drawn yourself.

02structure & schema

Browse it likeyou'd open it.

The vault as you'd actually open it — the fixed 00–90 skeleton, real notes inside. Open a note to read it; click any frontmatter field to see what it buys you.

apollo.md20-projects/apollo.md

Project Apollo

The main build for the quarter. Scope, milestones, and open

questions live here.

Milestones

M1 — prototype
M2 — first user

Grew out of kickoff

Owned with maya

00-inbox is the only doorway — captures land there, then get a real home. Every other folder has a hub note indexing what lives inside.

03the graph

The whole vault,as one brain.

Every node is a note; every line is a wikilink. Clusters are folders; the bright god-nodes are the hubs everything hangs off. Hover a node to light its links, or isolate an area.

areas

Every line is a wikilink — many drawn by the assistant, connecting notes you'd never have linked yourself. As the vault grows the graph densifies: knowledge compounding. Drag to orbit it.

04in practice

The difference,in a session.

Same question, same codebase. The only variable is whether the assistant has a structured vault to read.

without the vaultno context
> where does the Apollo project stand?
No context on this repo. Let me look around…
$ ls -R
$ cat README.md package.json src/**
$ grep -ri "apollo" .
reading 38 files…
…still reading…
⚠ partial answer — context window filling up
0 tokens·~18s
with the vault3 files
> where does the Apollo project stand?
Reading me.md · memory.md · apollo/context.md
✓ full context in 3 files
Apollo — build phase. 2 open questions,
blocked on the API key.
↳ related: [[kickoff]] · [[deep-work]]
notes you saved 3 months ago
0 tokens·~2s
−80% tokens · same answer, instantly

Illustrative — token counts vary by repo and task.

05how it stays clean

The AI isthe librarian.

The discipline is enforced, not remembered — every edit runs through rules an AI never tires of, so a vault that's tidy today is still tidy in a year.

see it run

Drop a file in the inbox — watch it file itself.

  1. 1Drop a source
  2. 2Claude reads · classifies · links
  3. 3It files itself
  4. 4The brain compounds
feed it
00-inbox/watching

Drag a source here

or tap one to feed it

claudefile-inbox
claude > watching 00-inbox/ …

Every drop compounds the brain — schema’d, linked, and queryable the moment it lands.

0 filed

and it audits itself, on a cadence

weekly reviewmonthly rolloverpost-move link auditstale sweep

It runs on a schedule — not just when you capture — sweeping the whole vault, fixing drift, and reporting what it changed. That's the difference between a folder and a second brain.