A big week for VoiceInfra
September 2026
As applications become increasingly LLM-driven, the line between building software and simply describing what you want keeps getting thinner. We’ve been building with that shift in mind, and recently shipped updates that move VoiceInfra further in that direction.
The one I’m most excited about is Sculptor. It’s a new way to build in the portal. You describe what you need in plain language, English or any other, and Sculptor chisels it out for you: an assistant, a workflow, functions, a knowledge base. Every change is presented as a clear before and after, and nothing is set in stone until you approve it. It lives in a docked panel that follows you through the app, or opens full screen as its own workspace. Sculptor is in beta while we continue to refine it.
Alongside it, we shipped MCP support. VoiceInfra now connects to the agent tools you already use: Claude, ChatGPT, anything that speaks the protocol. You can build agents and manage workspaces directly from those tools, with access visible and revocable through the Connected Apps page.
On the call-handling side, we added function lifecycle: functions can be tied to a specific point in a call: before the agent picks up, during the conversation as a tool, or after the call ends. Calls can now be transferred between agents and workflows, so you can build modular agents that connect. We also improved the widget and Test Chat: talk to any workflow in text, watch the canvas update live, and use the same workflow for voice and text, transfers included.
Together, these represent where we think building is heading: less configuration, more intent.