VoiceBlender is an open-source Go service that fuses SIP, WebRTC, and AI into a single real-time audio pipeline. It lets you spin up multi-party voice sessions, drop AI agents into live calls mid-conversation, and control everything through a simple REST API — no PBX required.
In this talk we'll walk through the architecture behind a system that treats a phone call and an AI agent as equal participants in the same audio mix. We'll explore the trade-offs of doing real-time media in Go, how to make legacy telephony and modern AI providers speak the same language, and why answering machine detection is still a surprisingly hard signal-processing problem in 2026.
You'll leave with a working mental model for building outbound dialers, AI-powered IVRs, and real-time voice agents — entirely in Go.


