Built-in Search

SmartDocs includes a fast, offline search bar powered by Fuse.js — a lightweight fuzzy search library.

How It Works

The search bar (Cmd+K / Ctrl+K) runs entirely in your browser. No server calls, no API costs, no latency.

The Index

Search uses the same page-index.json that powers the AI chat. One build artifact, two features. Each page entry contains:

{
  "slug": "/guides/configuration",
  "title": "Configuration Guide",
  "headings": ["Environment Variables", "Providers", "Security"],
  "keywords": ["smartdocs", "api", "environment", "variables"],
  "description": "SmartDocs is configured through environment..."
}

Fuzzy Matching

Fuse.js uses a modified Bitap algorithm for approximate string matching. It scores pages against your query across multiple fields with different weights:

Field	Weight	Why
title	3x	Most important — page name is the primary signal
headings	2x	Section titles are strong indicators of content
keywords	1x	Extracted terms from page body
description	0.5x	Supporting signal, least specific

The threshold: 0.4 setting determines how strict the matching is. Lower = more strict (only very close matches), higher = more forgiving (fuzzier matches). 0.4 means "quikstart" finds "Quickstart" but "deployment" won't match "Getting Started".

Instant Results

Results appear as you type — no debounce, no loading spinners. The page index is pre-loaded when you first open search (<100ms for typical docs). Subsequent searches are instant.

Keyboard Navigation

Key	Action
Ctrl+K / Cmd+K	Toggle search
↑ ↓	Navigate results
Enter	Go to selected page
Escape	Close search

Why Fuse.js?

Library	Size	Offline	Fuzzy	Speed
Fuse.js	13KB gzipped	✅	✅	⚡ Instant
Lunr.js	25KB	✅	❌ (stemming only)	Fast
Algolia	External SaaS	❌	✅	Fast (API)
Meilisearch	External service	❌	✅	Fast (API)

Fuse.js was chosen because:

Zero infrastructure — runs in the browser, no server needed
Fuzzy matching — handles typos, partial words, approximate queries
Tiny footprint — 13KB gzipped, loaded once
Configurable — per-field weights, threshold tuning
No API key — no external dependency, no cost, no privacy concern

Customizing Search

Adding More Fields

To index additional page metadata, modify the page-index.json generation in scripts/build-graph.js and add the field to Fuse.js keys in components/SearchBar.tsx.

Adjusting Sensitivity

Change the threshold value in SearchBar.tsx:

0.2 — very strict, exact matches only
0.4 — balanced (default)
0.6 — very lenient, many results

Excluding Pages

Filter the pages array before passing to Fuse.js. For example, to exclude draft pages:

const published = pageIndex.pages.filter(p => !p.slug.startsWith('/drafts'));
fuseInstance = new Fuse(published, { ... });

How the AI Works — The full retrieval pipeline
Karpathy's LLM Wiki — The inspiration behind the graph approach
Configuration Guide — Setting up your project