Skip to main content

Introduction to Notemd

💡TL;DR

Notemd (Note + EMD — Enhanced Markdown Documents) is an open-source Obsidian plugin that transforms LLM-powered reading into persistent knowledge. Unlike chat-based AI where insights vanish after the session, Notemd writes results directly into your vault as wiki-links, concept notes, research summaries, translations, workflows, and diagrams. It is built for researchers, students, and knowledge workers who want reading, research, and visual explanations to accumulate into a structured, evolving knowledge graph.

What is Notemd?

Notemd integrates 30+ Large Language Models (OpenAI, Anthropic, Google, DeepSeek, Qwen, Ollama, and more) into your Obsidian workflow to automate knowledge extraction, organization, translation, research, and diagram generation.

Key Difference: Ephemeral vs. Persistent Knowledge

AspectChat-based AI (ChatGPT, etc.)Notemd
Where results goChat history (disappears)Your Obsidian vault (persists)
FormatPlain text answersStructured files: [[wiki-links]], concept notes, diagrams
Long-term valueMust re-ask each timeAccumulates into a knowledge graph
Offline accessRequires internetWorks fully offline with Ollama

Core Capabilities

1. Automatic Wiki-Linking

  • LLM identifies key concepts in your notes
  • Inserts [[wiki-links]] at each occurrence
  • Optionally creates linked concept notes
  • Synonym suppression to avoid duplicates

2. Concept Note Generation

  • Extracts core concepts from papers, articles, notes
  • Generates dedicated concept files with backlinks
  • Customizable output paths and templates

3. Web Research Integration

  • Query Tavily or DuckDuckGo from within Obsidian
  • LLM summarizes results with source citations
  • Appends research findings to current note

4. Multilingual Translation

  • Translate selections or entire notes
  • Supports 21+ UI languages
  • Independent output language configuration
  • Batch translation support

5. Diagram Generation

  • Mermaid: Flowcharts, sequence, class, state, ER, Gantt
  • JSON Canvas: Obsidian native layouts
  • Vega-Lite: Data charts, time series, scatter plots
  • HTML / Editable HTML/SVG: Self-contained figure artifacts with semantic annotations
  • Draw.io / Drawnix artifact boundaries: Maintainer-facing export paths from the same semantic figure model
  • Circuit diagrams roadmap: circuitikz/TikZJax support is being designed around golden references, constrained prompts, render feedback, and topology/layout validation rather than raw unconstrained LLM TikZ
  • Preview diagnostics: Render artifacts can surface compile/render smoke diagnostics, and non-inline sources can be inspected without requiring a plugin-side LaTeX runtime
  • Syntax auto-fix for Mermaid errors

6. One-Click Workflows

  • Chain multiple actions into sidebar buttons
  • DSL-based workflow definition
  • Example: add-links > extract-concepts > research > diagram

Who Should Use Notemd?

Researchers reading papers and building literature reviews ✅ Students organizing study notes and creating concept maps ✅ Knowledge workers who want reading insights to persist ✅ Bilingual professionals needing translation + wiki-linking ✅ Privacy-conscious users wanting local LLM support (Ollama) ✅ Power users who customize prompts and workflows

Why Notemd + Obsidian?

Obsidian is a local-first, markdown-based knowledge base. Notemd adds AI superpowers:

  • Your data stays in your vault (not a cloud service)
  • Works offline with local models
  • Free and open source (MIT license)
  • Integrates with existing Obsidian plugins
  • Scales to tens of thousands of notes

Getting Started

  1. Install: Settings → Community Plugins → Browse → "Notemd"
  2. Configure: Add your LLM provider API key (or use local Ollama)
  3. Try it: Open a note → Right-click → "Process file (add links)"
  4. Explore: Check the sidebar for one-click workflows

👉 Installation Guide | Quick Start Tutorial

Diagram Capability Direction

Notemd's diagram work is moving away from "ask the model to write one syntax string" and toward a layered pipeline:

The current implementation already supports Mermaid, JSON Canvas, Vega-Lite, HTML fallback, editable HTML/SVG, Draw.io XML artifacts, a minimal Drawnix JSON subset, preview diagnostics/source-only fallback, and an offline CircuitSpec -> circuitikz prototype for common-source and CMOS inverter golden templates. Circuit diagrams are a harder class: circuitikz can express accurate electrical topology, but unconstrained LLM output often produces unreadable routing or non-rendering LaTeX. The next direction is to keep circuitikz constrained with golden-reference templates, node-grid layout rules, render diagnostics, and screenshot feedback loops.

Read the details in Diagrams.

Architecture

Notemd vs Other Obsidian AI Plugins

Most Obsidian AI plugins are conversation-first (you ask, AI answers, insights stay in chat). Notemd is write-first: AI processes your notes and writes structured results directly into your vault.

CapabilityNotemdCopilotSmart ConnectionsText Generator
Auto wiki-link insertionYesNoNoNo
Concept note generationYes (with backlinks + dedup)NoNoNo
Diagram generationYes (Mermaid, Canvas, Vega-Lite, HTML, editable artifacts)NoNoNo
Web research integrationYes (Tavily + DuckDuckGo)NoNoNo
Batch folder processingYesLimitedNoLimited
Per-task model routingYes (7 tasks, independent models)NoNoNo
One-click workflow chainsYes (DSL)NoNoNo
Translation (batch)YesNoNoNo
Chat with vaultNoYesNoNo
Semantic similarity searchNoNoYesNo
Template-based generationNoNoNoYes
LLM providers36 (cloud + gateway + local)3-52-33-5
Fully offlineYes (Ollama)PartialPartialPartial

When to choose Notemd: You want AI to build a persistent knowledge graph — not just chat about your notes.

When to choose Copilot: You want a conversational AI assistant inside Obsidian.

When to choose Smart Connections: You want to discover existing relationships between notes via semantic search.

Philosophy

Notemd believes that AI should augment human knowledge work, not replace it. The plugin:

  • Keeps you in control (review before applying changes)
  • Preserves context (all results link back to source)
  • Respects privacy (local LLM support, no telemetry)
  • Stays extensible (open APIs, custom workflows)

Open Source


Next: Installation →