Voice AI agent for inbound and outbound calls
Inbound calls are answered, qualified, and routed — including after hours and overflow. Outbound calls handle reminders, confirmations, follow-ups, and customer reactivation. The agent speaks the caller's language, records the outcome, and pushes structured data to your systems through your APIs.
A sample script based on our production confirmation scenario.
Agent: Hello, this is KernoVoice calling on behalf of City Medical Center. I'm confirming your appointment with Dr. Nowak tomorrow at 10:30. Can you make it?
Patient: Yes, I'll be there. Do I need to bring anything?
Agent: Please bring your ID and insurance card. We've also sent a reminder to your email. See you tomorrow.
Patient: Perfect, thank you.
Walkthrough
The agent answers your questions and hands off to a human specialist when needed. Latency is the hardest problem in voice AI — judge it live, not from a table.
Results
From our first production deployments — healthcare appointment workflows, June–July 2026. Client names are under NDA; sample sizes and definitions are below, because percentages without them mean nothing.
~90%
Outbound confirmation calls reached ~90% of patients within a calling window of about an hour, most on the first attempt — measured across 2,500+ patients over two weeks (daily range 83–97%). "Reached" means the patient answered; we count up to two dial attempts.
~80%
Confirmed, rescheduled or cancelled — recorded straight in the clinic's scheduling system. Median confirmed call: 18 seconds of connected conversation (n = 1,812).
4 / 5
In July, 4 out of 5 answered inbound booking calls were completed without a human — what the industry calls containment (500+ answered calls). The rest were transferred to staff by routing rule: medical questions and cases outside the booking scope.
Measurement basis: production database. Outbound is counted per patient with up to two dial attempts; inbound covers the supported booking workflows, not general reception.
What it does
A KernoVoice agent answers calls, qualifies requests, sends reminders, follows up, routes conversations, and captures structured outcomes for your team. Complex integrations are built around the client workflow instead of sold as a universal connector for everything.
Booking, rescheduling, cancelling and confirming appointments happen directly in the client's own management system, through the same administrative API its front-desk software uses. No parallel copy of the schedule: availability, prices, practitioners and patient records are read from the client system at call time, and every change is written back to it — the agent and the front desk work from the same data.
Every write is gated behind a confirmation step: the agent reads the booking back and commits only after an explicit spoken confirmation. Every write attempt — committed or not — is recorded in an audit log.
A single call can handle multiple appointments, each confirmed separately — in production, one caller booked three appointments in a call under two minutes.
Pilot languages: English, Polish, and Russian. Additional languages depend on the chosen speech stack — in the live demo, try switching to German mid-call.
Security and data
KernoVoice is early-stage, so we keep what is live today separate from what is on the roadmap — no badges, no future capabilities written in the present tense.
Live today
Calls run on a managed instance, or on a dedicated instance for pilots when you need stronger isolation, with your keys. Your data is not pooled into a shared platform.
Transport encryption protects application traffic; storage encryption comes from the hosting and database layers in the deployment. The data flow is documented and a DPA is available on request.
For production clients, speech recognition and synthesis run on infrastructure we operate — not routed through third-party speech APIs. We have also run live production calls with the LLM hosted in the same controlled environment.
Writes to client systems are gated behind a spoken confirmation step and recorded in an audit log — every attempt, committed or not.
KernoVoice does not claim formal third-party security certification. Security documentation and DPA terms are handled directly with clients.
Placement is yours to control: managed cloud, a dedicated environment, or your own infrastructure — in the region you choose.
Security questions: securitykernovoice.com. DPA available on request.
Common questions
Company
KernoVoice is built by Kerno Ops sp. z o.o., a Warsaw-based limited-liability company (LLC) bringing telecom, ISP, and contact-center operating experience into voice automation. The product is shaped by 20+ years of hands-on work with networks, telephony, and operational processes — operator judgment, not only application engineering.
Current stage: first production deployments — inbound booking with live write-back to the client's scheduling system, and outbound confirmation calls.
teamkernovoice.comPricing depends on the model stack (cloud or self-hosted), integration depth, scenario complexity, languages, and the level of quality controls — automated call scoring, guardrails, human review. We start from the workflow and constraints, then price the deployment. We target live pilot calls within 1–2 weeks of workflow review, given API access and timely client decisions.