INTRODUCING HZRELAY: ONE SDK FOR EVERY AI STREAM
Every team building an AI voice agent hits the same wall within the first week. You wire Twilio to receive the call, Deepgram to transcribe it, OpenAI to generate a response, ElevenLabs to speak it back — and then you spend two weeks debugging codec mismatches, reconnect failures, and the fact that Twilio sends mulaw 8kHz while every AI API expects PCM 16kHz.
That plumbing is not your product. It is infrastructure tax. Every team pays it separately. HZRelay pays it once.
WHAT HZRELAY IS
HZRelay is a real-time stream routing layer. It sits between your providers and your application code. You describe what you want — inbound source, STT provider, LLM, TTS provider, outbound sink — and HZRelay handles everything in between.
THREE PAYLOAD TYPES
HZRelay is not just a voice tool. The routing primitive works for any real-time payload.
[VOICE]
Phone calls (Twilio/Plivo) and browser WebRTC under one session model. Same events, same SDK, one config line to switch transport.
[TOKENS]
LLM token streams routed to multiple subscribers simultaneously. Fan-out Agent A output to Agent B, a web UI, and analytics — one source.
[EVENTS]
Webhook payloads from Stripe, GitHub, Twilio — normalized to a standard shape, fanned out to multiple services with guaranteed delivery.
TWO DEPTHS — ZERO LOCK-IN
Depth 1: use HZRelay as pure infrastructure. Routing, codecs, reconnects handled — pipe events into your own LangGraph or custom agent. Depth 2: add an agent: config block and we run the transcript→LLM→TTS loop automatically. Same session, same pricing, switch anytime.
GET STARTED
Free tier — 100 minutes/month, no credit card. Working AI phone call in 30 minutes. Read the quickstart or grab an API key.