Phone & Voice

pro

Get phone numbers onto your AI voice agent, route callers to people or voicemail, clone your own voice, and track call minutes and cost.

Phone & Voice puts a real phone number in front of your AI agent. Callers dial your business, the agent answers in their language, quotes your real services and prices, and books straight into your calendar — and when someone needs a person, the call goes to one. Every call is filed into your Contact Center inbox alongside WhatsApp and web chat, so your team works one queue, not three.

This guide covers the Voice section of the sidebar: getting numbers, deciding what answers them, reaching a human, voicemail, recordings and retention, the agent's voice (including cloning your own), and usage.

Voice is an optional add-on to the Contact Center and ships off. If you don't see Voice in the sidebar, it isn't enabled for your organization — contact support. The section is also manager-only, so non-manager staff won't see it either way.

What you get

A Numbers page listing every phone number you hold — where it rings, its status, what answers it, and its monthly cost.
Two ways to get a number: ask us to arrange one for you, or (where enabled) search the carrier's available numbers and buy one yourself, with every purchase going through platform approval.
A Calls page with the full call history: direction, duration, outcome, recordings, and a jump into the inbox conversation.
Voice settings for how each line answers, the agent's greeting and language, what the agent works toward on calls, and voice cloning.
Transfer-to-human, a keypad escape digit, a human fallback if the AI hits trouble, and voicemail that lands in your inbox.
A Usage page with the month's call minutes and voice cost.

Who can do this

Every page in the Voice section is manager-only — this is the domain-manager role (is_manager), not a privilege you can grant per staffer.

Capability	Who can do it
Open Numbers, Calls, Voice settings, Usage	Managers only
Request a phone number	Managers only
Search and request a number purchase (self-serve)	Managers only, and only when self-serve purchasing is enabled for your organization
Approve or reject a number purchase	Apolaris team (platform operators)
Edit voice lines, greeting, conversation settings; clone a voice	Managers only
Build call flows and place Transfer / Voicemail steps	Managers only — see Conversation flows
Set up transfer destinations, recording on/off, retention and disclosure policy	Apolaris team (platform operators)
Listen to voicemails and call recordings in the inbox	Anyone with inbox access — managers or staff with the Use contact center privilege (`can_use_ccaas`)

Find your way around

The Voice sidebar section has four pages, repeated as tabs at the top of each one:

Numbers — the phone numbers connected to your voice agent, plus the forms to get more.
Calls — persisted call history with outcomes and recordings.
Voice settings — everything about how the agent answers and sounds.
Usage — this month's metered voice volume and cost.

Get a phone number

Open Voice → Numbers. The table shows each number, its Site, Status, What answers it (a voice line or a call flow), and its Monthly cost. Below it you'll see one of two ways to add a number, depending on what's enabled for your organization.

An AI line never handles emergency calls (112 and similar). A virtual number is not a substitute for a basic or mobile emergency line.

Ask us to arrange it

The default path is a brokered request — real phone numbers involve a licensed carrier and identity checks, so Apolaris handles the paperwork:

On Numbers, fill in Request a phone number: the Country / area you need (for example "Spain, Madrid") and any Notes.
Press Request number. You'll see "Request received — we will set up your number and email you."
We arrange the number with the carrier and email you when it's live. It then appears in your numbers table.

Buy a number yourself, with approval

Where self-serve purchasing is enabled for your organization, the Buy a number card replaces the request form. You pick the exact number; a platform operator approves the purchase before any money moves.

Enter the Country (two letters) (pre-filled from your organization's country) and optionally an Area code.
Choose What the number must do — Voice or Voice + SMS. Most numbers outside the US and Canada are voice-only, so leaving this on Voice finds the most numbers.
Pick the Site the number should belong to (or All sites) and press Search numbers.
The carrier's available numbers are listed with their Monthly cost and One-time cost. Press Request purchase on the one you want.
Your request appears in the Purchase requests table as Pending approval. A platform operator reviews it; once approved and purchased, the number activates, is wired for inbound calls, and you get an in-app notification. If it's declined, the reason is shown on the request.

The costs shown at search time are the carrier's quoted prices, and they're what you'll be billed: the one-time cost when the purchase completes, then the monthly cost on your regular invoice (see Billing & usage).

Your plan sets how many numbers your organization can hold. Numbers you already own plus open purchase requests all count toward that allowance, and a request over the limit is refused — contact support if you need it raised.

