Talk to an agent that answers
in Gulf Arabic.
Pick who to call, allow your microphone, and say what you want the way you would say it on the phone: a dentist tomorrow, a flat to view, dinner delivered. It hears the dialect and answers in it, through nutq speech to text and text to speech, over LiveKit.
Who do you want to talk to?
- «أبغي موعد عند دكتور أسنان باچر»
- «عندي وجع ضرس من أمس»
- «العصر الساعة خمس يناسبني»
- «عندكم شقة غرفتين في المارينا؟»
- «كم الإيجار في السنة؟»
- «أقدر أشوفها يوم السبت؟»
- «أبغي أطلب برجر وبطاط»
- «عندكم شي حار؟»
- «التوصيل للجميرا، كم ياخذ وقت؟»
- Gulf Arabic, as it is spoken
- Three minutes a conversation
- Nothing is really booked
- No account
How the demo works
Pick who to talk to
A clinic, a real-estate office or a restaurant. Each has its own voice and one job to do, in Emirati Arabic.
Allow the microphone
Your browser asks once. The call runs over LiveKit, in this tab, and ends on its own after three minutes.
Talk
Say what you want in your own words, dialect and all. The agent hears it, works it out and answers. What it heard appears as you go, and the booking shows up as a card when it is made.
What is on the line
- Hearing
- nutq speech to text, streaming, built for Gulf dialect rather than Modern Standard
- Speaking
- nutq text to speech: Alia, Sultan and Amna, from the voice library
- Thinking
- A language model given one job per persona, and no more
- Transport
- LiveKit: WebRTC in the browser, an agent worker on the other side
- Build it
- The nutq plugin for LiveKit Agents, the same one the docs describe
- Limits
- Three minutes a conversation, a handful of conversations an hour per visitor
- Privacy
- Nothing is recorded: the audio is transcribed as it arrives, and the booking is pretend
The live demo, answered
Is this a real booking?
No. The clinic, the office and the restaurant are made up, and nothing is reserved, viewed or delivered. The conversation is real: the agent hears you, decides, and answers.
Which dialect does it understand?
Emirati, and the wider Gulf. Say it the way you would on the phone; it is built for that, not for Modern Standard Arabic read aloud. English words in the middle of a sentence are fine.
Why does it need my microphone?
To hear you. The browser asks once, the audio goes to the agent over LiveKit for as long as the call lasts, and the microphone is released when it ends.
Why did it end, or say every line is busy?
A conversation is capped at three minutes, and a few agents run at a time, so at a busy moment you may be asked to try again in a few minutes. Each visitor gets a handful of conversations an hour.
Can I build this for my own business?
Yes. The agent is a LiveKit Agents worker using the nutq plugin for speech to text and text to speech, with your own voice and your own instructions. The docs show how.