Kiara is an open-source, in-memory search library for Java and TypeScript. Load your records, call prepare() once, and search by any mix of fields, with typo correction and type-ahead built in.
Based on KapoorLabs' engineering article, October 2026
Every field you add to a filter panel doubles the combinations users can pick. Exact-key lookups need one key per combination. Composite indexes cut that down, but each one only serves its leading columns.
Kiara keeps one index per field and joins them at query time. Records are stored as paths in a tree, nodes are numbered, and conditions combine with integer range checks.
prepare() call.Source: "Your data has 10 attributes...", KapoorLabs on dev.to
Any field. Any mix. One prepare() call.Footage inside the type: Mixkit
A hand-written filter can match exact values. Users type faster and less carefully than that.
Fields marked oneEditAway can fix one wrong, missing or extra letter per word. It doesn't catch every typo, and one near the start of a word often stays uncorrected. The result tells you what was corrected, so your interface can show it.
Plurals work too. Keep a stemmed copy of the field and search both, so "laptops" finds laptop and the typo fixes still work.
// one field for typos, a stemmed copy for plurals textField("name", { oneEditAway: true }) textField("nameStem", { stemmedIndex: true }) keywordSearch.bestMatch("wireles keybord", { fields: ["name", "nameStem"], }) // correctedKeywords: { wireles: "wireless", keybord: "keyboard" }
A dictionary trie predicts whole words from the first letters. Type "wi" and get "wireless". It's the same data structure the search uses, so there's nothing extra to keep in sync.
const dictionary = new DictionaryTrie();
dictionary.addText("wireless headphones");
dictionary.predict("wi", 5)
// → ["wireless"]Users pick filters in unpredictable combinations. Kiara answers each one from the same prepared store, with no index planning per query.
store.search([
createCondition({ field: "category", operator: Operator.EQUAL, values: ["laptop"] }),
createCondition({ field: "price", operator: Operator.BETWEEN, lowerValue: 300, upperValue: 900 }),
createCondition({ field: "tags", operator: Operator.CONTAINS_ALL, values: ["usb-c", "eco"] }),
]);10,000 example products, searched by the real kiara-ts library inside your browser. Nothing is sent anywhere.
Median query time, lower is better. Kiara won all 16 scenarios both libraries can run at this size. The bars use a log scale, because the gap is too wide for a straight one.
Source: KapoorLabs benchmark on GitHub. One machine, mostly synthetic data, median of three isolated runs. Your results will differ.
From research notes to a library developers use. Photos: Unsplash
The docs also say clearly when not to use it: frequent writes, durable storage or relational joins.
A schema or decorators. textField("name")
One call builds every index. store.prepare()
In the browser or on a Java server. store.search([...])
Five questions, about one minute. If another tool fits your case better, it says so.
The demo runs in your browser in seconds. If it fits your data, the code is one install away, and the authors want to hear what you build.