Self-hosted Voice AI · Built in Saudi Arabia
Arabic & English voice agents that never leave the Kingdom.
A self-hosted voice AI receptionist for Saudi businesses — speech recognition and reasoning run on our own server in Dammam, not a US cloud. It answers, books, and follows up in Arabic and English, embedded right on your website.
It's like phone lines, not a phone bill. You buy simultaneous conversation capacity — not minutes. One flat price, no metered billing, no surprise usage charges.
- Response time
- ~2.8s
- Hosted in
- Dammam
- Region
- me-central2
Live on a shared Dammam node
Design target: ~12 concurrent conversations per node
Every lit line is one live conversation. Capacity is sold in lines, not minutes — so a client with two receptionists' worth of simultaneous calls buys two lines, once, and never worries about a metered bill again.
The problem
Every unanswered call is a customer you already paid to attract.
Calls go unanswered outside business hours
A missed call at 9pm is a lead your competitor answers first. Most Saudi SMBs can't staff reception around the clock, so after-hours and lunchtime inquiries simply go to voicemail — or nowhere.
Bilingual reception is hard to staff consistently
Callers switch between Arabic and English mid-sentence. Finding — and keeping — front-desk staff who handle both fluently, every shift, is a real hiring problem, not a training one.
Sending customer data outside the Kingdom is a real risk
Most AI voice tools route your callers' conversations through US cloud infrastructure. For PDPL-conscious businesses, that's a compliance question no one wants to answer with "we're not sure."
How it works
From spoken question to answered call, in under three seconds.
No routing through a US cloud for speech or reasoning. Every step below runs on our own GPU in Dammam, except the one clearly marked exception.
On your website
Widget answers the call
A caller speaks in Arabic or English through the voice widget embedded on your site — no app, no download.
Embeds on unlimited pages
Google Cloud · Dammam
Audio stays in-Kingdom
The stream goes straight to our server in me-central2. It does not transit a US hyperscaler for speech or reasoning.
g2-standard-8 · NVIDIA L4 24GB
Self-hosted STT
Whisper transcribes it
Whisper large-v3, running on our own GPU, converts speech to text in either language.
Whisper large-v3
Self-hosted LLM
Qwen2.5 decides the reply
A 7B-parameter model reasons over your business knowledge and drafts the response — in ~2.8 seconds end to end.
Qwen2.5-7B · ~2.8s
Response
The agent replies — and can act
The caller hears the answer, and the agent can book, log a lead, or hand off to CRM/WhatsApp on higher tiers.
Neural voice via Microsoft TTS today*
* Speech recognition and the language model are fully self-hosted in Dammam. The neural text-to-speech voice you hear currently still calls Microsoft's TTS service — a fully self-hosted TTS path is on our roadmap for clients with strict PDPL requirements. See FAQ.
Why self-hosted, in-Kingdom, matters
Built for PDPL data-residency alignment — stated plainly, not oversold.
Most voice-AI vendors route speech and reasoning through US cloud infrastructure. Ours doesn't: speech-to-text and the language model both run self-hosted, on our own server in Dammam. That's a meaningfully different data path for a Saudi business handling customer conversations.
We're precise about where the one exception sits today, and we won't claim a compliance status a court or regulator should be the one to confirm. Final PDPL, DPA, and ZATCA/VAT treatment is something we work through with your counsel, per client — not a one-line badge.
Status, as of today
-
Speech-to-text
Self-hosted (Whisper large-v3) on our own GPU in Dammam.
-
Language model
Self-hosted (Qwen2.5-7B) — no prompts sent to a third-party LLM API.
-
Hosting location
Google Cloud, Dammam region (me-central2) — in the Kingdom.
-
Text-to-speech
Currently calls Microsoft's neural TTS for the synthesized voice. Fully self-hosted TTS is on our roadmap for strict PDPL cases.
-
DPA / ZATCA / VAT treatment
Confirmed per client with Saudi legal counsel — not a blanket guarantee made on this page.
Pricing
Phone lines, not a phone bill.
You buy simultaneous conversation capacity, languages, branches, and features for one flat monthly or annual price. No metered minutes, no usage-based invoice.
Basic
One line, one language, one location
or SAR 9,900/yr · setup SAR 2,500
1 line
- Languages
- Arabic or English
- Sites / branches
- 1 website
- Agents
- 1 agent
- Designed for
- ~30 calls/day
- Receptionist, FAQ answering, lead capture
- Business-hours support
- Shared in-Kingdom node
Professional
The bilingual, multi-site default
or SAR 29,000/yr · setup SAR 5,000
2 lines
- Languages
- Arabic + English
- Sites / branches
- Up to 3 sites
- Agents
- 2 agents
- Designed for
- ~120 calls/day
- + Bookings and analytics dashboard
- Priority support
- Shared in-Kingdom node
Enterprise / Sovereign
Dedicated node, sized to you
Custom annual terms · setup from SAR 15,000
Dedicated node
- Languages
- Arabic + English + custom
- Sites / branches
- Unlimited
- Agents
- Unlimited
- Designed for
- Sized to client
- Telephony + full integrations
- Custom Arabic model on request
- 24/7 support, 99.9% SLA, signed DPA
Prices exclude 15% VAT (ZATCA e-invoicing). Annual = pay 10 months, get 12 (2 months free). 12-month minimum term on annual and discounted plans.
