VoxaLinkSupport
01Chapter 1 of 10

Overview & getting started

What a voice agent is, how to create one, drafts vs publishing, and choosing an AI engine.

Read time~ 5 min
Sections6
UpdatedJuly 27, 2026

A voice agent is an AI-driven assistant that answers or makes calls on your behalf — it can hold a real conversation, look things up in your knowledge, check availability and book appointments, and hand off to a person when it should. Find them under Call flow → Voice agents.

Voice agent vs. auto attendant An auto attendant plays a fixed menu and routes callers by the key they press. A voice agent actually converses — it understands what the caller says, answers from your knowledge, and can take real actions like booking a slot on your calendar — before ever routing the call anywhere. Use an auto attendant for a simple "press 1 for sales" menu; use a voice agent when you want callers to get real answers and get things done without waiting for a person.

The building blocks of an agent

Every voice agent is made up of the same pieces, each covered in its own chapter:

  • Assistant — who the agent is: its persona, voice, greeting, and the instructions that shape how it talks and what it will and won't do.
  • Playbook & transfers — what the agent does with each kind of request, and how callers reach a person — including matching anyone in your live directory by name or extension.
  • Knowledge — the approved information the agent answers from, so it never has to guess.
  • Calendars — connect a calendar so the agent can check availability and book, look up, cancel, or reschedule appointments.
  • Campaigns — have the agent place outbound calls to a list of contacts instead of only answering inbound ones.
  • Testing & publishing — how an agent goes live, and what changes take effect when.
  • Limits & troubleshooting — every size limit, cap, and known restriction in one place.

Creating a voice agent

Call flow → Voice agents → New agent walks you through three steps.

1. What should it do? Pick Answer incoming calls or Make outbound calls. This is the agent's mode and it can't be changed later, so if you're unsure, start with an inbound agent — that's what most people want first.

2. Pick a starting point Choose a template to seed the agent. This isn't optional — every agent starts from one — but you can change everything in it afterwards, so pick whichever is closest and adjust. You can enter your business name here to have it filled in for you.

3. Tell it about your business This is where the agent actually takes shape:

  • Business name and Agent name — the agent introduces itself with the agent name, so use something a caller would be comfortable hearing.
  • What does the business do? — a few plain sentences. This becomes the agent's business context and does a lot of the work, so it's worth writing properly.
  • AI engine and Voice — see Choosing an engine. Each engine has its own voices, so pick the engine first.
  • Language — English only at this stage; see The Assistant.
  • Agent label in your portal — the name in your agent list. Only your team sees this.
  • Account number — the internal number that reaches this agent (2 to 32 digits, must be unique, can't be changed later).

Everything you set here is editable afterwards in the Studio. There's no limit on how many voice agents you can create.

Draft, published, and what a caller actually reaches

Every change you make lives in a draft version until you publish it. A live call always runs whatever was published most recently — so you can freely edit an agent mid-day without any risk of changing a call that's already in progress; the change only takes effect the next time the agent is published. You can also roll back to an earlier published version if a change didn't work out.

An agent can be disabled — this stops it from being scheduled into new outbound campaigns, though it doesn't automatically detach it from any inbound routing. There's no separate "testing" or "sandbox" mode: to try an agent before relying on it, publish it and place a real test call through the outbound calling screen — see Testing & publishing.

Choosing an engine

Each agent runs on one AI engine, picked when you create it and changeable later in the Studio: Google Gemini (recommended), OpenAI Realtime, or xAI Grok. Which engines are available to you depends on what's enabled for your tenant — if only one is enabled it's used automatically; if you have access to more than one, you choose per agent. You pick the engine, not a specific model — VoxaLink keeps the underlying model current on your behalf.

How a call behaves

A voice agent always performs a warm transfer when it hands a call to a person — it puts the caller on hold, briefs whoever it's transferring to, and only connects the caller once that's accepted (there's no blind/cold transfer option). Callers can be routed to a fixed destination you choose, or matched by name or extension against your live directory — see Playbook & transfers. If the agent can't help — an error, or the caller needs a human — it falls back to whatever destination you've configured rather than leaving the caller stuck. Voice agents are fully voice-driven: keypad (DTMF) input isn't supported, so don't design a flow around callers pressing digits.

Everyday tasks

  • Build an AI receptionist that actually answers questions → create an agent, then fill in its Assistant and Knowledge.
  • Let callers book appointments → connect a Calendar and add a booking step to the agent.
  • Call a list of leads or customers automatically → publish an outbound agent, then run it as a Campaign.
  • Try an agent before pointing real traffic at it → publish it and place a manual test call — see Testing & publishing.
  • Check a specific limit or restriction → see Limits & troubleshooting.