kokey

Keyboard layout converter — restore text typed with the wrong layout. Korean (Dubeolsik ↔ QWERTY) built in; Russian, Ukrainian, Hebrew, Greek, Thai, Arabic and Georgian available as tree-shakeable subpath imports. TypeScript-first, zero dependencies, ESM/CJS dual package.

English · 한국어 · Русский · Українська · עברית · Ελληνικά · ไทย · العربية · ქართული · Live demo

Try it online: ⚡ StackBlitz — Vanilla · Vue · React · Svelte · Solid | 📦 CodeSandbox

Ever typed dkssud when you meant 안녕, or scanned a barcode while the Korean IME was on and got ㅇㄴㅁ쇼2068601 instead of DSATY2068601? kokey converts between what was typed and what was meant — in both directions, exactly the way a Dubeolsik IME composes Hangul.

The same slip exists in every language that toggles a non-Latin layout with QWERTY: Russians type ghbdtn for привет, Israelis akuo for שלום, Thais scan barcodes with Kedmanee on. kokey covers those layouts too — see Beyond Korean.

Numeric fields in the same form — amounts that need live comma grouping, a stable caret, and right alignment? That's kokey's sibling, numkey.

Install

npm install @devslab/kokey

Or straight from a CDN — no build step, everything under the kokey global:

<script src="https://cdn.jsdelivr.net/npm/@devslab/kokey/dist/kokey.global.js"></script>
<script>
  kokey.enToKo('dkssud')   // '안녕'
  kokey.toEn('привет안녕')  // 'ghbdtndkssud' — the CDN build ships every layout
  kokey.observe()          // auto-bind every <input data-kokey> / <input data-hangul>
</script>

Usage

import { koToEn, enToKo } from '@devslab/kokey'

// Hangul → the QWERTY keystrokes that produced it
koToEn('안녕')            // 'dkssud'
koToEn('값없는 닭갈비')     // 'rkqtdjqtsms ekfrrkfql'
koToEn('ㅇㄴㅁ쇼2068601')  // 'dsaty2068601' (wedge-scanner rescue)

// QWERTY keystrokes → composed Hangul (full IME automaton)
enToKo('dkssud')          // '안녕'
enToKo('gksrmf')          // '한글'
enToKo('ekfrl')           // '달기' (compound-final split, like a real IME)

Details that matter

• Shift is honored: R → ㄲ, r → ㄱ, koToEn('뛰다') === 'Enlek'
• Compound vowels/finals: ㅘ ↔ hk, ㄵ ↔ sw, …
• 받침 넘김 (final-consonant carry-over): enToKo('dkswk') === '안자'
• Pass-through: digits, punctuation, and unmapped letters are left as-is
• Round-trip safe for Korean text: enToKo(koToEn(s)) === s

Beyond Korean — any registered layout

Each layout is a subpath import (unused ones never reach your bundle) and plugs into the same machinery:

import { register, toEn, fromEn } from '@devslab/kokey'
import { ru, ruToEn, enToRu } from '@devslab/kokey/ru'
import { he } from '@devslab/kokey/he'

// direct, per-layout
ruToEn('привет')   // 'ghbdtn'  — the Punto Switcher classic
enToRu('ghbdtn')   // 'привет'

// or register + auto-detect by script, mixed strings included
register(ru, he)
toEn('안녕 привет שלום')  // 'dkssud ghbdtn akuo'
fromEn('ghbdtn', 'ru')    // 'привет'

Layout	Import	Notes
Korean 두벌식	built-in (ko)	full IME composition automaton
Russian ЙЦУКЕН	@devslab/kokey/ru	moved punctuation (ё on backtick, №, . on /) mapped faithfully
Ukrainian Enhanced	@devslab/kokey/uk	і/є/ї; AltGr-only ґ restored in reverse; ru/uk auto-disambiguated
Hebrew	@devslab/kokey/he	final forms, swapped brackets, caps-lock safe
Greek	@devslab/kokey/el	tonos/dialytika dead keys (;a → ά), final sigma
Thai Kedmanee	@devslab/kokey/th	full digit-row remap — barcode rescue works for Thai too
Arabic (101)	@devslab/kokey/ar	lam-alef لا on b, hamza forms, tashkeel
Georgian QWERTY	@devslab/kokey/ka	near-phonetic (gamarjoba ↔ გამარჯობა)

