@leuria/docs

Ask AI and search by meaning for any docs site, on the reader's own AI. The site holds no keys and pays for no model. @leuria/docusaurus wires it into Docusaurus in one line; this package works anywhere.

import { buildCorpus } from "@leuria/docs/build"          // at build time
const corpus = buildCorpus(pages, { site: "My docs" })     // pages: { url, title, markdown }[]

import { createDocs, setupDocs } from "@leuria/docs"       // in the page
const docs = createDocs(ai, { corpus: () => fetch("/corpus.json").then((r) => r.json()) })
setupDocs(docs, { suggestions: ["How do I get started?"] })

<leuria-search-button></leuria-search-button>
<leuria-ask-button></leuria-ask-button>

| Piece | What | | --- | --- | | buildCorpus(pages) | Pages to a corpus: cleaned Markdown, and sections with the site's anchors. light: true keeps titles and headings only, and pages are read from the site | | algoliaSearch(options) | The site's Algolia DocSearch, for createDocs({ search }) | | createDocs(ai, options) | The docs in the page: search() by words and by meaning (with @leuria/store), tools for an assistant, expose() for the browser's agents (WebMCP) | | <leuria-ask-button>, openAsk() | The Ask panel. AIs that can use tools search and read the docs; others (the browser's small model) get the best passages with the question | | <leuria-search-button>, openSearch() | The ⌘K search, with "Ask AI" for the query |

The elements keep Leuria Pearl's styles in their shadow root, as @leuria/connect does. Full guide: Ask AI for your Docusaurus site. Apache-2.0.