Build your first AI phone agent

Every missed call is a customer deciding between leaving a voicemail and calling your competitor. An AI phone agent removes that decision: every call answered on the first ring, at 2 pm or 2 am, in a voice that sounds like your business.

This guide takes you from a blank account to a live agent answering real calls. There's no code anywhere in it — you describe what you want, and Receptive's builder does the building.

What you're building

Receptive is AI phone infrastructure: the platform provides the AI receptionist, and you shape it into yours. An agent is three things working together:

A phone number it answers on (or places calls from).
A conversation flow — how it greets callers, what it can help with, what information it collects, and when it hands off to a human.
A voice — the actual sound of your business on the phone.

Optionally, a fourth: functions that connect the agent to your real systems, so it can check availability or look up an order instead of taking a message about it.

Step 1 — Create your workspace

Sign up and onboarding walks you through it: name your workspace (usually your company name), add your details, and pick a plan — every plan starts with a free trial, so nothing is billed while you build.

A workspace holds everything for one business: its locations, agents, numbers, and team. If you operate multiple brands, you can add more workspaces later and switch between them from the sidebar.

Then open Workspace in the dashboard and fill in your business context — what you do, your services, hours, and anything callers commonly ask. This isn't paperwork; it's what grounds your agent. The context you write here is what the agent draws on when a caller asks "how late are you open on Saturdays?"

Step 2 — Get a phone number

Head to Phone numbers and add one. You have two options:

Get a new number and publish it directly — the simplest start.
Keep your existing number and forward it. Most businesses go live this way: your published number stays the same, and your carrier forwards calls to the agent's number — either always, or only when the line is busy or unanswered. That second mode is a great first deployment: the agent only catches the calls you were already missing.

Assign the number to your agent and every call to it now reaches the agent.

Step 3 — Design the conversation

Open your agent from the Agents page. This is where you'll spend most of your time: a visual flow canvas on one side, a builder copilot on the other.

The fastest path is to just tell the copilot what you want:

"Greet callers, answer questions about our services and hours, and take booking requests — collect name, phone, and preferred time. If someone's upset or asks for a person, offer to take a message for a same-day callback."

It builds the flow; you refine it on the canvas. A good first agent has four parts:

A short greeting. One sentence: who they've reached and an open question. "Thanks for calling — how can I help you today?" Callers start talking when you stop.
The things it can do. Answer common questions from your business context, take booking requests, capture messages. Start with the two or three intents that cover most of your call volume — you can widen it later.
What it collects. For each intent, decide what the agent must capture (name, callback number, preferred time) and have it confirm details back before moving on. Numbers and names are exactly where phone conversations go wrong; a two-second "That's 555-0142, correct?" saves a bad callback.
An exit to a human. Decide up front what happens when the agent shouldn't handle it: transfer to a cell, take a priority message, or send the caller a text. The agent should offer this gracefully, not cling to the conversation.

Flow text supports variables — write "Thanks for calling {business_name}" once and every location's agent says its own name.

Step 4 — Give it a voice

Open Sandbox and audition voices with your own lines — paste your actual greeting, not lorem ipsum, because a voice that reads a sentence well and a voice that sounds like your front desk are different things. Pick for warmth and pace over novelty; callers relax when the voice sounds unhurried.

Step 5 — Connect your systems (optional, powerful)

Out of the box your agent works from the context and flow you gave it. The step-change happens when it can touch live data: check this afternoon's real availability, look up an order, create a booking that lands in your calendar.

You don't build this — you ask for it. On the agent's Integration view, tell the builder what software your business runs on ("we use NexHealth" / "our bookings live in Vagaro") and it writes the connection, asks for your API keys when it needs them (they're stored securely and can never be read back out), and tests everything against your real data before it goes live.

If your platform has a guide — like NexHealth for dental practices — it has a Deploy to Receptive button that kicks the whole thing off with one click.

Step 6 — Test, then go live

Call your agent. Seriously — before anything else, have the conversation a real caller would have. Then have the conversations they shouldn't: mumble, interrupt it, ask something off-topic, ask for a manager. You're checking that it recovers gracefully, not that it's perfect.

Every call appears in Conversations with the full transcript, the tools the agent used, and the outcome. This page is your feedback loop: skim the first days of real calls, notice where callers hesitated or repeated themselves, and adjust the flow. Ten minutes of transcript reading is worth more than an hour of guessing.

When it feels right, point your published number at it (or flip forwarding to "always") — you're live.

The habits of great phone agents

Teams that run agents callers actually like converge on the same rules:

Short turns. Phone attention spans are measured in seconds. One idea per sentence; never three options when two will do.
Confirm anything you'd write down. Names, numbers, times — repeat them back. It feels attentive, and it's your accuracy net.
Never dead-end. Every path ends in something concrete: a booking made, a message taken, a callback promised, a transfer completed.
Own being an assistant. Callers don't mind talking to an AI that's fast and competent; they mind one that pretends. A simple "I'll get that to the team right away" builds more trust than bluffing.
Review weekly. Call patterns shift. The Conversations page tells you what your callers actually need — let it steer what you build next.

Your first agent doesn't need to do everything. It needs to answer every call, handle the common cases cleanly, and hand off the rest. Ship that, read your transcripts, and grow it from there.