xAI's Voice Agent Builder lets you create production voice agents in 2 minutes without code

xAI has launched Voice Agent Builder in beta, a no-code platform that lets operators and developers build production-grade voice agents on Grok Voice in under two minutes. The platform combines telephony, knowledge retrieval, tools, guardrails, MCPs, observability, voice cloning, SIP, and call review into a single interface, eliminating the need to stitch together separate speech-to-text, language model, and text-to-speech APIs. Users write a plain-language description of how calls should flow, attach documents (plain text, Markdown, Word, PowerPoint, Excel, HTML, JSON, and more), and deploy a working agent instantly. Each account includes a free phone number, and existing numbers can be brought over via SIP. Agents can use any of 80+ built-in voices or a custom voice clone made from about two minutes of audio. The platform also supports tools for actions like scheduling appointments, checking order status, searching the web, or transferring calls to human teams. Every call is recorded and transcribed for review, with guardrails to limit unwanted behaviors.

xAI emphasizes that Grok Voice was trained on real-world calls with low-quality telephony audio, background noise, strong accents, interruptions, and 25+ languages. The τ-voice Bench measures agent performance under these conditions, and Grok Voice Think Fast 1.0 achieved 67.3%, significantly ahead of Gemini 3.1 Flash Live (43.8%) and GPT Realtime 1.5 (35.3%). Voice Agent Builder uses a speech-to-speech path that is tightly coupled to the model rather than assembled from separate components, reducing cost, latency, and failure modes. The platform is designed for high-volume production use cases such as customer support, sales, booking lines, and other conversational AI applications that require real-time reasoning, tool use, and multi-language support.