Languages whose IME needs a candidate-selection step (Chinese pinyin, Japanese kanji) are out of scope by construction — the keystroke ↔ text relation there isn't deterministic. Need another deterministic layout? It's one defineLayout({ id, script, fromKey }) table — PRs welcome.

DOM layer — enforce an input mode

Force an <input>/<textarea> to a specific mode regardless of the user's IME state — the field converts as you type, composition-safe, cursor preserved:

<input data-kokey="ko">   <!-- QWERTY keystrokes compose into Hangul -->
<input data-kokey="ru">   <!-- QWERTY keystrokes become Russian (register(ru) first) -->
<input data-kokey="en">   <!-- any registered script restored to QWERTY -->
<input data-hangul="ko">  <!-- legacy attribute, still supported -->

import { bind, observe } from '@devslab/kokey'

observe()                    // bind all [data-kokey]/[data-hangul] + watch for new ones
const unbind = bind(el, 'en') // or bind a single element explicitly

data-kokey="en" shines on invoice/e-mail/ID fields: whatever layout the user forgot to switch off — Korean, Russian, Thai — the field self-heals to Latin with no per-language branching.

Paste auto-correction

Fix wrong-layout gibberish on paste without forcing a mode on the input:

<input data-kokey-paste>  <!-- picked up by observe() -->

import { bindPaste, fixMistyped } from '@devslab/kokey'

bindPaste(el)               // imperative version
fixMistyped('dkssudgktpdy') // '안녕하세요' — or null if it looks fine
fixMistyped('hello')        // null

Detection is deliberately conservative and Korean-only, because composition itself is the signal: standalone vowel jamo mixed into text marks English typed in Korean mode (ㅗ디ㅣㅐ → hello), and Latin words that recompose into complete syllables with nothing left over mark Korean typed in English mode (dkssudgktpdy → 안녕하세요). Real Korean — including ㅋㅋㅋ/ㅠㅠ laughter — and real English pass through untouched; a single short word must yield at least three syllables before it fires. Other layouts have no such validity signal (any Latin string maps to Cyrillic), so use explicit modes for those. Before replacing, a cancelable kokey-paste CustomEvent fires with detail: { pasted, fixed } — preventDefault() it to veto, or use fixMistyped directly to build a suggest-UI instead.

Vue / React / Svelte / Solid

For plain (uncontrolled) inputs, the directive/hook:

<script setup>
import { vKokey } from '@devslab/kokey/vue'
</script>
<template>
  <input v-kokey="'ko'">
  <input v-kokey="'ru'">
</template>

import { useKokey } from '@devslab/kokey/react'

function Field() {
  return <input ref={useKokey('en')} />
}

For v-model / controlled inputs, use the KokeyInput component — it converts inside the framework's data flow, so your bound state always holds the converted value (the ref-based bindings mutate the DOM after the framework reads it, which fights v-model/value=):

<script setup>
import { KokeyInput } from '@devslab/kokey/vue'
const name = ref('')
</script>
<template>
  <KokeyInput v-model="name" mode="ko" />
  <KokeyInput v-model="memo" mode="en" as="textarea" />
</template>

import { KokeyInput } from '@devslab/kokey/react'

function Form() {
  const [v, setV] = useState('')
  return <KokeyInput mode="en" value={v} onChange={(e) => setV(e.target.value)} />
}

Svelte gets an action — bind:value works, the action re-syncs the binding after converting (and it imports nothing from svelte, so there is no peer dependency at all):