Choose what answers each number

The What answers column on Numbers shows one of two things:

Voice line — the number rings straight into the AI agent, using the line's settings (voice, greeting, language, conversation goal — see below).
Call flow — the number runs a flow you designed: menus, conditions, transfers, voicemail, and AI steps in the order you choose. You build flows on the Flows page and bind a published flow to a number there — see Conversation flows.

A number with no flow simply uses its voice line, so the agent answers from day one and you can add a flow later without downtime.

Calls don't have to come from a phone at all. The Use it anywhere card on Voice settings gives you a Call link to share anywhere (social bio, email signature, a QR code) and an Embed code that puts a floating AI call button on any HTTPS website — even one not built with Apolaris. The button renders nothing while voice is switched off, so it's safe to leave in place. For the voice widget on your Apolaris-built site, see Your public website.

When the caller needs a person

Not every call should stay with the AI, and the platform gives callers real ways out:

Transfer steps in a flow. Add a Transfer step to a call flow and pick a Transfer target — one of your organization's approved destinations (a front desk, a mobile, a manager). The caller stays with the agent while the destination rings, and the agent only leaves the call once a person has actually answered. If the line is busy or unanswered, the agent says so honestly and keeps helping — the caller is never dumped into dead air.
The keypad escape digit. At any point in a call, the caller can press the escape digit (0 unless configured otherwise) to be put through to your configured person, no matter where they are in the conversation.
Human fallback. If the AI itself runs into a technical problem mid-call, the call is handed to your configured fallback destination instead of failing silently.

The list of transfer destinations, the escape digit, and the fallback destination are set up for you — contact support to add or change them. Flow authors then pick from those approved destinations; a flow can never send a caller to an arbitrary number.

A voice line can also put calls through to a person by schedule, or after a number of AI replies. On Voice settings, each line's card has Who attends this number — the AI at all hours, only inside the ticked hours, or only outside them (your team takes the rest) — and Hand off to your team after N AI replies, counted within one call. Outside the AI's hours, or once the cap is reached, the caller is put through to your configured transfer destination exactly like the escape digit. With no destination set up, the AI simply keeps answering, so no caller is ever left in silence. WhatsApp numbers have the same settings — see Attention hours.

Voicemail

A Voicemail step in a call flow decides what happens when a call shouldn't continue — after hours, or from a menu option like "leave us a message". Its Voicemail action is either Hang up (say the step's message and end the call) or Leave a message, which records the caller:

The caller hears your prompt, then a short tone.
They speak for up to the Max message length (seconds) you set (10–300, default 120). The recording ends when they stop talking for a few seconds — and hanging up mid-sentence still saves the message.
Your Thank-you text (after the message) is spoken once the message is safely stored. Leave it empty to end the call silently.

The message lands in your Contact Center inbox as a playable audio attachment on the caller's conversation, transcribed when possible, and raises the normal new-message notification so staff see it like any inbound text. It also shows on Voice → Calls with the outcome Voicemail. Voicemail audio expires with your voice retention policy (below).

Recordings, disclosures, and retention

What callers hear is enforced by the platform. At the start of a call, callers hear a short standard notice — that they're speaking with an AI assistant and, when recording is on, that the call is recorded. Which notices are required is policy-driven per country and cannot be switched off where the law requires them. Your custom greeting plays after the notice, never instead of it.

Recording is off by default. No voice line records calls until recording is explicitly enabled for it — contact support to turn it on. When it's on, recordings appear on the Calls page and on the conversation.

Retention is real deletion. Recordings and transcripts are kept for your organization's retention period (90 days unless agreed otherwise) and then permanently deleted — audio files and spoken transcripts are physically removed, not archived. The call's basic record (date, duration, outcome) is kept so your history and stats stay intact. The retention period is part of your organization's policy, managed with Apolaris.

Give the agent its voice — or yours

Open Voice → Voice settings. Under Your voice lines, each line has a Voice picker with the platform's standard voices — plus any voices you've cloned. The same card holds the line's Name, Default booking location, Serves these locations, and the AI answers and Active switches.

