A Whizz Tech product · voice agents
Answers the phone.
In their dialect.
Whizz Voice builds voice agents that pick up in under a second and talk like a local — Egyptian, Saudi, Gulf, Levantine and 10 more Arabic dialects, plus 32 languages. They listen, reason over your knowledge, call your tools, and log every call. One price, all-in, per minute.
$5 free credit · web calls in minutes · no card
Idle · pick a dialect or language and place a call
Scripted replay · the agent answers in the selected dialect or language and handles the request end to end.
One call, eight moves. No hold music.
Every call runs the same loop — answer → route → listen → act → log — with first audio back in about half a second. This is the loop, stage by stage.
Ring — picked up before the third ring
A web widget, a platform number, or your own SIP trunk rings the agent. It answers in well under a second — no queue, no 'press 2 for Arabic', no hold music.
0.4s to answer
channel web_widget · concurrency 6 / 15
no queue · no IVR · no hold music
session opened → agent "Support · AR/EN"
One call, eight moves. No hold music.
Every call runs the same loop — answer → route → listen → act → log — with first audio back in about half a second. This is the loop, stage by stage.
01 · ring
Ring — picked up before the third ring
A web widget, a platform number, or your own SIP trunk rings the agent. It answers in well under a second — no queue, no 'press 2 for Arabic', no hold music.
inbound — +966 · web · sip0.4s to answer
channel web_widget · concurrency 6 / 15
no queue · no IVR · no hold music
session opened → agent "Support · AR/EN"
02 · greet
Greet — in the caller's own dialect
The agent opens with your scripted first line, spoken by a dialect-native voice. An Egyptian caller hears Egyptian; a Riyadh caller hears Saudi. The greeting sets the language before the caller says a word.
first_message — dialect-nativevoice Noura · dialect saudi · gender female
حياك الله في خدمة عملاء وِز، معك نورة — كيف أقدر أخدمك اليوم؟
spoken in-dialect — not MSA, not translated
03 · route ⑂ ar / en
Route — Arabic forks from the rest
The first utterance decides the lane. Arabic and code-switched calls go to the Dialect Engine; English and 31 other languages go to the Global Engine. The router is invisible to the caller — one agent, two engines behind it.
router — one utterance, two lanes▲ Dialect Engine
ar · 14 dialects
code-switch aware
Global Engine
en +31 languages
low-latency
detected ar-EG → routed to Dialect Engine
04 · listen
Listen — full-duplex, barge-in ready
Streaming speech recognition transcribes as the caller talks — dialect as spoken, digits code-switched to English, no flattening to MSA. The caller can interrupt mid-sentence and the agent yields, like a real conversation.
asr_stream — live, barge-incaller الرقم four-nine-two-zero-seven بس تأخر كتير
partial streaming · 120ms behind speech
digits captured in en · dialect kept as spoken
caller interrupted → agent yielded (barge-in)
05 · think
Think — grounded in your knowledge
A frontier language model reads the live transcript against your persona, knowledge base, and rules, then decides what to say and which tool to call. Retrieval keeps answers grounded in your content, not the model's guesses.
reasoning — grounded turnintent track_order · entity order=49207
kb shipping_policy.md → 2 chunks retrieved
plan lookup_order → if stalled → expedite
answer grounded in your content — not a guess
06 · speak
Speak — first audio in ~480ms
The reply streams back as dialect-native speech while the model is still finishing the sentence. Latency stays conversational, so the caller never hears the dead air that gives a bot away.
tts_stream — first audio 480ms480ms first audio
لقيت الطلب، هو في الطريق ووصوله المتوقّع خلال يوم…
streamed while the model finished the sentence
conversational latency — no bot dead-air
07 · act
Act — tools, transfers, bookings
The agent calls your webhooks and functions mid-call: look up an order, book a slot, send an SMS, or hand off to a human with full context. It doesn't just talk — it does the task the call was about.
tool_calls — mid-call⚙ lookup_order(id=49207) → in_transit · ETA 26h
⚙ expedite_shipment(id=49207) → ok
⚙ send_sms(confirmation) → queued
it did the task — not just talked about it
08 · log
Log — every call, receipts included
The moment the call ends you get the recording, a diarized transcript, the tools it called, sentiment, an outcome, and a webhook to your stack. Nothing is a black box — every call leaves a paper trail.
call.completed — webhook payloadduration 0:52 · outcome resolved · handoff none
sentiment ▲ positive · csat 5 / 5
artifacts recording · transcript · tool_log
POST call.completed → your endpoint
Not Arabic.
