Wiki-Links

💡TL;DR

Notemd automatically adds [[wiki-links]] to key concepts in your notes. The LLM reads your content, identifies important terms in context, and inserts Obsidian-style wiki-links at each occurrence. Optionally creates concept note files with backlinks. Supports synonym suppression, link integrity on rename/delete, and pure extraction mode (no file modification). Unlike Auto Link which only matches existing note titles, Notemd uses AI to identify new concepts and creates corresponding notes. This is part of the Obsidian AI Knowledge Management Guide.

Overview

Wiki-linking is Notemd's core feature. It transforms plain text into a connected knowledge graph by:

1. Analyzing your note with an LLM
2. Identifying key concepts (terms, people, methods, theories)
3. Inserting [[wiki-links]] at each occurrence
4. Creating concept notes (optional) with backlinks

How It Works

Process

Example

Before:

Machine learning models use neural networks to learn patterns from data.
The transformer architecture revolutionized natural language processing.

After:

[[Machine learning]] models use [[neural networks]] to learn patterns from data.
The [[transformer architecture]] revolutionized [[natural language processing]].

Usage

Basic: Add Links to Current Note

1. Open a note
2. Right-click in editor → "Process file (add links)"
3. Wait a few seconds
4. Concepts are now linked!

Batch: Process Multiple Notes

1. Right-click a folder in file explorer
2. Select "Notemd: Process folder (add links)"
3. Configure:
   • Concurrency (how many files in parallel)
   • Overwrite existing links (yes/no)
4. Click Process

Selective: Link Specific Text

1. Highlight text to process
2. Right-click → "Process selection (add links)"
3. Only highlighted portion is analyzed

Notemd vs Auto Link

Obsidian has two approaches to automatic wiki-linking:

	Auto Link	Notemd
Link source	Existing note titles in vault	LLM-identified concepts in content
Can link new concepts	No — title must already exist	Yes — AI identifies concepts and creates notes
Synonym handling	No	Yes — synonym suppression
Concept note creation	No	Yes — with backlinks and dedup
Batch processing	No (single file)	Yes (folder-level)
Per-task model routing	No	Yes

Auto Link is title-matching: if a note named "Machine Learning" exists, it wraps occurrences in [[Machine Learning]]. If the note doesn't exist, nothing happens.

Notemd is AI-driven: the LLM reads your content, understands context, identifies concepts that should be linked — even if no note exists yet — and creates both the link and the concept note.

Features

Synonym Suppression

Problem: "transformer", "transformers", "Transformer architecture" → 3 separate concepts

Solution: Notemd detects near-duplicates and uses canonical form.

Configuration:

Settings → Advanced → Synonym Suppression
Threshold: 0.8 (0 = off, 1 = aggressive)

Link Integrity

When you rename a concept note:

• All wiki-links automatically update (Obsidian core feature)
• Backlinks remain intact

When you delete a concept note:

• Links remain but show as "unlinked mentions"
• You can recreate from any occurrence

Pure Extraction Mode

Extract concepts without modifying the original:

1. Right-click → "Extract concepts (no linking)"
2. Concept notes are created
3. Original file untouched

Use case: Processing read-only content or final drafts.

Concept Note Generation

Automatic Creation

When enabled (default), Notemd creates:

---
tags: [concept, auto-generated]
created: 2026-06-13
source: [[Original Note Name]]
---

# Machine Learning

A branch of artificial intelligence that enables computers
to learn from data without explicit programming.

## Occurrences in Your Vault

- [[Original Note Name#Section]]
- [[Another Note#Header]]

## Related Concepts

- [[Neural Networks]]
- [[Deep Learning]]
- [[Supervised Learning]]

Configuration

Output folder:

Settings → Output → Concept Folder
Default: concepts/

Hierarchical structure:

Settings → Output → Use Hierarchical Folders
If enabled:
  papers/my-paper.md → papers/concepts/Concept.md
If disabled:
  → concepts/Concept.md

Template:

Settings → Output → Concept Template
Customize with variables:
  {{concept}} — Concept name
  {{description}} — LLM-generated description
  {{backlinks}} — List of source notes
  {{date}} — Creation date

Advanced Options

Context Window

How much surrounding text to send:

Settings → Linking → Context Window
Options: Sentence | Paragraph | Full Note
Default: Paragraph

Larger = better accuracy, higher cost.

Minimum Occurrences

Only link concepts that appear multiple times:

Settings → Linking → Min Occurrences
Default: 1 (link all)

Set to 2 or 3 to focus on recurring themes.

Exclude Patterns

Skip certain words:

Settings → Linking → Exclude List
Example: note, idea, example, thing

Prevents over-linking generic terms.

Custom Prompts

Override default LLM instructions:

Settings → Advanced → Custom Linking Prompt
Default:
  "Identify key concepts, theories, methods, and technical
   terms in the following text. Return as a list..."

Modify for domain-specific needs (e.g., "Focus on medical terminology").

Tips & Best Practices

✅ DO

• Process notes with >100 words — Short notes yield few concepts
• Use powerful models for better concept identification (GPT-4o, Claude)
• Review before accepting — Check suggested links make sense
• Build iteratively — Process 5-10 notes, review graph, adjust settings

❌ DON'T

• Over-link — Not every noun needs a link
• Process drafts repeatedly — Concepts may shift, wait until stable
• Ignore synonyms — Enable suppression to avoid "ML" vs "Machine Learning"

Performance

Speed

Note Size	GPT-4o-mini	Claude Sonnet	Ollama (local)
500 words	2-3 sec	3-5 sec	5-10 sec
2000 words	5-8 sec	10-15 sec	20-40 sec
5000+ words	Chunked (multiple calls)	Chunked	Chunked

Cost Estimation

Example: 1000-word note with GPT-4o-mini

• Input: ~1500 tokens
• Output: ~200 tokens
• Cost: ~$0.001

Batch processing 100 notes: ~$0.10

Troubleshooting

No Links Added

Check:

1. LLM call succeeded (Settings → Diagnostics)
2. Note has enough content (>50 words)
3. Concepts are technical/specific (not just pronouns)

Try:

• Use a more powerful model
• Increase context window
• Check API key validity

Too Many Links

Solutions:

1. Increase minimum occurrences (2 or 3)
2. Add common words to exclude list
3. Use a less aggressive model

Wrong Concepts Linked

Fixes:

1. Use custom prompt for domain specificity
2. Enable synonym suppression
3. Manually review and unlink

Links Break After Rename

This is normal Obsidian behavior.

To update all links:

1. Rename the concept note
2. Obsidian automatically updates [[old]] → [[new]]

---

Next Steps

• 📖 Concept Notes — Deep dive into concept note generation
• 🔍 Research Integration — Combine linking with web research
• 🎨 Diagrams — Visualize your knowledge graph