AI telephony is a phone line answered by an AI voice agent instead of a person: it picks up the call, handles the conversation according to its instructions and, once the call ends, leaves a recording, a transcript and an analysis behind. This article follows a short demo of Tridan AI telephony and explains what happens to a call from the first ring to the final report.
The video above is in Serbian and runs under four minutes. Below is the same flow in English, step by step, with the settings behind each step. A Serbian version of this article is also available.
Which channels does the AI agent answer?
The Tridan AI agent talks to people through three channels: telephony (mobile and landline), voice conversations in the browser and the text chat widget on your website. This video covers the first two, telephony and voice.
For telephony, it does not matter where the call comes from. A customer can call:
- a mobile number,
- a landline number,
- a toll-free line,
- or start a call directly from your website.
In every case the agent answers automatically and handles the conversation according to its instructions and goal: answering a question, booking an appointment, collecting details or passing the case to a person. The telephony layer runs on LiveKit and Twilio, and companies that already have their own phone system can connect it by following the PBX and SBC setup guide.
If you want to hear what it sounds like, call the Tridan AI demo line at +381 11 635 5917. An AI agent will answer.
What happens to a call after the conversation?
Every call is recorded and processed. That is the key difference from a traditional phone line, where the conversation disappears the moment the caller hangs up and everything that was said depends on the notes an operator managed to take.
A recorded conversation is processed in this order:
- Recording: the whole conversation is stored as audio.
- Transcription: the recording is turned into text automatically.
- Diarization, on request: the transcript is split by speaker, so it is clear what the agent said and what the caller said.
- AI analysis and reports: a summary of the conversation, a score, the sentiment and a check for misunderstandings during the call.
Both inbound and outbound calls are analyzed, so the same reports cover a support line and a sales campaign. You can read more about this part of the platform on the Voice Analytics page.
Where is the data processed: in the cloud or on your servers?
Processing can run in two ways. In the cloud, the platform uses model providers and speech services. Tridan AI works with four model families (OpenAI, Anthropic, Meta and DeepSeek), and uses providers such as Deepgram, ElevenLabs and OpenAI for speech recognition and voice.
The other option is on premises: all processing runs on the client's infrastructure, with local models via Ollama. The demo shows exactly this kind of data flow. The recording, the transcript and the analysis never leave the client's network. Details about running the platform on your own infrastructure are on the self-hosted page.
What does a call review look like in the portal?
In the portal, every recorded conversation has its own page where you can:
- listen to the call right in the browser,
- see details about the recording itself,
- review the analytics produced for that call.
The analytics shown in the demo answer the questions managers care about most: did the call end successfully, was a meeting booked, was a PDF sent to the customer, what was the sentiment of the conversation and was the call transferred to a human operator.
These indicators are not fixed. Depending on what the client needs, the call analytics are adjusted and new features are added for that specific company. A sales team cares about booked meetings, while a support team wants to know whether the problem was solved on the first call.
Does the same apply to voice conversations on the website?
It does. Everything that applies to telephony also applies to voice chat, meaning a conversation with the agent through the browser. A visitor clicks the voice button in the chat widget and talks to the agent without picking up a phone. That conversation is recorded, transcribed and analyzed just like a phone call. The AI chat widget page explains how the widget combines text, voice and phone.
What does call monitoring show?
The last part of the demo is monitoring. For every conversation you can track the technical details:
- which model handled speech to text and which handled text to speech,
- which model ran in real time and which was used for knowledge,
- which tools were called during the conversation,
- how fast the responses were,
- what the network conditions were.
This data is not there just to look at. It is what the agent is tuned with afterwards: switching a model, shortening the instructions or speeding up a tool that slows the conversation down, so the agent meets its goal as well as possible.
Who benefits most from AI telephony?
AI telephony helps most where calls arrive in waves or outside business hours: clinics, service companies, shops and support teams. The agent answers every call, at midnight and at peak time alike, and in the morning the team gets the transcripts, summaries and a list of calls that need a person. Staff then spend their time on conversations that really need a human instead of repeating the same answers.
Where to go next
The AI Call Center page explains how AI telephony is used for inbound calls, outbound campaigns and handing conversations over to people. If you would like to see how an agent like this would work on your own number, get in touch through the Contact page and we will set up a short demo.
- AI call center
- Inbound calls
- Call analytics
Last updated · Reviewed by Nemanja Milivojevic, Founder