How this site works

This site is built on a simple but powerful idea: a living wiki that an AI helps maintain, but that I edit and control. It’s inspired by Andrej Karpathy’s “LLM Wiki” pattern and the way digital gardens work.

The three layers

The garden is built in three layers, and the difference between them is who’s allowed to touch what.

GARDEN / EDITING RIGHTS

Different layers. Different permissions.

The schemaAGENTS.md: conventions, linking, loggingGOVERNS THE AI

Raw sourcesArticles, notes and ideasAI READS ONLYimmutable inputs

AI drafts

The wikiInterlinked published pagesI APPROVE

I set these editing rules in AGENTS.md: raw sources stay immutable, the AI drafts pages, and I approve changes.

Why it’s different from a normal blog

A blog is a stream. A wiki compounds: pages get revised as my thinking changes, contradictions get flagged, and connections between ideas become as valuable as the ideas themselves. Nothing is re-derived from scratch; the knowledge is kept current.

Who’s in the loop

I am. 🧑‍✈️ I curate the sources, ask the questions, and approve edits (every change is a git commit, like Wikipedia’s edit history). The AI does the bookkeeping no human enjoys: summarizing, filing, and keeping links consistent.

It’s all open

• The content and the renderer (Quartz) live in one public repo: FVossebeld/FVossebeld.github.io.
• Every page has a full version history.
• Anyone can propose an edit via pull request; I merge what I agree with.

How a page gets published

PUBLISHING / REVIEW GATE

Review comes before publication.

1. Writemarkdown
2. AI draftson a branch
3. Review and mergehuman review
4. GitHub Actionbuilds the site
5. Site is live

I review the branch draft before merging it. The publishing workflow then builds the Markdown pages and deploys the site.