jev-search

# Jev Search for Obsidian

[English](README.md) | [日本語](README.ja.md)

[](https://github.com/jh1373/jev-search/releases/tag/v0.1.0)
[](LICENSE)
[](https://obsidian.md)
[](https://typesafe.ai)

**Next-generation hybrid AI search for Obsidian that surfaces the exact knowledge you need from thousands of notes in milliseconds.**

Combining an ultra-fast local BM25 search engine with TypeSafe's revolutionary "Jev" System One judgment model. Narrow down candidates across 10,000 notes in 15 milliseconds offline, then let calibrated AI judgment elevate the note that genuinely answers your question directly to #1.

---

## Why Jev Search? (The Problem vs. Solution)

### The Limitations of Standard Obsidian Search
As your Obsidian vault grows to hundreds or thousands of notes, locating specific thoughts becomes increasingly difficult:
- **Exact-Keyword Brittleness**: Searching for "API design tips" fails completely if your note mentions "endpoint creation guidelines".
- **Drowning in Notes**: Notes with a single accidental keyword match bury your most critical insights at rank #20 or #30.
- **Flaws in Existing LLM Plugins**: Traditional generative LLM plugins attempt full-text answers or summaries. They introduce 5–15 second waiting times, fabricate information (hallucinations), and incur significant recurring subscription or token costs.

### The Jev Advantage
Jev Search integrates **TypeSafe's flagship "Jev" model**—the world's first System One model designed specifically for focused software judgments:

1. **System One Judgment: Evaluative, Not Generative**
   - Jev does not generate conversational prose. Instead, it evaluates whether a given note excerpt directly answers your query, outputting calibrated probability distributions and structured scores.
   - With no generative