What This Actually Is
A phone agent you build by talking to it
xAI (Elon's AI company, the one behind Grok) shipped a thing called the Grok Voice Agent Builder. It's a no-code page where you describe an AI agent in plain English, hand it a phone number, and it starts answering and making real calls.
Not a chatbot on a website. An actual phone line. Someone dials the number, a human-sounding voice picks up, talks like a real person (sub-second replies, handles interruptions and "actually, one more thing"), and gets the job done: books the appointment, answers the pricing question, takes the message.
Setting this up used to mean weeks of duct-taping a tool like Twilio to a voice model to a calendar. Now it's a browser tab and about two minutes. That's the whole reason this guide exists.
Before You Start
What you need
A browser and an xAI account
That's the whole list. You build, test, and launch the agent from the browser. No install, no code, no separate phone service.
Whatever the agent should know (optional)
Hours, pricing, FAQs, policies. A PDF, a Markdown file, or plain text. You drop it in later so the agent answers from real info instead of guessing.
2 minutes of clean audio (only if you want to clone a voice)
Skip this and use one of their 25+ built-in voices. Voice cloning is US-only right now (except Illinois), so if that's you and you want your own voice, record a quiet 90-second clip.
Step 1
Open the builder
Go to x.ai/voice and hit Try It Free. It opens the builder right in your browser (over at console.x.ai). No download.
It asks what you want to build and gives you four starting points. Pick the closest one; you'll customize everything next.
Customer Support
Answers calls, looks up info, resolves issues.
Sales Associate
Qualifies callers and books the next step.
Personal Assistant
Handles your calls, scheduling, and follow-ups.
Custom
Blank slate. Start here once you know what you want.
Step 2
Write the playbook
This is the part that matters most. The "playbook" is just plain-English instructions telling the agent how a call should go. You write it like you'd brief a new front-desk hire, in simple Markdown headings: a greeting, what it can do, how to handle the main job, how to wrap up.
You could type "you answer calls for a dental office, book appointments, and answer pricing questions" and it works. But the more specific the playbook, the better the agent. So here's a full one you can copy, paste, and edit. This is the actual structure I'd use.
Copy-paste playbook: Dental office receptionist
## Greeting Answer warmly as the front desk for Sunrise Dental. Say the practice name, then ask how you can help. Keep it to one sentence. ## What you can do - Book, reschedule, and cancel appointments on the calendar. - Answer questions about services, hours, insurance, and pricing using the knowledge base. - Take a message for the office when the caller needs something you can't handle. ## Booking a new appointment 1. Ask what the appointment is for (cleaning, checkup, emergency, cosmetic). 2. Ask for the caller's full name and a callback number. 3. Offer the next two open slots that match. Confirm the one they pick out loud. 4. Ask if they are a new or returning patient. If new, tell them to arrive 15 minutes early to fill out forms. ## Pricing questions Quote only the ranges in the knowledge base. If they ask about something not in there, say the office will confirm the exact cost and offer to take a message. ## Wrap up Recap what you booked or did, confirm the caller has nothing else, then thank them by name and end the call. ## Rules - Never guess a price, a doctor's availability, or a medical answer. If you're unsure, offer to have the office call back. - Never read back or ask for a full credit card, SSN, or insurance ID over the phone. - If the caller is in pain or says "emergency," skip the small talk and get them the soonest available slot.
Step 3
Feed it a knowledge base
The playbook is how it talks. The knowledge base is what it knows. Drop in the docs the business already has, PDF, Markdown, or plain text, and the agent answers from them instead of making things up.
For that dental agent: the services menu, price ranges, insurance list, hours, and the "new patient" instructions. For a contractor: service area, warranty terms, the stuff you say on every call. Upload it once and it's baked in.
Tip: keep the pricing in the doc, and tell the playbook to only quote from it. That's how you stop the agent from inventing a number.
Step 4
Pick a voice (or clone your own)
Two options. Use one of the 25+ built-in voices (they're warm and expressive, not the robotic phone-tree kind), or clone your own from about two minutes of audio.
Cloning is free (30 clones per team) and takes one clip. A few things make a clone actually good:
Record 90–120 seconds
Longer is better. Under 30 seconds comes out thin. 120s is the max.
Quiet room, one voice
Kill the fan and notifications. Background noise gets cloned right along with you.
Talk like the calls will go
Cloning a support agent? Record a natural support-style exchange, not a stiff script read.
US-only for now
Voice cloning is limited to the US (except Illinois). Outside that, use a built-in voice.
Step 5
Grab your free phone number
This is the headline feature. xAI hands you a real phone number, free, and your agent can make and receive calls on it immediately. No Twilio account, no telephony setup, no per-number fees to wrestle with.
Already have a business number you want to keep? You can bring your own. It works with any major phone provider through direct SIP, so the agent picks up on the line customers already call.
