On July 1, 2026, xAI, the AI startup led by Elon Musk, released a beta of Voice Agent Builder, a tool that lets anyone create an AI voice agent without writing code. You simply describe how a phone call should flow in plain language inside a web browser, and a working agent that actually answers calls can be up and running in as little as two minutes. Built on the company's Grok Voice model, it targets uses such as customer support, appointment handling, and personal assistance. This article walks through its features, pricing, and enterprise safety measures.

Build an AI Voice Agent in About 2 Minutes

The defining feature of Voice Agent Builder is that it is no-code, meaning you can assemble the whole thing without any programming. Even operators without a dedicated development team can prepare an AI that handles phone conversations just by working through the settings in a browser.

The setup is simple. First, you write out, in everyday language, the order in which you want the conversation to proceed. You then attach supporting materials such as internal manuals, the tools you want the AI to use, and guardrails (safety limits that keep the agent's responses in bounds). From a blank start, a working agent comes together in around two minutes. The underlying Grok Voice keeps response latency under one second, aiming for natural, human-like exchanges.

80+ Voices and Cloning Your Own Voice

You can pick the agent's voice from more than 80 built-in options. On top of that, by recording about two minutes of audio, you can create a cloned voice that reproduces your own brand's voice. The platform is said to support more than 25 languages, keeping multilingual handling in view as well.

According to xAI, Voice Agent Builder is designed with a speech-to-speech path tightly coupled to Grok Voice, processing audio directly as audio. Compared with the conventional approach of stitching together three separate stages — speech recognition, text generation, and speech synthesis — this makes it easier to reduce latency and unnaturalness.

Phone Numbers, SIP, and Service Integrations

Every new account is granted one free phone number, usable straight through from a first test call to live inbound traffic. If you want to keep using a number you already have, you can connect the lines of major telephony providers directly via SIP (a protocol for making phone calls over the internet).

Integrations with outside services are plentiful. It connects with Gmail, Outlook, Google Calendar, Linear, Notion, OneDrive, and more, letting the agent check schedules or log information during a conversation. It also supports a custom Model Context Protocol (MCP, a standard that connects apps and data directly to an AI), so you can plug in internal databases, customer management systems, and various APIs to tailor a voice agent to your own operations.

Pricing and Enterprise-Grade Safety Features

Pricing is usage-based, charged by the minutes of a call, at 0.05 USD (about 8 yen) per minute. The 80-plus voices are included in this rate, and there is no separate platform fee. When you use the phone line on your free provisioned number, an additional 0.01 USD (about 1.6 yen) per minute is added on top. *Converted at 1 USD = 161 JPY (as of July 4, 2026)

There are also features aimed at enterprise use. In addition to SIP connectivity for major telephony providers, there are guardrails that let you finely set the scope of responses, along with compliance support such as SOC 2, HIPAA eligibility, and GDPR. You can also restrict the AI from reading out sensitive information during a call, giving you a design that automates customer handling while staying mindful of how information is treated.

Summary

xAI's Voice Agent Builder stands out for pairing the convenience of no-code setup in around two minutes with practical, business-oriented features such as 80-plus voices, SIP connectivity, and MCP integration. The clear per-minute pricing, starting at 0.05 USD per minute, also lowers the barrier to adoption. For now it is a beta, so actual call quality and how well it works in Japanese are things to confirm as availability expands, but it looks set to become one option for operators considering automating their phone operations.