Integrate client-side search engines

Each example is independent and uses the standard profile. Install the package from source and the matching engine version listed in installation. Run the snippets as ESM in Node or bundle them for the browser. The adapters do not supply a UI or replace engine ranking.

MiniSearch

import MiniSearch from "minisearch";
import { faMiniSearch } from "fa-search-kit/minisearch";
const fa = faMiniSearch();
const index = new MiniSearch({
  fields: ["title"], storeFields: ["title"], ...fa,
  searchOptions: { ...fa.searchOptions, boost: { title: 2 } },
});
index.addAll([
  { id: "books", title: "كتابهاي قديمي" },
  { id: "car", title: "ماشین قرمز" },
]);
console.log(JSON.stringify(index.search("کتاب").map(({ id, title }) => ({ id, title })))); // => [{"id":"books","title":"كتابهاي قديمي"}]

The adapter supplies index tokenize, identity processTerm, and query versions under searchOptions. Merge that object when adding boosts, prefix, or fuzzy options. Replacing it loses query-mode analysis. autoSuggest also uses those search options.

Default matching is MiniSearch’s OR. Use faMiniSearch({ combineWith: "AND", lexicon }) for AND and full-profile lemma defaults. Import lexicon from fa-search-kit/lexicon first. When changing combineWith later, keep verb behavior intentional and rebuild if terms change. Fuzzy matching and prefix search are engine features; evaluate their ranking effects separately from rescue.

Orama

import { create, insertMultiple, search } from "@orama/orama";
import { faTokenizer } from "fa-search-kit/orama";
const db = create({
  schema: { title: "string" },
  components: { tokenizer: faTokenizer() },
});
await insertMultiple(db, [{ id: "books", title: "كتابهاي قديمي" }]);
console.log(JSON.stringify((await search(db, { term: "کتاب" })).hits.map(h => h.id))); // => ["books"]

Do not also pass language to create with a custom tokenizer. Orama supplies a property name while indexing and no property during queries; the adapter uses this to select modes.

exactTerms: true is the default: it appends _ to each term to prevent ordinary prefix lookup from matching longer words. Set exactTerms: false to retain prefix matching and rebuild the index. This is different from Orama’s exact: true: in the tested 3.1.18 version that option uses a word-boundary test unsuitable for Persian. Engine typo tolerance remains separate; no tolerance is enabled by this adapter. Full-profile verbs default to stem.

FlexSearch

import FlexSearch from "flexsearch";
import { faDocument, faEncode } from "fa-search-kit/flexsearch";
const options = { document: { id: "id", index: ["title"] } };
const index = faDocument(FlexSearch, options);
index.add({ id: "books", title: "کتاب‌خانه" });
console.log(JSON.stringify(index.search("کتاب خانه", { merge: true }).map(h => h.id))); // => ["books"]
const simple = new FlexSearch.Document({ ...options, encode: faEncode() });
simple.add({ id: "books", title: "کتاب‌خانه" });
console.log(JSON.stringify(simple.search("کتاب خانه", { merge: true }))); // => []

faDocument wraps synchronous add, append, and update to use index mode, then restores query mode for search. It replaces any encode or encoder supplied in Document options. Other options, including stored fields, remain yours.

Do not combine this wrapper with workers or *Async writes: encoding can run after the temporary mode has been restored. The drop-in faEncode uses query mode on both sides, so it cannot add compound parts, alternate half-space terms, or madda-less index alternatives. It is a deliberate reduced-capability path. Full-profile verbs default to lemma.

Lunr

import lunr from "lunr";
import { faLunr } from "fa-search-kit/lunr";
const fa = faLunr(lunr);
const index = lunr(function () {
  this.use(fa);
  this.ref("id");
  this.field("title");
  this.add({ id: "books", title: "كتابهاي قديمي" });
});
console.log(JSON.stringify(fa.search(index, "کتاب").map(h => h.ref))); // => ["books"]

The plugin resets both English pipelines and replaces the builder tokenizer. Search through fa.search(index, query); index.search() uses Lunr’s own parser and bypasses the adapter tokenizer.

The helper adds analyzed terms with usePipeline: false; any term may match. Lunr syntax characters such as :, ~, ^, +, -, and * are treated as input text/separators, not operators. Advanced field queries or required/prohibited clauses need your own deliberate integration through Lunr’s query API. The plugin assigns token sequence metadata, not original character positions. Full-profile verbs default to stem.

See shared configuration, highlighting, and adapter signatures.