Everything I know, in one place

For a long time, what I knew was scattered.

Across bookmarks. Across open tabs. In forgotten Discord chats. In GitHub repos I could no longer find. In my own head.

The problem was never learning. It was retrieving.

The starting idea

I wanted one place — a single one — where everything I know exists in a form I can query, link, and grow.

Not an app. An app can disappear, change its business model, shut down.

Files. Plain text. Markdown. UTF-8. Things that will still exist in twenty years, readable by whatever tool comes along.

The file system

The vault lives in ~/Brain/. Its structure:

Brain/
├── inbox/         Raw capture — everything lands here first
├── notes/         Permanent atomic notes
├── projets/       Active projects with a defined goal
├── ressources/    Reusable, timeless knowledge
├── archives/      Finished, abandoned, obsolete
└── _systeme/      Templates, scripts, MOCs, AI sessions

1 note = 1 concept. That's the rule that never changes.

Every file has a stable identifier: YYYYMMDD-HHmm-slug.md. The name never changes, even if the title does. It's the permanent address of an idea.

Neural connections

Folders store. Links navigate.

Writing [[note-name]] creates a link between two concepts. These links accumulate. A strong note has several incoming backlinks — several ideas pointing to it.

This network of pointers simulates the brain's neural connections. Not a metaphor — a real architecture: concepts form a graph, and the most-traveled paths reveal the most important ideas.

When I'm working on Bloko and write a balance-management pattern, the link points to ressources/competences/convex/mutation-patterns.md. When I reuse the same pattern on Pixel-Mart, same link, same destination.

Two projects. One piece of knowledge. Visible in the graph.

The 4 layers

Brain has a layered architecture:

LAYER 4 — AGENTS      Claude Code · bash scripts · Brain Assistant
LAYER 3 — INTERFACE   Obsidian · hub CLI · terminal
LAYER 2 — SYSTEM      Templates · MOCs · AI sessions · CLAUDE.md
LAYER 1 — DATA        .md with YAML frontmatter · stable naming

Layer 1 is untouchable. It's what guarantees the system survives any change of tooling.

Layers 2-4 evolve. Whatever editor or AI agent comes along tomorrow — the files are still there.

From Brain to the site

This site is an extraction of the vault.

The pipeline:

~/Brain/projets/portfolio/drafts/  →  content/<section>/  →  git push  →  Vercel

The articles you read here started as Markdown notes in ~/Brain. The YAML frontmatter is the same. Cross-links between articles get resolved automatically at build time by a remark plugin.

I don't write "for the site." I write in Brain. The site is a window onto what's already there.

The long-term goal

To precisely isolate the skills that are genuinely mine, and make them useful for clearly defined needs.

Not a generic profile. A map of what I can do, in what context, under what constraints, for what market.

The Brain Assistant — a local LLM connected to the vault via RAG — can already query this knowledge in real time. 183 files. 2,000 chunks. Queryable in natural language.

This is the first version of that map turned into a machine.

Next step: every piece of knowledge documented in the vault, precisely tagged, tied to concrete evidence — projects, commits, decisions — and isolable by domain.

Not a résumé. A system of verifiable skills.

→ The Brain → site pipeline, in detail → The AI assistant connected to the vault → How I work with Claude Code day to day