Step 6
Test it live, then hit create
Before a single real call, you can talk to the agent right in the browser. Change a line in the playbook and hear the difference that second. This is where you catch the awkward stuff, the spot where it over-explains, the answer that's a touch off, and tighten it.
When it sounds right, hit create. That's it. Your AI voice agent can now make and receive real calls. Every call gets recorded and you can play any of them back start to finish, which is how you keep improving it after launch.
Make It Actually Useful
Wire in the tools it needs
An agent that only talks is a nicer voicemail. The value is in the doing, and it plugs straight into the apps a business already runs on, so it can actually take action mid-call.
Calendars & email
Google Calendar, Outlook, Gmail. It books the slot, schedules the follow-up, and sends the confirmation on the call.
Work tools & CRMs
Notion, Linear, OneDrive, and CRMs. Log the lead, update the record, trigger a workflow.
Anything else, via custom MCP
Got an internal system or database? Connect it with a custom MCP and the agent can pull records and take action there too.
Hand off to a human
When a call needs a person, it transfers. You set the rules for when.
Keep It Safe
Set guardrails you decide
You define what the agent will and won't do, and it stays inside those lines on every call. The obvious one: no PII. Tell it never to read back a card number or ask for an SSN, and it'll verify identity another way instead.
This is also what makes it usable for real businesses, not just a toy. It's built to be enterprise-grade on the compliance side (SOC 2, HIPAA eligible, GDPR), so a dental office or a law firm can actually run it. Put the guardrails right in the playbook, like the "Rules" section in the copy-paste example above.
The Honest Part
What it actually costs
"Free phone number" is real, and so is building and testing the agent for nothing. But calls aren't free forever, so here's the straight version.
Building + the phone number → free
Build the agent, test it live in the browser, and get the number at no cost. Voice cloning is free too (30 clones).
Live calls → $0.05 / minute
That's the speech-to-speech rate once the agent is on a real call. A 4-minute booking call costs about 20 cents.
The rest of the voice suite (if you go deeper)
Speech-to-text is $0.10/hr batch or $0.20/hr streaming. Text-to-speech is $15 per 1M characters. Live translation is contact-sales.
Put it in perspective: five cents a minute for a phone line that never sleeps, never misses a call, and never needs a lunch break. The thing it replaces used to be weeks of engineering.
The Money Play
Build these for businesses and charge a retainer
Here's the part most people scroll past. Every local business with a phone that goes to voicemail is losing money. The dentist who misses a booking because the front desk is with a patient. The contractor who can't answer while he's on a roof. The med spa that's closed on Sundays.
You can build one of these agents in an afternoon, wire it to their calendar, and charge a monthly retainer to run and improve it. The cost to you is pennies a minute. The value to them is the calls they were dropping. That's the whole business.
So here's a second copy-paste playbook, a lead qualifier for a home-services company, so you've got two different shapes to work from: one that books, one that qualifies.
Copy-paste playbook: Lead qualifier (home services)
## Greeting Answer as the booking line for Northwest Roofing. Say the company name and ask how you can help in one sentence. ## Goal Qualify the caller and book a free inspection for anyone who's a real fit. You are not trying to close a sale on the call. ## Qualify (ask these in order, one at a time) 1. Are you the homeowner? (If renter, politely explain you can only schedule with the homeowner or property owner.) 2. What's going on with the roof? (leak, storm damage, age, selling the house) 3. What's the property address and roughly how old is the roof? 4. Is this covered by insurance, or paying out of pocket? ## Book the inspection - Offer the next two open inspection windows. Confirm the one they pick. - Get their full name, best callback number, and the address. - Tell them an inspector will call 30 minutes before arrival. ## If they're not a fit Renters, commercial buildings, or anyone outside the service area: thank them, explain briefly why, and offer to take a message for the office. ## Wrap up Recap the appointment (date, time, address), confirm nothing else, thank them by name, end the call. ## Rules - Never quote a repair price. Say the inspector gives the exact number on site, free. - Never promise the insurance company will cover anything. - One question at a time. Wait for the answer before moving on.
The Whole Thing in 6 Steps
Quick recap
Open the builder. x.ai/voice → Try It Free → pick a preset.
Write the playbook. Plain-English instructions (steal the copy-paste ones above).
Feed it the docs. Drop in pricing, hours, FAQs as PDF or text.
Pick or clone a voice. Built-in, or your own from 2 minutes of audio.
Grab the free number. Or bring your own via SIP.
Test live, then create. Talk to it in the browser, tighten it, ship it.
Work with Me
Want me on your problem for an hour?
Book a 1:1 call and we build the fix live — the workflow you want automated, the tool you can’t crack, whatever’s stuck. Direct, hands-on, no pitch waiting at the end.