Developers
Two battle-tested SDKs, one straightforward HTTP API, and a streaming surface that doesn't make you build back-pressure yourself. Install, get a key, make the call.
One line, any terminal:
Code that runs on first copy
The same three endpoints back the landing demo, the playground, and your production pipeline. Every example below is the shortest working call, no fake helpers, no hidden config.
TTS with Hakim Fast v1.3. Streaming MP3 to disk, a voice you already heard on the landing.
import { Hakim } from '@tryhakim/voice';
const client = new Hakim({ apiKey: process.env.HAKIM_API_KEY });
const audio = await client.audio.speech.create({
model: 'hakim-fast-v1.3',
voice: 'reem-khaleeji',
input: 'أهلاً وسهلاً بكم في حكيم.',
format: 'mp3',
});
await audio.writeToFile('hello.mp3');STT with Hakim Arab v2. Verbose JSON so you get timestamps, speakers, and confidences without a second call.
import { createReadStream } from 'node:fs';
import { Hakim } from '@tryhakim/voice';
const client = new Hakim({ apiKey: process.env.HAKIM_API_KEY });
const transcription = await client.audio.transcriptions.create({
model: 'hakim-arab-v2',
file: createReadStream('call.wav'),
language: 'ar',
response_format: 'verbose_json',
});
console.log(transcription.text);Every parameter lives in the API reference.
SDKs + bindings
Typed surface, idiomatic naming, zero bundled telemetry. Node and Python are stable and on every release. HTTP is the common denominator, cURL always works.
ESM + CJS. First-class streams. Runs on Node 18+, Cloudflare Workers, Deno, Bun.
npm install @tryhakim/voice
Sync + async clients. Typed with mypy. Runs on CPython 3.9+ and PyPy.
pip install hakim
Idiomatic Go client with context-aware streaming. Targeted for v1.1.
Coming soonZero-cost async client using tokio + tower. Targeted for v1.1.
Coming soonBuild in five minutes
We're not grading the stopwatch, we're saying the flow is small enough that if you get stuck, something's wrong with us, not you.
No card, no trial clock, no automatic Starter plan. Sign up with email or GitHub and you're on the free tier.
Dashboard → API keys → new key. Named, scoped, revokable. We never show the raw key again after you close the modal.
npm install @tryhakim/voice or pip install hakim. Or skip the SDK and hit the HTTP API with cURL, everything works the same way.
Paste one of the snippets above, replace the text, run. You'll hear Reem in Khaleeji, or you'll see the transcription for your own audio file.
Streaming
TTS streams PCM or MP3 chunks from the first viable frame. STT streams interim + final tokens over WebSocket. We don't hold a whole utterance hostage to buffer on our side.
Streamed synthesis starts emitting audio before the full utterance is generated, so your user hears the opening phonemes while the rest is still decoding.
Both SDKs honour consumer back-pressure, stop reading and we stop generating. No phantom bills for audio you never played.
Long STT sessions survive reconnection via a resume token. Lose the socket, reconnect, continue, no re-uploading the audio you already sent.
Every link you want bookmarked
The docs are the source of truth, this page is the side entrance.
