AI & Voice / Voice agents

On-premise AI voice agents for call centres

A voicebot that answers real phone calls, lets callers interrupt and hands the call to a person with a summary. Installed on your servers.

Talk about a deployment

A voice agent that answers the phone for real

Most calls to a support line are about the same few things: the connection is down, a payment failed, where is my order. An AI voice agent can answer those calls at any hour, with no queue, and leave your staff the cases that genuinely need a person.

We do not build polished demos that fall apart on a real line. Our voice agent already answers subscriber calls for an internet provider, every day. Everything on this page comes from that work.

What the agent does

  • Listens and talks at the same time. Callers can cut in mid-sentence, as they would with a person. Getting turn-taking right did more for callers than any change to the language model.
  • Handles routine calls using your business rules and your data.
  • Hands the call to a person when the case needs one, with a short summary so the caller never has to repeat themselves.
  • Logs every call in your CRM or ticketing system, so nothing lives only in a recording.
  • Respects access boundaries: what the agent may see and what only trusted staff can see is part of the architecture from day one.

On-premise voice AI

By default the agent is installed on your servers or in a private cloud. Calls and customer data stay inside your network, with no third-party cloud in the conversation. For banks, insurers, healthcare, telecoms and the public sector, that is often the difference between being allowed to use voice AI or not.

Recordings follow a written retention rule, and we keep a log of every change to the system.

Built for your phone system

Our team has more than 20 years of experience building phone systems. The agent connects to what you already run: SIP, PBX, carrier numbers, inbound SMS and one-time-code verification flows. You do not need to change carrier or PBX to start.

How we start

  1. Pick the right calls. We start with the most frequent, simplest requests and route everything else to people. That keeps the risk low and produces real data quickly.
  2. Pilot on live calls. Quality for a language and a call type is confirmed in a pilot on real calls before any launch.
  3. Measure what matters. The best test is simple: did the caller have to say the same thing twice?
  4. Widen step by step, with your team watching every step.

European languages, starting with Portuguese

We have a working full-duplex voice prototype in European Portuguese, trained on real speech: podcasts, public TV and radio, and parliament sessions. Training continues on our own GPUs. Our speech engine has 646 languages and dialects in development; none is called ready until it has passed a live pilot.

Voice agent, voicebot or IVR?

A traditional IVR (“press 1, press 2”) only follows menus. A classic voicebot recognises fixed phrases. An AI voice agent understands the request in the caller's own words, checks your systems and answers in conversation, and it knows when to hand over to a person.

Working with a team in Portugal

We are in Lisbon, on London time and one hour behind Central Europe, and we work in English and Portuguese under EU rules, GDPR included. How we run projects for companies abroad: nearshore in Portugal.

Who it is for

Internet and telecom providers, high-volume customer support, bookings and confirmations by phone, and any team that misses calls outside office hours today. If most of your callers ask for simple, repeated things, a voice agent pays for itself quickly. For AI beyond the phone, see AI integration for business.

Case studies

  • An AI call centre for a telecom and internet provider

    Our first live deployment answers subscriber calls for a telecom and internet provider.

    • We started with the boring calls: connection down, payments, router restarts. Everything else goes to a person.
    • We fixed turn-taking first: early versions waited too long and sometimes talked over callers.
    • Every hand-over to staff now carries a short summary of the call.
    • Everything runs inside a private perimeter; recordings follow a written retention rule.

    We wrote up the rollout in the Journal, mistakes included: Putting an AI voice agent on a real phone line.

  • auto.moto.pt: an AI assistant on the website, a separate product

    On our vehicle marketplace the AI assistant has been live since September 2026. It searches real listings, estimates values, calculates taxes with the site's simulators, and takes voice notes and a voice call right in the browser.

    It is not the phone agent on this page and it does not run on our servers: it uses external cloud language and speech models. It lives on the website, not on the phone system. We show it because it shows how we connect a language model to live data with rules and limits; if your project needs everything to stay inside your network, the answer is the on-premise agent described above.

    Visit auto.moto.pt(opens in a new tab)

Frequently asked questions

Can the voice agent run on our own servers?

Yes, that is the default. We install it on your servers or a private cloud, with no third-party cloud in the conversation.

Which languages does it speak?

Quality is confirmed per language and per call type in a pilot on live calls before launch. Our European Portuguese full-duplex prototype is working and still in training; our speech engine has 646 languages and dialects in development.

What happens when the agent cannot help?

It hands the call to a person on your team with a short summary, so the caller does not have to repeat what they already said.

Does it work with our PBX?

In most cases, yes. We connect over SIP to PBXs and carrier numbers. Tell us what you run and we will confirm in the first call.

How long does it take to go live?

It depends on the calls you pick and the integrations. We always start with a pilot on real calls for a small set of requests, and widen from there.

What is the difference between an AI voice agent and an IVR?

An IVR follows keypad menus. An AI voice agent understands the request in the caller's own words, checks your systems and answers in conversation.

Want to know if a voice agent fits your support line?
Tell us what calls you get and which systems you use. We usually reply within two working days.
Talk about a deployment

Talk to us

Tell us what you are working on. We usually reply within two working days.

Lisbon / WET (UTC+0/+1)
Estrada Nacional N10-1, loja 189, 2815-892 Sobreda, Portugal