Obsidian Stats logoObsidian Stats

VaultSearch

by erayaydn0
5
4
3
2
1
Score: 35/100
Install Code

Description

This plugin has not been manually reviewed by Obsidian staff. Local-first hybrid search for your vault — combines keyword (BM25), semantic (vector embeddings), and fuzzy title matching. Multilingual, on-device, no cloud.

Reviews

No reviews yet.

Stats

stars
422
downloads
0
forks
1
days
NaN
days
NaN
days
0
total PRs
0
open PRs
0
closed PRs
0
merged PRs
0
total issues
0
open issues
0
closed issues
0
commits

Latest Version

Invalid date

Changelog

See all version on GitHub

README file from

Github

VaultSearch

Local-first hybrid search for Obsidian — combines BM25, vector similarity, and fuzzy title matching. Runs entirely on-device. No cloud, no telemetry.

VaultSearch makes finding notes in your Obsidian vault feel instant and smart. Instead of relying on plain keyword search, it understands what you mean — even if your query doesn't exactly match the words in your notes, even across multiple languages.

Everything happens on your machine. Your notes never leave your device.

Why VaultSearch?

Obsidian's built-in search is great at exact matches, but it falls short when:

  • You remember an idea but not the exact words you used.
  • Your vault mixes multiple languages.
  • You want results ranked by relevance, not just appearance order.
  • You want to discover related notes you forgot you had.

VaultSearch fixes all of that — without sending a single byte to the cloud.

Features

  • 🔍 Hybrid search — three retrievers run in parallel and are fused into one ranked list:
    • BM25 keyword search (powered by SQLite FTS5) for exact term matches
    • Vector / semantic search that understands meaning, not just words
    • Fuzzy title matching for when you almost remember the filename
  • 🌍 Multilingual — works across 50+ languages out of the box
  • 🔒 100% local — no cloud, no API keys, no telemetry, no analytics
  • ⚡ Zero configuration — install, enable, and it just works
  • 🧠 Related notes sidebar — discover connections between notes automatically
  • 🎯 Smart highlights — see why a note matched, with snippet excerpts
  • 📦 No native dependencies — pure JavaScript and WebAssembly, works everywhere Obsidian works

How it works

When you first enable VaultSearch, it does three things:

  1. Downloads a small AI model (~47 MB, one-time) to understand the meaning of your notes
  2. Indexes your vault in the background — splits notes into chunks, computes embeddings, builds a full-text index
  3. Watches for changes — every time you create, edit, or rename a note, the index updates automatically

When you search, three engines run at once and their results are intelligently merged:

your query
    │
    ├─→ keyword search    (finds exact terms)
    ├─→ semantic search   (finds related ideas)
    └─→ fuzzy title       (finds approximate filenames)
            │
            ▼
       combined ranking → your results

All of this happens locally, in milliseconds, on your own computer.

Installation

⚠️ VaultSearch is in early development. Expect rough edges and breaking changes.

Usage

  • Search: run the VaultSearch: Open search command from the command palette (or assign your own hotkey under Settings → Hotkeys), type your query, hit Enter.
  • Related notes: open the right sidebar — VaultSearch shows notes related to whatever you're currently viewing.
  • Settings: customize chunk sizes, ranking weights, and indexing behavior under Settings → VaultSearch.

Privacy

VaultSearch is built around one rule: your vault is yours.

  • No telemetry, no analytics, no tracking
  • No cloud uploads
  • The only network call is the one-time AI model download from Hugging Face on first use — after that, the plugin works fully offline
  • The full list of constraints lives in DEVELOPMENT.md §8

Requirements

  • Obsidian 1.4.0+
  • Desktop only (Windows / macOS / Linux) — mobile is not supported because Obsidian mobile cannot run the underlying WebAssembly modules

Contributing

Contributions are very welcome! Whether it's a bug report, an idea, or a pull request — please read CONTRIBUTING.md first.

For questions and discussions, head to GitHub Discussions. For bugs, open an issue.

If you want to dive into the architecture and internals, DEVELOPMENT.md is the developer's guide. (AGENTS.md is the equivalent reference for AI agents working in the repo.)

Tech stack

Built with TypeScript, sql.js (WASM SQLite + FTS5), @huggingface/transformers (ONNX/WASM embeddings), esbuild, and Bun as the toolchain. Default embedding model: paraphrase-multilingual-MiniLM-L12-v2.

Status

Early development. The indexer, storage layer, and hybrid search engine are working. The MCP server module is intentionally frozen until its design is finalized.

License

MIT © Eray Aydın

Made with care for the Obsidian community. If VaultSearch helps you, a ⭐ on the repo means a lot.

Sponsors