Knowledge graph note-taking, in practice.
Knowledge graph note-taking is the habit of linking pages instead of nesting them in folders, then navigating by following those links. The graph is computed from the links you make while writing — not assigned by hand.
What it actually means
In a folder-first system, the path to a note is its address — and only that path. In a graph-first system, every link is also an address. You don't have to file each note in exactly one place; you just write the links you naturally want, and the graph forms itself.
The basic moves
- Write atomic notes — one idea per page. Easier to link, easier to reuse.
- Use [[wikilinks]] when you mention another idea. Don't pre-build folders; let the graph emerge.
- Skim the backlinks panel on important pages to see what else you've been thinking about.
- Pan the graph view when you want to see neighborhoods you forgot existed.
- Periodically promote frequently-linked notes into "maps of content" — index pages that link out to a cluster.
Why this works
- Folders force you to pick one home for a note. Graphs let it live next to all its neighbors.
- Linking on write is cheaper than filing later.
- The graph reveals structure you didn't consciously plan — useful for research that crosses domains.
- Modern search + a graph view together cover the cases folders are actually good at.
MindWiki's implementation
- [[wikilinks]] are first-class in the editor. Backlinks are computed automatically.
- Graph view on macOS pans/zooms across the whole vault.
- Hybrid search (keyword + vector) means you can find notes by half-remembered phrases.
- MCP/REST API exposes the graph via the mindwiki_graph tool / GET /vault/graph endpoint — your AI can navigate it too.
Frequently asked questions
Do I still need folders?
You can use folders, just don't depend on them as your only retrieval system. Folders are useful for projects with finish lines (active/) or capture inboxes (capture/). Link-first notes work best for ideas and reference material.
How is this different from tagging?
Tags are one-dimensional labels. Links are directional and carry context — "X relates to Y because…" — and they show up in both directions automatically.
Will AI clients use the graph?
Yes — MindWiki exposes mindwiki_graph and mindwiki_similar tools over MCP, and there's an HTTP equivalent. Your AI can walk the graph when answering, not just keyword-search.