What is voice AI, and how does it work?
Voice AI, in plain language
Voice AI is software that conducts a spoken conversation with a person over the phone in real time: it listens, understands what the caller wants, decides what to do about it, and speaks back, in a loop measured in milliseconds. It is built from four parts working together — speech recognition that turns the caller's audio into text, a language model that understands the request and picks the next action, tool calls that carry the action out (look up the account, book the slot, update the record), and speech synthesis that speaks the response. When the loop is fast enough and the model is grounded in your business, callers experience it as a competent phone agent, not a phone menu.
On a live call the pipeline runs continuously from the first greeting: the phone network delivers the call, the agent answers, streaming speech recognition transcribes the caller as they speak, the language model reads the transcript against your knowledge base and business rules, any needed action executes through an API, and speech synthesis turns the reply back into audio while the caller is still listening. Each stage streams into the next, which keeps the rhythm of the conversation in the range of a human exchange — and the whole call, transcript and actions included, lands on one record.
Devotel Orbit ships this as voice AI agents inside an integrated communications platform: the same agent runtime that answers the call also answers SMS, WhatsApp, RCS, and email, reads and writes the same per-contact timeline a human agent works from, and escalates into the same shared inbox and queues as your team. The calls themselves terminate on Devotel's own wholesale softswitch, where Devotel is the carrier of record — so the voice path, the messaging path, and the agent layer are one accountable infrastructure, not three vendors stitched together.
How it works: one call, six stages
Every stage streams into the next, which is what keeps a voice AI conversation in the rhythm of a human one:
1.The call arrives and the agent answers
The phone network routes the call to Orbit and the voice AI agent answers with a greeting grounded in your business, on a call carried on Devotel's own carrier-of-record softswitch.
2.Speech becomes text
A streaming speech-recognition model transcribes the caller's audio as they speak, so understanding starts before the caller stops talking.
3.The model understands and decides
A language model reads the transcript against your knowledge base, account data, and business rules to determine what the caller needs and what the right next action is.
4.The agent acts
Through tool calls the agent executes the action: look up the order, book the appointment, update the contact record, or check consent and quiet hours before any outbound step.
5.The response is spoken back
Speech synthesis turns the reply into natural audio, streamed so the conversation keeps the rhythm of a human call.
6.The outcome lands on one record, with a human handoff when needed
Transcript, actions, and outcome are written to the contact's timeline; when the caller needs a person, the agent transfers into the same queue and shared inbox your human agents work, with the full history attached.
Voice AI agents on Devotel Orbit
A voice AI agent is only as good as the infrastructure it sits on. On Orbit the agents are not a layer bolted onto someone else's telephony — they run inside the platform:
Voice infrastructure, owned end to end
Every call a voice AI agent answers or places terminates on Devotel's own wholesale softswitch, where Devotel is the carrier of record. There is no resold telephony hop between your caller and the agent, which is where the latency and quality of a real-time conversation are won.
Messaging, on the same agent runtime
The agent runtime is channel-agnostic: the same configuration that answers calls can answer SMS, WhatsApp, RCS, and email, so a conversation that starts on the phone can continue by message without the customer repeating themselves.
The contact center, shared with humans
Voice AI agents sit in the same routing engine, queues, and shared inbox as your human agents. Routing decides whether a call goes to an agent or a person first, and escalation carries the live transcript and the contact record with it.
One record, one bill
Every call, message, and agent action writes to a single per-contact timeline, and usage meters to one pay-as-you-go bill at published per-country rates — no per-seat agent licence on top.
Voice AI — frequently asked
- What is voice AI?
- Software that holds a spoken conversation over the phone in real time. It combines speech recognition, a language model, and speech synthesis in a loop fast enough to feel like a person, and it can take real actions — lookups, bookings, record updates — while the call is live. On Devotel Orbit, voice AI agents answer calls on Devotel's own carrier-of-record network and share one record and one inbox with human agents.
- How does voice AI work?
- In a loop: the caller's audio is transcribed by a streaming speech model, a language model understands the request against your knowledge base and business rules, the agent executes any needed action through tool calls, and a speech model speaks the reply back. Each stage streams into the next, which keeps response latency in the range of a human conversation.
- What are voice AI agents?
- Voice AI deployed as participants in your contact center: they answer calls in your queue, work from your knowledge base, write outcomes to the contact's timeline, and hand off to human agents in the same shared inbox. The term distinguishes a deployed, accountable agent from a demo or a standalone bot.
- How is voice AI different from IVR or a phone menu?
- A phone menu routes on the digits you press; voice AI understands what you say and why, and can act on it. IVR can tell a caller which department handles refunds; a voice AI agent can process the refund, confirm it, and log it, escalating to a human only when judgement is required.
- Can voice AI agents hand off to human agents?
- Yes. On Orbit the handoff lands in the same queues and shared inbox your team already works, with the live transcript and the contact record attached, so the human continues the conversation instead of restarting it.
- Whose network carries the calls?
- Devotel's. Agent calls terminate on Devotel's own wholesale softswitch, where Devotel is the carrier of record, rather than on a resold upstream provider. That ownership keeps the real-time voice path under one operator's control.
Explore more
Put voice AI agents on your calls
Voice AI agents, AI chat agents, the shared inbox, and the carrier-of-record network they run on are one account on Devotel Orbit, on one pay-as-you-go bill. Start free, or talk to our team about your rollout.