It speaks textbook Arabic
A voice reading formal Modern Standard Arabic sounds like a news bulletin on a customer-service line. Callers notice in the first sentence, and many switch to English or hang up.
AI voice agents · AI receptionists · Arabic and English
A voice agent that sounds like it is from here.
Most voice AI is built for American English and an American phone line. Your callers speak Gulf or Levantine Arabic, switch to English mid-sentence and expect an answer before the silence gets awkward. We build voice agents for UAE and Saudi businesses that handle exactly that, connected to the systems that hold the answers.
What is an AI voice agent?
An AI voice agent is software that holds a spoken conversation. It listens, understands what the caller wants, answers out loud and acts on it: booking an appointment, checking an order or a balance, taking a message, or passing the call to a person with the transcript attached. It runs on a phone line, in an app or on a website.
Businesses we’ve worked with
Live on stage, not just in a demo
For Emirates Group, with KalaMena AI, we built Maryam 3.0, a virtual AI panelist that answered the moderator out loud at ForsaTEK for a full hour, responding in 100 to 200 milliseconds and recognising each panelist by voice.
Arabic as people actually speak it
Gulf, Saudi, Levantine and Egyptian Arabic, English, and the switching between them. We pick the voice by testing it on native speakers of your callers’ dialect, not by reading a vendor’s language list.
It does the job, then writes it down
Bookings land in your calendar, orders and tickets in your systems, and every call leaves a transcript and a summary your team can search.
Where voice agents go wrong
Six reasons callers hang up on voice AI.
Voice is less forgiving than chat. A slow, robotic or confused agent loses the caller in seconds, and these are the usual causes.
It is too slow to feel like a conversation
Every step, hearing the caller, thinking, and speaking, adds delay. Past a second or so of silence the caller starts repeating themselves and the call falls apart.
It cannot tell when the caller has finished
People pause mid-sentence, trail off, or say "yalla" and keep going. An agent that jumps in early talks over the caller; one that waits too long leaves dead air.
It mishears the details that matter
Names, phone numbers, plate numbers and order references, read out in a mix of Arabic and English, are where general speech recognition fails. A booking under the wrong number is worse than no booking.
It can talk but cannot act
If the agent cannot see your calendar, your orders or your customer records, it can only take a message. The value is in completing the request while the caller is still on the line.
There is no clean way to a person
Some calls need a human. Without a handover that passes the caller on with the conversation so far, they have to start again, angrier than before.
How we build it.
One kind of call, live and holding up, before we add the next. A voice agent that handles one job well beats one that handles ten jobs badly.
Pick the call
We listen to how your calls actually go and find the one that repeats most: bookings, order status, opening hours, a first line for support. That one goes first.
Choose the voice with native ears
We shortlist voices and play them blind to native speakers of your callers’ dialect, then tune the wording until it sounds like a person from here rather than a translation.
Connect the systems
Calendar, booking system, customer records, order database. The agent answers from your live data and writes the outcome back, which is what makes the call worth automating.
Go live on one line, then widen
One number or one entry point first, with a human fallback, reviewed against real calls in the first weeks, then extended once it is genuinely holding up.
What ships
What a voice agent build includes.
Everything needed to answer real calls in production, not a demo that only works when the room is quiet.
Call design from your real callsArabic dialect and English voices, tested blindTurn-taking tuned for natural pausesBooking, order and account lookupsCalendar, CRM and database integrationHuman handover with the transcriptTranscripts and call summariesPhone line, app or website entry pointsMonitoring of every callMonthly tuning against real calls
What we will not build
We do not build unsolicited outbound calling campaigns or agents designed to hide that the caller is talking to an AI. Callers are told up front, and the agent stays inside what your business would say on a recorded line. If your plan is mass cold calling, we are the wrong partner.
Questions people ask.
What is an AI voice agent?
An AI voice agent is software that holds a spoken conversation: it listens, understands the request, answers out loud and acts on it, such as booking an appointment, checking an order or handing the call to a person with the transcript. It can run on a phone line, inside an app or on a website.
Can an AI voice agent speak Gulf or Emirati Arabic?
Yes, if the voice is chosen for it. Many voices that claim Arabic speak formal Modern Standard Arabic or carry the wrong accent. We test candidate voices blind with native speakers of your callers’ dialect and adjust the wording until it sounds local, before anything goes live.
What is the difference between an AI receptionist and an AI voice agent?
An AI receptionist is a voice agent with one job: answering incoming calls, handling routine requests such as bookings and opening hours, and routing everything else to the right person. A voice agent can also handle longer conversations, such as support, order changes or qualification, and can work in an app or on a website as well as on the phone.
How fast does it respond?
Fast enough to feel like a conversation, which is the hard part of the work. On stage at Emirates Group’s ForsaTEK, the AI panelist we built with KalaMena AI answered in 100 to 200 milliseconds. On a phone line the network adds some delay, so we measure it on your real line before launch.
Can it work with our existing phone number?
Usually, yes. Calls can be forwarded to the agent or routed to it from your existing phone system, so customers keep calling the number they already know. We confirm what your provider supports on the first call.
How long does it take to build, and what does it cost?
A first call type usually goes live within a few weeks. Each build is scoped to a fixed price after a short discovery call, so you know the number before you commit, and running costs depend on call volume, which we estimate with you up front.
Who builds AI voice agents in the UAE?
Hephon AI (Hephon L.L.C-FZ) is an AI agency headquartered in Dubai. We build Arabic and English voice agents, AI receptionists and chat agents for businesses in the UAE and Saudi Arabia, and stay on after launch to tune them against real calls.
Start with the call your team answers most.
Tell us which calls fill your team’s day and we will tell you honestly whether a voice agent can take them.
Prefer email? WhatsApp us or email hello@hephon.ai — we reply within one business day.