The Vibe Typer Voice API puts the same fast, accurate voice pipeline behind any text field in your app. Your users speak naturally in any of 99 languages, and clean, corrected text lands moments after they stop. It costs US$0.50 per audio hour.
Using Claude Code, Cursor or Codex? Paste the prompt in and your agent adds voice typing to your app for you. You just add your API key.
um can we move the review to noon actually make that 3 PM on Friday and uh send the agenda the day before
Can we move the review to 3 PM on Friday and send the agenda the day before?
Everything that makes Vibe Typer fast and accurate on desktop and iPhone, running behind your text fields.
Text lands moments after your user presses stop. In Chrome and Edge the audio streams while they speak, so there's no upload to wait for.
Transcription runs on large cloud speech models, the same ones behind the Vibe Typer app. There's no model to pick and nothing to tune.
99 languages, detected on every recording. Users can switch language between recordings, and there's no language picker to build.
It hears “no wait”, “scratch that” and “actually make that” and types what the user meant. Filler words go, and punctuation and capitals are fixed. It's included in the price, and you can turn it off.
One key on your server, one route and one element. Or copy the prompt and let your coding agent do it.
Sign in and press Get API key. Add a card at checkout, and Polar's confirmation page shows your key. After that, Get API key opens your billing portal, where the key is under Benefit Grants. Save it as an environment variable on your server.
VIBE_TYPER_API_KEY=VTSK-...Install @vibetyper/voice and add a route behind your existing login. It trades your secret key for a token that covers one recording, so the key never reaches the browser.
import { voiceTokenResponse } from '@vibetyper/voice/server';
export async function POST(request) {
const user = await requireUser(request);
return voiceTokenResponse({
apiKey: process.env.VIBE_TYPER_API_KEY,
userId: user.id,
});
}Put <vibe-voice> next to any input or textarea. The user presses record, speaks, presses stop, and the text lands at the cursor. Rich editors listen for the text event instead.
<script type="module" src="https://cdn.jsdelivr.net/npm/@vibetyper/voice@0.1/dist/voice.js"></script>
<textarea id="message"></textarea>
<vibe-voice token-url="/api/voice-token" for="message"></vibe-voice>A Whisper API alternative that handles the microphone, the upload and the text field, not just the transcript.
The element opens the microphone, records, downsamples and uploads. Your app only ever receives text.
The browser only holds a token for one recording, issued by your own login-protected route. A token copied from DevTools is worth one recording at most.
Chrome, Edge, Firefox and Safari, including every browser on iPhone. It works as a voice typing React component in React 19, or as a plain custom element anywhere else.
Billed by the second
Get API key starts pay as you go at US$0.50 per audio hour. Add a card at checkout, and you're billed monthly with no fixed fee. Your limit starts at US$100 a month, and usage is non-refundable. By getting a key you agree to the Voice API Terms.
Sign in on this page and press Get API key. Add a server route that trades your secret key for a recording token, then put the <vibe-voice> element next to any input or textarea. Users press record, speak, press stop, and the text appears in the field.
It supports 99 languages and detects the spoken one on every recording, so there's no language setting to build or pass. Users can switch languages between recordings without changing anything. Magic Formatter cleans up the text in whatever language was spoken, and you can turn it off with magic-formatter="off".
Yes. It works in Safari on the Mac and in every browser on iPhone and iPad, as well as Chrome, Edge and Firefox. Chromium browsers stream the audio while the user speaks. Safari and Firefox upload it when the user presses stop, which adds about a second to a 20 second dictation.
No. Audio is processed in memory to produce the text, then discarded. We don't store recordings or transcripts, and we don't use them to train models. For each recording we keep only its length, a timestamp and the user ID your server sent, so we can bill you and enforce limits.
Yes. The same element runs in Electron, Tauri and other web-based desktop apps once you allow microphone access. Native apps built with Swift, C# or Qt record 16 kHz audio and call the HTTP API directly. Either way, tokens come from your server. Never ship the secret key inside the app.
Every account starts with a US$100 monthly limit, which is 200 hours of audio. We email you at 90% and again at 100%. At the limit, new recordings pause until your next billing period and the element shows a short message. Reply to either email and we can raise it.
Whisper gives you a raw transcript, and you still build the recording, audio conversion, upload and key handling yourself. The Vibe Typer Voice API ships all of that as one element, keeps your key on your server, and runs Magic Formatter so filler words and punctuation are fixed before the text reaches your app.