Awesome Search KG

Your Obsidian Vault Is a Knowledge Graph. Here's How to Make It Think (quickly).

4 min read

Your Obsidian Vault Is a Knowledge Graph. Here’s How to Make It Think (quickly).

Author: Alexander Shereshevsky (Graph Praxis) Source: https://medium.com/graph-praxis/your-obsidian-vault-is-a-knowledge-graph-heres-how-to-make-it-think-quickly-1487614a7682

Following the yes LLM Wiki / no LLM Wiki debate, I want to share my experience from the last five years of building a second brain in Obsidian. Connecting it to Claude Code showed me what I’d been leaving on the table.

Summary

A practitioner account of connecting Claude Code to a 5,000-note Obsidian vault. The core insight: a vault maps onto a codebase almost perfectly (notes = source files, wikilinks = imports, folders = directories, frontmatter = config), so Claude Code can navigate it with the same fluency it uses for software projects. The article focuses on two things: how to configure the integration via CLAUDE.md, and what graph-quality analysis becomes possible once you have it.

The Core Insight

Most Obsidian-AI integration guides start with plugins and configuration. They miss the fundamental point: Claude Code was built to navigate codebases — read files, follow references, make targeted edits, execute multi-step plans. An Obsidian vault maps onto this almost perfectly:

CodebaseObsidian Vault
Source filesMarkdown notes
Import statementsWikilinks
Directory structureFolder hierarchy
ConfigurationYAML frontmatter

The simplest integration:

cd ~/my-vault
claude

No plugins, no servers, no configuration. Claude Code can now read every note, create files, edit content, and search with regex.

CLAUDE.md: Onboarding Documentation for Claude

When claude runs in a directory, it automatically reads CLAUDE.md. Think of it as onboarding docs for a new hire who forgets everything between sessions.

Example structure:

# Vault Context
[description of vault purpose and scope]
 
## Structure
- Projects/ - active client and personal work
- Areas/   - ongoing domains
- Resources/ - reference material, book notes, papers
- Archive/  - completed work (still searchable, not active)
- _attachments/ - images and PDFs (NEVER modify)
- _templates/   - Obsidian templates (NEVER modify)
- _ai-drafts/   - staging area for AI-generated content
 
## Conventions
- Internal links: always [[wikilinks]], never bare URLs
- Tags: hierarchical #domain/topic format
- MOCs: prefixed "MOC - " (Maps of Content)
- Source notes: must include `source:` in frontmatter
 
## Rules
- NEVER modify _attachments/, _templates/, or .obsidian/
- NEVER delete notes without explicit confirmation
- Always include YAML frontmatter: date, tags, aliases
- When synthesizing, use ONLY content from vault notes
- Output drafts to _ai-drafts/ — never directly into vault
 
## Active Context (update before each session)
- Working on: [current project]
- Open questions: [current unknowns]
- Recent focus: [relevant folders]

Three non-obvious practices:

  1. Keep Active Context fresh. Stale context sends Claude down yesterday’s priorities. Ask Claude to append a session log at the end of each session for continuity.
  2. Reference, don’t inline. Keep interests and project context in separate notes; CLAUDE.md just references them. Avoids burning tokens on irrelevant context.
  3. Include negative instructions. “Never modify _templates/” is more effective than hoping Claude figures it out.

The Knowledge Graph Structure

Every Obsidian vault is an implicit graph database:

  • Nodes = notes
  • Edges = wikilinks (directed: A → B, but Obsidian backlinks make navigation bidirectional)
  • Subgraphs = tags
  • Node attributes = frontmatter properties

Obsidian’s Graph View shows this structure — but it’s passive. It can’t tell you which notes are hubs, can’t identify clusters that should be connected but aren’t. That’s where Claude Code fills the gap.

Graph Metrics That Actually Matter

1. Centrality Ranking

Identifies true hub notes — concepts that connect to the most other concepts. Actual hubs often don’t match “official” MOCs.

Example using TurboVault MCP:

get_centrality_ranking(limit=5)

1. Feedback Loops          - 38 connections (bridges 3 domains)
2. MOC - Distributed Systems - 34 connections
3. Bayesian Reasoning      - 29 connections (bridges 3 clusters)
4. Second-Order Effects    - 26 connections
5. Margin of Safety        - 24 connections

A hub with 38 connections that has no corresponding topic/MOC note is a restructuring signal.

2. Orphan Detection

Finds notes with zero incoming and outgoing links — ideas captured but never connected. Common after migrations or rapid note-taking.

In a 5,000-note vault: 340 orphans found. Some worth integrating; most were archive candidates.

3. Cluster Analysis

Reveals natural groupings and, crucially, which clusters are disconnected from each other. Disconnected clusters with obvious conceptual overlap represent missed integration opportunities.

Investing notes and systems thinking notes were two separate islands despite strong conceptual overlap — an integration opportunity missed for years.

4. Bridge Note Identification

Finds the rare notes that already connect otherwise isolated clusters. These are structurally the most valuable notes in the vault — the ones that do cross-domain synthesis.

People

Graph View

Backlinks