Subsystem deep-dive

The Brain Vault

The shareable knowledge base: an Obsidian vault organized PARA-style, version-controlled in git, with strict rules about what lands where. The thing every other layer reads from.

0
Notes
0
Sessions logged
0
Active projects
0
Git commits
The layout
PARA, with a place for everything.

A documented folder map (MAP.md) is the source of truth for where things live, so the structure survives even when Claude is the one moving files.

Projects/

65 notes

Active, outcome-oriented work: pitches, builds, decisions.

Resources/

graph-visible

Reference material: articles, news digests, podcast summaries.

Stack/

system notes

AI-stack build notes and these architecture diagrams.

Sessions/

686 notes

Session-close logs. The running record of work.

Reviews/

generated

Weekly reviews and monthly rabbit-hole sweeps.

AI/Skills/

64 skills

Source of truth for personal skills.

The routing protocol
Four streams, never tangled.

Keeping content properly routed is what stops a second brain from feeling like a junk drawer.

Project synthesis

Pitches, status, decisions → Projects/ + Sessions/. Graph-visible.

Consumed content

Articles + podcast summaries → Resources/. Lean twins are graph-visible; full transcripts stay out.

System notes

How the machine works → Stack/, AI/Skills/, MAP.md.

Identity docs

Who I am → Context/. The single canonical home; never duplicated.

Version control
Git is the memory of what changed.
Private remote. The vault pushes to a private GitHub repo: a real backup and history, not a public page.
Auto-commit + auto-push. Ending a session commits the vault and pushes it, gated by a privacy/secret check that runs before any push.
The graph stays out. graphify-out/ is gitignored. It's regenerable and concentrates distilled personal facts, so it's never versioned.
Shareable by default. Because raw personal data lives outside the vault, everything in Brain is safe to commit.
The biggest tributary
The podcast pipeline.

Most of the vault's volume is automated listening: episodes fetched, enriched, summarized, and split into graph-friendly twins.

3,530 lean summaries

Each episode gets a transcript-free twin in Resources/Podcast-Summaries/ (graph-visible), while the heavy full transcripts stay ignored.

Per-show citations

A scraper pulls each episode's reference list into the note, so sources land in the graph too.