← Back to feed

xAI launched Voice Agent Builder in beta on July 1, 2026, a no-code platform that turns Grok Voice Think Fast 1.0 into a configurable call-centre agent without a single line of code. The tool is available at x.ai/voice.

Building a voice agent previously required direct integration of xAI’s voice API, custom telephony plumbing, and tooling logic wired by hand. Voice Agent Builder replaces that with a point-and-click interface: pick a voice, describe what the agent should do, connect external tools, and deploy to a phone number that xAI provisions automatically on account creation.

What ships in beta

The builder provides access to over 80 pre-set voice tones across multiple accents and registers. Operators who need a specific voice can record two minutes of audio from any speaker and clone it directly in the platform — xAI does not require studio-quality recording.

External integrations available at launch include Notion and Gmail. The documented use cases are call-centre flows (a customer calls, the agent looks up order status via API, provides an answer, creates a follow-up ticket) and reservation desks (caller states a time, agent books a Google Calendar entry and sends a confirmation email). Both workflows run without human intervention once configured.

The API cost is $0.05 per minute of conversation. For a business handling 10,000 minutes of inbound calls per month, that is $500/month in voice AI costs before telephony charges. Human agents in a US-based call centre typically cost $20-35/hour fully loaded.

The model underneath

Grok Voice Think Fast 1.0 is xAI’s existing production voice model, released in June 2026 with a 70% autonomous resolution rate at Starlink’s internal customer support desk. xAI has benchmarked it as producing higher quality audio than Google’s Gemini 3.1 Flash Live and OpenAI’s GPT Realtime 1.5, though independent audio quality comparisons have not yet been published by a neutral evaluator.

The model supports conversation in Japanese as confirmed by xAI’s own demo materials; the full supported-language list for the builder has not been published.

Strategic context

Voice agent builders have existed in various forms since 2023. What changes in 2026 is the quality floor. First-generation voice AI was reliably recognizable as non-human within three seconds; GPT Realtime, Gemini Live, and Grok Voice Think Fast are now generating audio that users in blindfolded tests identify as human at rates above 60%.

xAI’s move puts a production-grade voice agent infrastructure in front of developers who previously needed to integrate OpenAI Realtime API, Eleven Labs, or Twilio separately. The $0.05/minute price is aggressive — Twilio Flex charges $1/user/hour in its agent-facing product, and OpenAI’s Realtime API runs $0.10/minute for audio input plus output. xAI packages telephony, model inference, and tool execution into a single per-minute charge.

The competing products are Retell AI (same no-code category, $0.07/minute), Vapi (developer-first, $0.05/minute, more configuration overhead), and Synthflow ($0.13/minute for comparable features). xAI enters at the bottom of that pricing range with a frontier-tier model.

Caveats

Voice Agent Builder is in beta. Production reliability, latency SLAs, and telephony coverage by country have not been disclosed. The voice cloning feature raises standard consent and misuse questions that xAI’s terms of service will need to address explicitly — voice cloning at two-minute turnaround is an identical threat model to any deepfake audio tool. xAI’s existing track record on safety guardrails for Grok has been mixed.

The beta also operates without published uptime guarantees, which matters for businesses that route real customer calls through the system.