NOUZ — Semantic Knowledge Graph for Obsidian

One server. Three approaches. Your notes find their own place in the graph.

Why NOUZ?

You write in Obsidian. A lot. Over time your vault turns into a mess — hundreds of notes connected somehow, with no system or logic.

NOUZ fixes this. It reads your notes, analyzes the content, and builds the knowledge graph itself — what belongs where, what "sign" each note has, where the semantic connections between different branches are.

No API keys needed. NOUZ works with your own embedding model.

What NOUZ Does

Builds the Graph

Add parents: to a note — and NOUZ automatically places it in the hierarchy. No manual folder sorting.

Classifies by Content

Using embeddings (local model), NOUZ understands what the note is about and assigns it a "sign" — T (technology), S (science), H (humanities) or any other you define.

Finds Hidden Connections

Notes from different branches can be semantically close. NOUZ finds these bridges and suggests linking them.

Tracks Knowledge Drift

Over time the content of a branch changes. NOUZ shows when the actual composition of notes diverges from the declared topic — this is core_drift, a signal that the branch has grown beyond its domain.

Three Modes for Different Tasks

Start with LUCA — just connect and add links. Move to PRIZMA or SLOI when you want semantics.

Quick Start

# Install via PyPI (recommended)
pip install nouz-mcp

# Run
OBSIDIAN_ROOT=/path/to/vault nouz-mcp

Or from source:

git clone https://github.com/KVANTRA-dev/NOUZ-MCP
cd NOUZ-MCP
pip install -r requirements.txt
OBSIDIAN_ROOT=./vault python server.py

Connect to Claude Desktop, Cursor, Opencode, or any other MCP client:

{
  "mcpServers": {
    "nouz": {
      "command": "nouz-mcp",
      "env": {
        "OBSIDIAN_ROOT": "/path/to/vault",
        "MODE": "prizma",
        "EMBED_API_URL": "http://127.0.0.1:1234/v1"
      }
    }
  }
}

How It Works

1. You Write a Note

---
type: module
level: 3
sign: T
parents:
  - Machine Learning
---

Your note here.

2. NOUZ Builds the Graph

• Reads YAML → builds DAG (directed acyclic graph)
• Vectorizes text → calculates proximity to your cores
• Proposes sign, level, parents
• Finds bridges to other branches

3. Structure Emerges from Content

T (core)
 ├── TH (pattern: AI)
 │   ├── TH (module: ML)
 │   │   ├── TH (quant: neural-networks.md) — T
 │   │   └── TS (quant: transformers.md) — T
 │   └── TH (module: Ethics)
 │       └── TH (quant: ai-safety.md) — T
 └── TS (pattern: Physics)
     └── ...

The more you write — the smarter your knowledge base becomes, and the agent working with NOUZ.

How Classification Works

Cores — Coordinate Axes

In config.yaml you define 2–5 domains as text descriptions. The server converts them into reference vectors — coordinate axes in the multidimensional embedding space.

mode: prizma

etalons:
  - sign: T
    name: Technology
    text: "programming software architecture machine learning neural networks"
  - sign: S
    name: Science
    text: "physics chemistry biology mathematics formal logic theorems"
  - sign: H
    name: Humanities
    text: "philosophy psychology sociology history literature ethics"

**On core

... View full README on GitHub