The @docmd/plugin-search plugin provides a client-side search experience for your documentation site. It uses MiniSearch to build a compressed index during compilation, enabling readers to search technical documentation instantly without requiring server-side databases or third-party crawling services.

Configuration Options

Search is enabled by default across standard docmd templates. Configure indexer parameters and header placement in docmd.config.json:

Option Type Default Description
enabled boolean true Enable or disable full-text search index generation.
placeholder string 'Search...' Input placeholder text in search dialogs.
maxResults number 10 Maximum number of search results returned in the modal window.

Header Integration Example

docmd.config.json
{
  "layout": {
    "optionsMenu": {
      "position": "header",
      "components": {
        "search": true
      }
    }
  }
}

How Keyword Search Works

1. Build-Time Indexing

During site compilation (npx @docmd/core build), @docmd/plugin-search traverses every page on your site. It extracts headings, titles, and body prose to generate a compressed search-index.json bundle:

  • Deep Linking: Registers heading anchors (#, ##) as direct search jump targets.
  • Relevance Weighting: Page titles receive highest weighting, followed by section headings, then prose paragraphs.

2. Client-Side Retrieval

When a user opens the search modal (press Ctrl+K or /), the browser fetches search-index.json. Queries execute locally with prefix matching and fuzzy string distance matching to accommodate minor typos.

Customising Search Scope

To exclude specific pages from the search index, add noindex: true to Page Frontmatter:

---
title: "Internal Draft Specification"
noindex: true
---
Privacy & Compliance

Because search queries execute entirely within client memory, zero search input or keystroke telemetry leaves the user’s browser.

@docmd/plugin-search includes support for local semantic search powered by docmd-search. Semantic search uses client-side embedding models to process queries conceptually rather than matching literal keywords.

  1. Install docmd-search in your documentation workspace:
npm install docmd-search
  1. Enable semantic indexing in docmd.config.json:
docmd.config.json
{
  "plugins": {
    "search": {
      "semantic": true
    }
  }
}

Semantic Search Options

Option Type Default Technical Purpose
semantic boolean false Enable vector embedding search.
showConfidence boolean false Display similarity percentage badges on search results.
showFilters boolean true Display version filter controls in search dialogs.
model string 'Xenova/all-MiniLM-L6-v2' HuggingFace embedding model ID.
chunkSize number 512 Token chunking limit per document section.

Supported Embedding Models

Model ID Download Size Best For
Xenova/all-MiniLM-L6-v2 (Default) ~23 MB English technical documentation
Xenova/paraphrase-multilingual-MiniLM-L12-v2 ~118 MB Multilingual sites (German, Chinese, French)
Xenova/multilingual-e5-small ~118 MB Broad international language coverage
Automatic Fallback

If docmd-search is enabled but vector embedding dependencies cannot be loaded, the search plugin falls back gracefully to standard MiniSearch keyword indexing.

Previous Overview Next AI Assistant