Their Arabic.
Most voice stacks flatten Arabic to one robotic MSA voice. We serve 14 dialects with native voices — the caller hears their own, and the agent code-switches to English the way people actually talk.
Egyptian
مصري
4 voices
Press play to hear a real sample of any voice. Inside a call these voices speak your knowledge base in the dialect's colloquial register, and code-switch to English.
And 32 languages, worldwide.
Arabic dialects are the moat — but the Global Engine answers your European and international callers too, at the market's lowest all-in rate. Eight core languages get the fastest, cheapest path; 23 more are one click away.
See every language & dialectCore languages · fastest & lowest cost
23 more languages
Three calls to
a live agent.
Create an agent, give it a widget or a number, and let it take or make calls. REST from end to end — no black-box console required. Test keys (wz_test_) ship with the trial.
- — Agents · phone numbers · calls · campaigns — all REST
- — wz_live_ bearer keys · signed webhooks (whsec_)
- — The router picks the engine — you never name a vendor
# 1 · Create a voice agent — Arabic + English, dialect-nativecurl -X POST https://voice.whizztech.ai/v1/agents \-H "Authorization: Bearer wz_live_k3XJ************Qf9A" \-H "Content-Type: application/json" \-d '{"name": "Support · AR/EN","languages": ["ar", "en"],"dialect": "saudi","voice": "noura","first_message": "حياك الله في خدمة عملاء وِز، كيف أقدر أخدمك؟","persona": "Warm, concise support agent. Confirm the order before acting.","tools": ["lookup_order", "expedite_shipment", "send_sms"]}'# HTTP/2 201 — the router picks the Dialect Engine for Arabic automatically# {# "id": "ag_7Q2c51b84a07",# "status": "draft",# "engine_route": "dialect",# "lang_tier": "arabic"# }
Plug and play.
Or bring your carrier.
Every agent works over the web widget out of the box — zero telephony cost. Add a phone number when you need one; the PSTN leg is billed at cost, never marked up.
01
Web widget
Embed one script tag and the agent takes WebRTC calls in the browser. No number, no carrier, no per-minute telephony fee — just your all-in minute rate.
02
Platform number
Buy a local number in a click — KSA, UAE, Egypt and more. $2/mo per number, $0.015/min PSTN, both at cost. Inbound and outbound on the same line.
03
Bring your own carrier
Point your existing SIP trunk or Twilio number at the agent and we charge $0 for the telephony leg. Keep your carrier contract, keep your numbers.
04
Outbound & campaigns
Fire a single outbound call or a batch campaign over the API, with per-recipient variables and concurrency you control. Every call logs back to your stack.
One rate.
All in.
The speech engine, orchestration, knowledge base, tools and analytics are all in the per-minute price. Telephony is the only separate line, billed at cost.
Elsewhere, a $0.05 headline becomes $0.31 once you add the speech, the LLM, the platform fee. Ours is the price.
Starter
$29 /mo
$29 call credit · 5 lines
3 agents · web + 1 number
Growth
Most popular
$99 /mo
$105 credit · 15 lines
20 agents · campaigns
Scale
$299 /mo
$330 credit · 40 lines
volume rates unlock
Lite $0.05 · Global $0.09 · Extended $0.13 · Arabic dialects $0.15 per minute — all-in retail. Telephony billed at cost. $5 free credit to start, no card.
Full pricing & calculator→Stop making customers press 2 for Arabic.
They called to be understood. Give them an agent that answers in their dialect, does the task, and hangs up having actually helped. $5 free credit, no card.