graphkasten

Graphkasten is an AI-native evolution of the Zettelkasten: graph-centric note-taking where connections matter as much as the notes themselves. Built for Obsidian, usable with any agentic system.

Most vaults treat Graph View as a gimmick. Here everything revolves around it, a global graph keeps the vault auditable, a local graph gives instant neighbourhood around any note. AI does the indexing; the graph makes human review cheap.

See it live: graphkasten.com

This repo ships a working vault — 462 notes across computer science, finance, and personal knowledge management — so you can read the conventions at scale before adopting them.

Three Components

1. Design Pattern — four primitives

Folders (LIFT). Graph filter scoped to one folder:

Links. Wiki-links are the graph — reserve [[wiki-link]] for structural edges (MOC↔note, hub↔child). Everything else: plain text or an ` URL (clickable, no edge). Placeholder links to unwritten notes are encouraged. Start from folder hierarchy, then rewire to match how you actually think.

Tags. Shallow on purpose — one level of nesting max (gen_ai/agents, never gen_ai/agents/x). Each tag has a colour, so the graph clusters by topic.

Frontmatter. One unified schema across the vault (declared in types.json), filled from Obsidian templates per note type.

2. Review Process

Medallion-inspired: Inbox → AI indexes → Graph View → Human review → useful graph with correct tags and links.