xAI launches Voice Agent Builder for AI voice agents

xAI has launched Voice Agent Builder in beta, a no-code platform for creating voice agents that can handle calls, look up information, and take actions in a company’s systems.

The tool runs on Grok Voice and brings together several pieces that are usually purchased and connected separately: telephony, document search, external tools, security controls, and call monitoring.

Create an agent in minutes

You start with a description written in plain language. This is where you define how the agent should behave during a call. You can then add documents, connect tools, and set limits on what it can do.

xAI says you can go from zero to a functional agent in about two minutes. You can also test it in your browser before connecting it to a phone number.

The knowledge base accepts text files, Markdown, Word, PowerPoint, Excel, HTML, and JSON, among other formats. Documents are grouped into collections that can be shared across several agents, for example to keep return policies, product details, or internal manuals in one central place.

The agent does more than answer questions. It can:

• Check an order’s status or issue a refund through an API.
• Book appointments in Google Calendar or Outlook Calendar.
• Send confirmations by email.
• Search for current information on the web or on X.
• Manage tickets in tools such as Linear or Notion.
• Transfer the call to a person when necessary.

The connectors can also communicate with MCP servers, a standard for connecting AI models to external tools and data. Companies can use the included services or connect their own APIs.

A single path for the conversation

Many voice platforms combine three separate services: one turns speech into text, another generates the response, and a third turns it back into audio. Each step can add cost, delay, and new points of failure.

Voice Agent Builder uses an integrated voice-to-voice path with Grok Voice. The model processes the conversation in real time, even with background noise, strong accents, interruptions, or the typical change of thought halfway through a sentence.

xAI says it trained Grok Voice on real calls and evaluates agents under those conditions using τ-voice Bench. The company mentions support for more than 25 languages, although the announcement does not provide specific results from the new evaluation system.

Voices, controls, and monitoring

You can choose one of the available voices or create a copy of a brand’s voice from about two minutes of audio. Each account includes a free phone number for testing the service and receiving traffic. You can also move an existing number using SIP, a common protocol for connecting calls over the internet.

All calls are recorded and transcribed. The team can listen to the audio, review the conversation, and check which tools the agent used. Real-time notifications let you intervene or follow its actions while the call is in progress.

Security controls, called guardrails, restrict specific behaviors. For example, they can prevent the agent from reading card numbers or moving beyond approved topics.

How much it costs

The announced rate is $0.08 per minute of audio, with voices included and no additional platform fee. The free phone number adds $0.01 per minute for telephony.

This model aims to simplify billing compared with platforms that charge separately for speech recognition, the language model, speech synthesis, and platform usage. Even so, the final cost will depend on call volume and the external connections each company uses.

For you, the most important change is that setting up a phone agent no longer requires integrating several APIs from day one. But the beta does not remove the need to test it with difficult cases: confusing calls, angry customers, sensitive data, and situations where it must know when to hand control to a person. That is where we will see whether the promise of quick setup also holds up in real operations.