Vault Mind

Local AI-powered semantic search for Obsidian. Your notes never leave your machine.

What it does

Vault Mind lets you search your vault by meaning, not keywords. Ask "what did I write about productivity?" and it finds relevant notes even if none of them contain those exact words.

All embeddings run locally via Ollama. No API keys. No cloud. No data leaves your machine.

Requirements

• Obsidian 1.0.0+
• Ollama running locally
• The nomic-embed-text model (or any other Ollama embedding model)

Setup

1. Install Ollama

Download from ollama.com and run:

ollama serve

2. Pull the embedding model

ollama pull nomic-embed-text

This is a ~274MB download. It runs on CPU — no GPU required.

3. Install Vault Mind

Until this plugin is in the community marketplace:

1. Download main.js, manifest.json, and styles.css from the latest release
2. Create a folder: <your-vault>/.obsidian/plugins/vault-mind/
3. Place the three files in that folder
4. Enable the plugin in Obsidian Settings > Community Plugins

4. Index your vault

Click the brain icon in the left ribbon, or run the command Vault Mind: Update index (incremental).

The first index takes time — roughly 1-2 seconds per note depending on your hardware. Subsequent runs only process changed files.

Usage

Click the brain icon (or use the ribbon) to open the search panel. Type anything — questions, concepts, partial phrases. Results update as you type with a 300ms debounce.

Each result shows:

• Similarity score (0-100%)
• Note title
• Relevant snippet with matched terms highlighted
• Click to open the note

Settings

Setting	Default	Description
Ollama endpoint	http://localhost:11434	URL of your Ollama server
Embedding model	nomic-embed-text	Model for generating embeddings
Max results	10	Results to show per search
Auto-index on startup	On	Check for changed files on load
Batch size	10	Files per indexing batch

How it works

1. Indexing: Each note is converted to a 768-dimensional vector using nomic-embed-text. Vectors are stored in a binary file (.obsidian/plugins/vault-mind/data/vault-mind.bin) alongside a JSON metadata index.

2. Search: Your query is embedded with the same model, then compared to all note vectors using cosine similarity. The top-k matches are returned in milliseconds.

3. Incremental updates: File modification times are tracked. Only changed notes are re-embedded on subsequent index runs. File watchers update individual notes 3 seconds after you stop editing.

Storage

Index files are stored at:

<vault>/.obsidian/plugins/vault-mind/data/
├── vault-mind.bin         # Float32Array binary (4 bytes × 768 dims × N notes)
└── vault-mind-index.json  # Note metadata (paths, mtimes, snippets)

For 10,000 notes: ~30MB binary + ~5MB JSON. For 20,000 notes: ~61MB binary.

Model recommendations

Model	Dimensions	Size	Notes
nomic-embed-text	768	274MB	Default. Best balance of speed and quality.
mxbai-embed-large	1024	670MB	Higher quality, slower, larger index.
all-minilm	384	46MB	Fastest, smallest. Good for low-RAM machines.

If you change models, run Re-index vault (full rebuild) from settings or the command palette.

Commands

• Vault Mind: Open semantic search — Open the search panel
• Vault Mind: Update index (incremental) — Embed new/changed notes
• Vault Mind: Re-index vault (full rebuild) — Wipe and rebuild the entire index

Privacy

Nothing leaves your machine. Vault Mind communicates only with your local Ollama process at localhost:11434. No telemetry, no analytics, no network requests to external servers.

License

MIT