Your patients don'tread your forms.They'll talk to this.
Four in ten Pakistani adults cannot comfortably read a form, a menu or an SMS — but almost all of them will answer a phone. Montegritty builds voice agents that hold the call in Urdu, wires them into the systems you already run, and stays accountable for the number they were hired to move.
Anyone can generate an agent. Almost nobody runs one.
The hard part was never producing a voice. It is producing one that survives a real caller, in a real language, wired to a real system — and then proving what it did.
You can hear ours working
Eight agents, eight full recorded calls in Urdu, with a bilingual transcript you can follow line by line. Not mock-ups, not a template list — conversations, synthesised end to end. Listen before you believe anything else on this page.
Listen to a call →Not locked to one voice vendor
Uplift AI is our primary engine because it is the best Urdu on the market. But ElevenLabs, Vapi and open-source models are all on the table, chosen per project — and open-source self-hosted when the data cannot leave your building. A product built on a single vendor inherits that vendor's roadmap and outages.
See the engines →We build it and run it
You get an agent wired into your booking system, your records, your order feed — and a dashboard built from its own calls. Not a link you are left to figure out. We agree one number up front and are accountable for it.
How a project runs →Listen before you believe us
Every voice here is a working Montegritty agent handling a real call in Urdu, synthesised end to end. Press play, step through with the arrows, or open any of them for the full recording and a transcript you can follow line by line.
Hospitals, Clinics & Diagnostic Labs
Ayesha
Appointment & No-Show Desk
Fills the empty chair. Confirms, reschedules and preps every patient.
Hear the full call →Three sectors, in depth
Not every industry — three where the calls are highest-volume, the literacy gap is widest, and we have done the work. Banking and telecom are already crowded with contact-centre vendors; we would add nothing there.
Healthcare
Clinics, hospitals, labs and diagnostics
Appointment confirmation, pre-arrival intake and follow-up, in the language the patient speaks.
What we build for healthcare →Education
Schools, colleges and training networks
Admissions enquiries, fee reminders and absence follow-up — reaching the parent, not the schoolbag.
What we build for education →Front desk
The businesses that live on the phone
Order confirmation, bookings, enquiries and reminders for the desks that never stop ringing.
What we build for front desk →One agent. Whichever engine suits it.
Voice AI is assembled from three swappable parts — speech in, reasoning, speech out. We pick each per project instead of taking whatever one vendor happens to offer, which is why a confidential deployment and a public helpline can be the same agent on different rails.
Uplift AI
In productionPrimary — Urdu & regional
The best Urdu speech on the market, built in Pakistan on native-speaker data. Our default for anything a Pakistani caller will hear.
ElevenLabs
In productionEnglish & multilingual
Where an agent serves English-speaking or international callers and the priority is naturalness across accents.
Vapi
AvailableTelephony orchestration
Call routing, SIP and phone-number provisioning when an agent has to answer a real number rather than a browser.
Open-source, self-hosted
On requestWhen data cannot leave
Whisper and open speech models running on your own hardware. No third party ever receives a recording, and there is no per-request bill.
Which engine a project uses is a decision we make with you and can revisit later. Nothing about the agent is rewritten when the engine underneath it changes.
You also get a dashboard
Every client gets reporting built from their own calls — not a spreadsheet somebody assembles on Monday. Here are three of the shapes we ship, depending on whether you are running a support line, a calling campaign, or a regulated operation that has to prove what was said.
Operations
Live call volume, outcome mix and per-agent performance, rebuilt from every call as it ends. This is the one shipped in Urdu — open it and watch the feed arrive.
Sample data. Every dashboard is built from the agent’s own transcripts and outcomes, so the system answering the phone is the system reporting on it.
Confidential operations, covered
Most voice AI is a wrapper around somebody else's API, which means your patients' conversations travel to a third party before you hear them. That is a non-starter for a hospital — so it is not how we deploy for one.
It can run entirely inside your walls
Open-source speech models self-hosted on your own servers or private cloud. Recordings and transcripts never leave your environment, and there is no per-request call to anyone else's API.
You decide what is kept
Retention is a setting, not our policy. Keep every recording for audit, keep transcripts but discard audio, or redact identifiers before anything is written down. Deletion is real deletion.
The agent only knows what you show it
Integrations are scoped to the fields a call actually needs. An appointment agent can see a calendar; it cannot see your patient records, and it cannot be talked into reading one out.
Every call is on the record
Full transcripts, timestamps and outcomes for anything an auditor might ask about — including what the agent said, not just what it did.
Pakistan runs on voice, not text
This is not a bet on a trend. It is arithmetic about a country where almost everyone can answer a phone and four in ten adults cannot comfortably read the message you sent them instead.
national literacy. In Khyber Pakhtunkhwa it is 51.1%, and female literacy nationally is 52.8%.
mobile penetration, with smartphone usage at 71.6%. Connectivity has outrun literacy.
annual growth in healthcare voice AI globally — the fastest-growing vertical in the category.
of Pakistani online orders are cash on delivery, and roughly one parcel in five comes back undelivered.
Every call these agents make is already being made today — badly, expensively, and only to a fraction of the list. That makes this a cost-and-coverage argument rather than a new-behaviour one, which is the easier case to win.
Which call would you hand over first?
That is the whole scoping conversation. Tell us the call your team makes most often and we will tell you, within a week, whether an agent is worth building for it — and what it would take.