Skip to content
const portfolio = { loading: true, status: 'initializing...'};
0%

00~/renan-rios$cat ~/brain/README.md

My second brain

A second brain is a knowledge base you keep outside your head — notes that link to each other, so retrieving something depends on how it connects rather than on where you filed it. Mine is a folder of plain Markdown that AI agents read, write and cross-reference. It is the single source my design work draws from, and the reason I can answer “why did we decide that?” six months later.

101

linked notes

32

folders

489

connections

32

pages kept private

01~/renan-rios$define second-brain

What a second brain actually is

Not a note-taking app. A structure that makes retrieval a property of the notes rather than of my memory.

Most notes die where they are filed. You write something down, put it in a folder, and never find it again — because finding it depends on remembering the folder you chose months ago.

A second brain inverts that. Notes link to each other, so any one of them is reachable from every related idea, and the connections carry as much meaning as the text. Ask it a question and you follow edges rather than guessing at a hierarchy.

Mine is a folder of plain Markdown files. That is the whole storage format — no database, no vendor, readable in any editor in twenty years. What makes it useful is that AI agents read and write it directly: they pull a PRD apart into linked notes, cross-reference a new meeting against six months of decisions, and flag when two pages contradict each other.

It is the single source my design process draws from. When someone asks why a decision was made, the answer is a note with a date and a source on it — not a recollection.

02~/renan-rios$npm run brain -- serve

PINE

PINE reads the vault and renders it as a graph: every note a body, every link a line. It runs locally with zero dependencies, and the deployed copy is password-gated because the vault holds a career, not a portfolio.

Personal Intelligence & Notes Engine

built for this vault · lives in the same repository

Obsidian is where I write. PINE is where I look. It indexes the vault, scores every note against every other for topic similarity, and draws the result as a graph you can fly around — so the shape of what I know becomes something I can actually see, including the parts that are thin.

Full-text + TF-IDF searchSimilarity graphConfidence ringsPrivacy guardZero dependencies

03~/renan-rios$open ./viewer

Two ways to read it

The same vault, drawn twice. One view shows mass and orbit, the other shows lineage.

pine — cosmos

Cosmos

The vault as an orrery. The root sits at the centre and each folder orbits it, sized by how much it holds — so the shape of the graph is the shape of the career.

pine — constellation

Constellation

The same data as a branching tree. Folder hierarchy only, no similarity edges — useful when I want the structure rather than the associations.

pine — a note, opened

A note, opened

Click any node for its provenance: where it came from, when it was captured, what it is tagged with, and every note it connects to.

04~/renan-rios$cat AGENTS.md

The rules that keep it honest

A knowledge base an AI can write to needs rules an AI cannot talk its way around. These four do most of the work.

Every claim carries its confidence

Each page is marked EXTRACTED (transcribed from something I wrote or said), INFERRED (an agent derived it, unverified) or AMBIGUOUS (sources disagree). The graph rings unreviewed pages in amber and contradictory ones in red, so I can see at a glance which parts a machine invented. Nothing enters unmarked.

Links are typed, not decorative

A wikilink can carry a relationship — worked_at, evidences, supersedes, contradicts. That turns the vault into something an agent can traverse deliberately instead of grepping, and makes a contradiction between two notes a thing the graph can show rather than something I have to remember.

The privacy guard fails closed

Anything inside a folder named private/ is withheld from the build, alongside a regex scan for personal data. The builder then re-reads its own output and deletes the whole site if a withheld page made it through. A published page that merely quotes a private one is caught too.

A fact lives in exactly one place

Duplicating prose across a CV, a portfolio and a notes file is how a wrong claim survives being corrected. Reusable prose lives once, and a lint pass checks the numbers in the CV against the pages that source them.

05~/renan-rios$cat tools.json

What it runs on

Deliberately boring. The vault outlives every tool that reads it.

Markdown

Plain text files. No database, no lock-in, readable in any editor.

Obsidian

Offline-first editor for the vault when I am writing rather than querying.

TypeScript

The indexer and viewer. Node strips the types at runtime — no build step.

Zero dependencies

Nothing installed. The whole tool is source files in this repo.

Claude · Copilot

The agents that read the vault, write back to it and check it for contradictions.

GitHub Actions

Rebuilds and redeploys the vault on every push, and refuses to upload if the password gate is missing.

Hostinger

Hosts the deployed copy in a password-protected directory, on the same server as this site.

06~/renan-rios$open https://pine.renanrios.com.br

Open the brain

The deployed copy is live and asks for a password. It is a career record, not a portfolio piece — happy to walk anyone through it on a call.

69 pages published, 32 withheld. Anything naming a colleague or carrying personal data never leaves my machine.

Open PINE

pine.renanrios.com.br · password required · opens in a new tab