Customize your plan
Add exactly what you need, nothing you don't.
Not sure which tier?
Answer 3 quick questions.
Roughly how many calls or inquiries do you get each day?
Who it's for
Built for steady, everyday inbound volume.
The common thread across every good-fit business below: a steady trickle of questions across the day, where a rare second caller waiting a moment is perfectly fine.
Health & personal care
- Clinics & dental centers
- Salons, spas & beauty businesses
Book appointments and answer insurance/availability questions in Arabic or English, day and night, without pulling reception staff off the floor.
Property & professional services
- Real-estate agencies & brokers
- Law firms & consultancies
- Accounting, marketing & professional-services agencies
Qualify inbound leads and answer routine questions immediately, so a call at 9pm doesn’t wait until 9am to get a reply.
Food & hospitality
- Restaurants & cafés
- Small hotels & hospitality
Take reservations and answer menu/availability questions during peak hours, when every line is already busy.
Retail & vehicles
- Car dealerships & service workshops
- Small e-commerce & retail
Handle order status, stock, and service-booking questions without adding headcount for a seasonal spike in calls.
Education & trades
- Training institutes & schools
- Contractors & trades
Answer enrolment, pricing, and scheduling questions consistently, in the language the caller opens with.
Not what this is built for
On the shared-node tiers above, we deliberately don't sell to businesses expecting call-center-scale saturation. If that's you, the right conversation is Enterprise / Sovereign on a dedicated node.
- Call centers & BPOs
- Telecom / big-bank mass support
- Emergency hotlines
- Large-scale outbound telemarketing
- SAMA-regulated banking operations
- Anyone expecting 50+ simultaneous callers or 24/7 saturation
FAQ
Straight answers, including the caveats.
Including the one about billing, the one about Microsoft, and the one about what "compliant" actually means here.
Is this billed per minute, like a call center service?
No. You buy a fixed number of simultaneous lines, languages, sites, and features for one flat monthly or annual price — the same way you'd buy phone lines, not a phone bill. There is no per-minute metering and no usage-based invoice at the end of the month.
What happens if I outgrow my tier's call volume?
Each tier lists a “designed for” daily-volume guide (for example, ~120 calls/day on Professional). It's fair-use guidance for choosing the right tier, not a billed limit. If you consistently exceed it, we'll recommend an upsell to the next tier or an add-on line — you will never see a surprise overage charge.
Where is our data actually processed and stored?
Speech-to-text (Whisper large-v3) and the language model (Qwen2.5-7B) run on our own server in Google Cloud’s Dammam region (me-central2) — nothing is routed through a US hyperscaler for those steps. This is built for PDPL data-residency alignment. See the compliance note below for the one exception (text-to-speech) and what still needs sign-off from counsel.
How long does setup take, and what does the setup fee cover?
Setup is a one-time, non-refundable fee that covers onboarding, configuring your agent(s) for your business, and training the assistant on your FAQs, services, and tone of voice. Typical setup runs from a few business days (Basic) to a couple of weeks (Premium/Enterprise, depending on integrations like CRM or WhatsApp).
Does the AI actually speak, or does it call Microsoft for that part?
Speech recognition and the reasoning/response are fully self-hosted on our Dammam server. Today, the neural text-to-speech (the synthesized voice you hear back) still calls Microsoft’s TTS service — we are transparent about this. A fully self-hosted TTS path is on our roadmap for clients with strict PDPL requirements. See the compliance note for detail.
Can the assistant work across more than one branch or location?
Yes — Professional supports up to 3 sites/branches, Premium up to 10, and Enterprise is unlimited across a dedicated node. Each tier’s branch allowance is listed in the pricing table above.
Is this suitable for a call center or high-volume outbound campaigns?
Not on the shared-node tiers. Basic through Premium are built for steady, everyday inbound volume — clinics, agencies, restaurants, and similar businesses that rarely have more than a few callers at once. If you run a call center, expect 24/7 saturation, more than ~50 simultaneous callers, or handle heavily regulated data (e.g. SAMA-regulated banking), that belongs in an Enterprise / Sovereign conversation on a dedicated node.
Get started
See it answer a real call in Arabic and English.
Book a 20-minute demo on your own scenarios, or send us a few details and we'll size the right tier for you before we talk.
- Response time
- ~2.8 seconds
- Hosting
- Dammam · me-central2
- Languages
- Arabic + English