xAI launches Voice Agent Builder with $0.05/min pricing and SIP routing

• xAI opened a beta Grok Voice Agent Builder launch post for building Grok-powered voice agents in the developer console, and TestingCatalog's console screenshot shows it as a first-class Voice surface alongside text-to-speech, speech-to-text, and a voice library.
• The builder is positioned as no-code: ai_for_success's feature rundown says agents can be assembled in under two minutes, while WesRoth's launch summary says teams can add instructions, company knowledge, APIs, MCP tools, and guardrails without stitching separate speech and TTS stacks.
• xAI priced Grok Voice at $0.05 per minute of audio, and WesRoth's pricing breakdown adds another $0.01 per minute when calls use an xAI-provided phone number.
• Telephony is part of the launch surface: WesRoth's deployment details says every account gets a phone number and existing numbers can connect over SIP, while ai_for_success's reply about phone numbers called the included number one of the standout bits.
• The beta also ships with broad voice and connector coverage, because ai_for_success's launch thread lists 25-plus languages, 80-plus built-in voices, voice cloning, and integrations for Gmail, Google Calendar, Outlook, Notion, and MCP-based tools.

You can watch the console flow in xAI's launch clip, skim TestingCatalog's screenshot of the new Voice Agents page, and WesRoth's walkthrough adds the deployment details that matter most for engineering teams: SIP, call transcripts, tool logs, and the telephony surcharge.

The console UI in TestingCatalog's screenshot exposes Voice Agents as a beta product with prebuilt templates for customer support, sales, appointment scheduling, personal assistant, and lead qualification. The same screen also shows the broader voice stack in the sidebar: agents, text-to-speech, speech-to-text, and a voice library.

The attached demo in xAI's launch clip ends on a Deploy button, which is the clearest signal that xAI is selling this as an end-to-end surface rather than a raw model primitive.

Between ai_for_success's launch thread and WesRoth's walkthrough, the builder flow breaks into a short, pretty standard voice-agent stack:

• instructions for the agent
• connected company knowledge
• APIs and MCP tools
• guardrails
• voice selection or cloning
• phone deployment
• recorded calls with transcripts and tool-use review

The video attached to WesRoth's walkthrough shows that last part clearly: a live phone interaction with real-time transcription and visible tool execution logs.

The pricing model is simple enough to quote in one line. Grok Voice is $0.05 per minute, with no separate platform fee, and calls placed on xAI-provided numbers add $0.01 per minute in telephony charges, according to WesRoth's pricing breakdown.

Routing is a little more interesting. WesRoth's deployment details says every account includes a phone number, while existing business numbers can be connected through SIP, which puts the product closer to a deployable contact surface than a sandbox voice demo.

The launch pitch in ai_for_success's feature rundown bundles together the two things most builders check first:

• 25-plus supported languages
• 80-plus built-in voices
• optional custom voice cloning
• Gmail integration
• Google Calendar integration
• Outlook integration
• Notion integration
• MCP support

The sample templates in TestingCatalog's console screenshot line up with that connector story. The personal assistant template explicitly points at email and calendar access from connected apps.

One launch wrinkle surfaced in the thread itself. In a reply about outbound calling compliance, ai_for_success said outbound calling still needs legal clearance and compliance steps, with a hope that broader support lands later.

That is not in tension with the launch materials, which focus on inbound-style phone deployment, SIP connection, and call review. It does add one concrete boundary: the beta shipped with telephony hooks, but the thread discussion suggests outbound workflows are still gated by compliance realities rather than just product switches.