Clone your own voice

Where voice cloning is enabled for your organization, the Clone your own voice card lets you create a voice from a sample:

Enter a Voice name and pick an Audio sample — about 10 seconds of clear speech, recorded somewhere quiet, in a common audio format (WAV, MP3, M4A, OGG, WebM, and more; phone recordings work).
Tick the consent checkbox. It's required: use your own voice whenever possible, and otherwise get documented permission from the speaker — you are responsible for applicable voice, biometric, privacy, publicity, and AI-disclosure rules.
Press Create voice. The finished voice appears as a saved voice and in every line's Voice picker.

Uploading a new sample under an existing voice name replaces that voice safely — lines using it switch over only once the replacement is ready. Removing a cloned voice returns any line using it to the default voice.

If cloning isn't self-serve on your plan, managers get a Request voice cloning page instead (under Customers → Contact Center in the sidebar): submit the consented sample with notes, and the platform team performs the clone — the finished voice then appears in your Voice settings, with the request's status tracked on the page.

Hear it before callers do

The Test your voice card gives every cloned voice a tester:

Write a short sentence in Text to speak and pick the Voice.
Leave Language on Auto to detect the language from your text — after generating, the selector shows which language was used — or pick one of the supported languages explicitly.
Press Generate preview and listen right on the page.

Tune what the agent works toward

Two cards on Voice settings shape every call.

Greeting & language applies to your AI agent across chat and voice: the Agent greeting, Language, Formality, and Reply length.

Conversation is what the agent does on calls:

Goal of the conversation — the one line the agent works toward on every call (for example "book appointments and answer questions about our services").
Featured services — up to 10 services the agent knows by heart and leads with. Until you pick any, it leads with your most-booked services, and the card shows you which those currently are.
Required before booking — whether the agent must collect the Caller's name and the Caller's phone number before it can book. It reads the phone number back digit by digit and asks the caller to confirm it, so a misheard number never lands on a booking. Untick both only if you want the agent to book without collecting caller details.
Booking confirmation links — after the agent books or changes an appointment on a call, it can text the caller a link to that appointment, where they can see the details and cancel it themselves. Choose WhatsApp if available, else SMS, or pin one channel. The agent always asks the caller first and only sends after a clear yes; that consent is recorded with the caller's own words, the channel and the timestamp, so you can demonstrate it later. It needs a confirmed phone number, so it pairs with Required before booking. Leave it on Do not send links to switch the offer off entirely.
Behavior instructions (tone, promos, boundaries…) — free-text guidance for how the agent behaves.
Personality, Expressiveness and Laughter — how the agent sounds. Pick a personality (Professional, Friendly, Cheerful, Funny, Empathetic, Calm, Efficient, Enthusiastic, Elegant, plus two marked ⚠ whose register can read as unprofessional on a business line), then how much of it reaches each sentence: None is the plain, neutral agent, Subtle adds a little warmth, Expressive reacts before answering and varies its tone, Theatrical leans in hard. The choice changes both what the agent says (reactions, pauses, a written laugh) and the voice itself (its tone and pace); prices, availability and everything it must collect before booking stay exactly the same. Laughter decides whether it may laugh softly at something genuinely funny; it never laughs at a complaint or a problem. Try a personality on your own line before you settle on it: save, call the number and listen.

Changes apply to new calls, not ones already in progress.

Review your calls

Voice → Calls is the persisted call history, 50 rows per page: when the call happened and on which line, its Direction and Transport, the Caller, the Duration, and its Outcome — Booked, Transferred, Voicemail, Abandoned, or Answered. When a recording exists, it plays inline in the Recording column, and Open conversation jumps to the call's thread in the inbox.

The page shows calls that actually exist in the ledger. Browser-based web calls (from the website call button) don't create call-history rows yet, so they appear in the inbox as conversations but not on this page.

Track usage and cost

Voice → Usage shows the current month at a glance — Call minutes, TTS usage (speech the agent generated), STT minutes (speech it transcribed), and Voice cost — followed by the Metered events table with every individual charge. Speech served from cache is listed as free, and the same voice totals appear on your Billing & usage page, where phone-number rental also shows on your invoice.

Tips

Get the number first, flow later: a fresh number answers with its voice line immediately, so you can go live and add menus or voicemail when you're ready.
Feature your highest-value services in the Conversation card — the agent leads with what you pick, not just what's most booked.
Test a cloned voice with a sentence in each language your customers actually speak before assigning it to a line.
Keep Required before booking on unless you have a reason not to: a confirmed name and number is what makes phone bookings reliable.