Joplin Note Categorization Plugin

A local-first AI plugin for Joplin that clusters notes semantically, suggests tags and notebook structures.

Features

• On-Device Embeddings: Uses @huggingface/transformers (all-MiniLM-L6-v2) inside a Web Worker. Automatically utilizes WebGPU (fp16) on supported platforms (macOS/Windows) with WebAssembly (q8) fallback on Linux. Reuses Joplin's native AI search embeddings when available.
• Clustering Algorithms:
  • K-Means: Automatically determines the optimal number of clusters ($k$) using Silhouette score evaluation.
  • HDBSCAN: Density-based clustering to detect natural topic groupings and isolate outlier/noise notes.
  • UMAP: Reduces vector dimensionality for consistent distance projection.
• Topic & Tag Extraction: Generates cluster titles and tag suggestions using TF-IDF term scoring or generative AI naming.
• Interactive Panel: Side-by-side strategy comparison, drag-and-drop note re-assignment between clusters, cluster renaming, and custom category creation.
• Organization Modes: Move notes into generated sub-notebooks, apply tags, or both.
• Undo System: Full change-tracking log allowing one-click rollback of notebook moves and tag assignments via the panel, Tools menu, or Joplin Settings.

Demonstration

Quick Demo

export-1786849255881.mp4

Full Video Walkthrough

• YouTube: Watch the full walkthrough

How It Works

1. Ingestion & Chunking: Notes are fetched via the Joplin Data API and split into 200-token chunks with js-tiktoken (cl100k_base).
2. Embedding Generation: Reuses native Joplin AI Search vectors if available, or generates embeddings locally using all-MiniLM-L6-v2 via WebGPU (fp16) or WASM (q8). Chunk vectors are combined using mean pooling.
3. Dimensionality Reduction & Clustering: Embeddings are projected via UMAP, then clustered using K-Means (Silhouette-optimized auto-$k$) or HDBSCAN (density-based with outlier isolation).
4. Topic & Tag Extraction: Cluster names and tags are derived through either generative Joplin AI Naming or offline statistical TF-IDF keyword extraction.
5. Execution & Rollback: Reorganizations are reviewed in the interactive panel and applied directly to Joplin notebooks and tags, with full state logging for one-click undo.

Installation

1. Open Joplin.
2. Go to Tools -> Options -> Plugins (or Joplin -> Preferences -> Plugins on macOS).
3. Search for Note Categorization and click Install.
4. Restart Joplin.