Najva · Voice AI for enterprises
Najva is a voice AI platform built for Persian first. Its agents answer and place phone calls, hold natural conversations and carry out tasks in your systems. Its analytics turn every recorded call into transcripts, quality scores and follow-ups.
Enterprise soft launch · by demo request
Why we built Najva
Najva started with a problem we kept meeting in contact centers and service teams. Calls were recorded and never heard again, operators spent their day on routine requests, and the voice tools on offer weren’t built for the language customers actually speak.
Persian has a large market but very little high-quality speech data. Most voice models learn mainly from English and other well-resourced languages, and it shows in their Persian.
Phone lines are narrowband and noisy. People speak colloquially, with accents, numbers, names and English terms mixed in. Models tuned on clean, read speech struggle on exactly the calls that matter.
The strongest voice services are cloud APIs abroad: data leaves the organization, cost grows with every minute and Persian quality varies. Banks, insurers, utilities and public bodies need a system on their own servers, and good ones were rare.
Our answer
So we built Najva: Persian first, tuned for real phone audio, able to run entirely on your servers, and connected to the phone and business systems you already use.
What Najva does
Customers tell you what they need, what went wrong and what they think of you, mostly on the phone. Najva answers those calls and makes sure what was said doesn’t get lost.
Answer and place calls over your phone system, and get the job done.
Review every recorded call, not a sample, and act on what you find.
The core
Every Najva product combines the same four building blocks with a language model, so improvements to the core reach every agent and every report.
Accurate transcription of spoken Persian: the base for search, analysis and archives.
A natural, human-like voice for replies, announcements and audio content.
Who said what in a conversation: the backbone of call reviews and meeting minutes.
Recognizing a speaker by voice, for verification, fraud checks and personal service.
Building an agent
A voice agent is configured, not coded. Your team decides how it sounds, what it asks, what it collects and where the result goes.
Name, instructions and voice. Choose how it handles interruptions and when it decides the caller has finished speaking.
The steps of the conversation, its branches, and when to hand the call to a person.
The fields the agent must collect, each with a type and whether it’s required.
Map each field to your endpoint and test the connection before going live.
Connect your SIP trunk and number, run the final checks and publish.
The agent takes calls. Every call is kept with its recording, transcript and extracted data.
One call, end to end
A customer calls a restaurant. Najva answers, takes the order and confirms it. Halfway through, the customer changes their mind. The agent stops talking, listens, updates the order and carries on.
When the call ends, the order is already in the restaurant’s own system, sent as structured data through its API. Nobody had to type it in.
POST /v1/orders
{
"items": [
{ "name": "Margherita pizza", "qty": 1 },
{ "name": "Lemonade", "qty": 2 }
],
"delivery": "saved_address"
}Restaurant system201 Created
Call analytics
Every finding links back to the exact words and moment in the call, so reviewers can check the evidence in seconds.
The call becomes text, separated by speaker and time-stamped.
Reason for the call, topics and details, each linked to the words that support it.
The agent’s performance is scored against your quality checklist.
Word choice, meaning, voice and call data combine into one risk score.
Flagged calls go to a review queue. Najva flags; a person decides.
In production
Average results across our contact center customers, after they moved call review from manual sampling to every call.
Deployment and security
Customer conversations are sensitive business data. Najva can run entirely inside your network, or start in the cloud and move on-premise as volume grows. Both options are available to soft-launch customers.
| Cloud | On-premise | |
|---|---|---|
| Setup | Fast, with no infrastructure to run | GPU servers and an initial setup |
| Where data lives | Processed by the cloud service | Stays on your own servers |
| Cost | Pay per use, which grows with volume | Fixed, and the most economical at high volume |
| Persian quality | Depends on the underlying service | Improves with training on your own Persian data |
| Best for | Pilots and smaller volumes | Banks, insurers, utilities, government and high volume |
Accounts are created by your administrators.
Each person sees only what their role allows.
Configurable limits for every account.
Access levels and secure storage for every conversation.
Availability
Najva is in soft launch. We are working closely with a small number of enterprise customers before opening it more widely.
Available to enterprises by demo request. Voice agents and call analytics are in soft launch; meeting minutes and on-premise installation are available to soft-launch customers, not yet publicly released. We set up each deployment with your team: telephony, business systems and the vocabulary of your industry.
Request a demoA general panel that anyone can sign up for, to build a voice agent, try it and put it to work. It isn’t available yet.
Najva makes sure none of it is lost. See it on your own calls.