<script>
  import { kokey, kokeyPaste } from '@devslab/kokey/svelte'
  let name = ''
</script>

<input use:kokey={'ko'} bind:value={name} />
<input use:kokeyPaste />

Solid gets a use: directive (reactive to a mode signal) and a ref factory — Solid's delegated onInput already reads the converted value:

import { kokey, useKokey } from '@devslab/kokey/solid'

<input use:kokey={mode()} onInput={(e) => setV(e.currentTarget.value)} />
<input ref={useKokey('en')} />

All are thin wrappers over the DOM layer — vue/react/solid-js are optional peer dependencies (Svelte needs none), so the core stays zero-dependency. The legacy vHangul / useHangul names still work.

API

Function	Signature	Description
koToEn	(text: string) => string	Decompose Hangul syllables/jamo into their Dubeolsik QWERTY key sequence
enToKo	(text: string) => string	Compose QWERTY key sequence into Hangul via the standard IME automaton
toEn	(text: string) => string	Restore any registered script to QWERTY, auto-detected per run
fromEn	(text, layoutId) => string	Compose QWERTY keystrokes into the given registered layout
register	(...layouts) => void	Register layouts for toEn and the DOM data-kokey modes
defineLayout	(def) => Layout	Build a table-driven layout ({ id, script, fromKey })
bind	(el, mode?) => unbind	Enforce a mode on one input/textarea (mode defaults to its data-kokey/data-hangul attribute)
observe	(root?) => stop	Bind every [data-kokey]/[data-hangul] under root and keep watching via MutationObserver
createRefBinder	(mode?) => (el | null) => void	Framework-agnostic ref-callback factory (what useKokey wraps)
fixMistyped	(text) => string | null	Correct wrong-layout gibberish, or null if the text looks fine (heuristic, Korean-only)
bindPaste	(el) => unbind	Auto-correct wrong-layout pastes on one input (data-kokey-paste via observe)
vKokey	@devslab/kokey/vue	Vue 3 directive: v-kokey="'ko'" (legacy vHangul kept)
KokeyInput	@devslab/kokey/vue · /react	Component for v-model / controlled inputs (mode, as="input|textarea")
useKokey	@devslab/kokey/react · /solid	Hook/ref factory returning a ref callback (legacy useHangul kept)
kokey	@devslab/kokey/svelte · /solid	Svelte action / Solid directive for use:kokey (+ kokeyPaste in both)
convert	(text, mode) => string	One-shot conversion for a mode ('en' or a layout id)
applyToInput	(el, mode) => boolean	Convert an input's value in place, caret preserved

Per-layout modules also export direct converters: ruToEn/enToRu, heToEn/enToHe, thToEn/enToTh, … Low-level Korean tables (CHOSUNG, JUNGSUNG, JONGSUNG, JAMO_TO_KEY, KEY_TO_JAMO) are exported for advanced use.

Roadmap

• v0.2 — DOM layer ✅ shipped
• v0.3 — Vue directive / React hook ✅ shipped
• v0.4 — multi-layout: ru/uk/he/el/th/ar/ka + toEn auto-detection ✅ shipped
• v0.5 — Svelte action / Solid directive + paste auto-correction
• v0.6 — browser extension (fix mistyped text on any site — context menu + hotkey) — 🚧 in repo (extension/), store submission pending

Why not inko?

inko pioneered this space but has been unmaintained since 2019 and predates modern TypeScript/ESM packaging. kokey is a from-scratch implementation: typed, tree-shakeable, dual ESM/CJS, tested against real IME behavior (compound finals, carry-over, shift handling).

Contributing

Issues and PRs welcome — new layouts especially. See CONTRIBUTING.md for the dev setup and the two hard rules for layout tables (anchor verification + round-trip tests).

Family

• numkey — the numeric sibling in the "-key" input family: live thousands grouping, Korean amount UX, string-first canonical values
• More open source from devslab

License

MIT © devslab