Engineering, tutorials, and product updates from the core team.
Twilio sends G.711 µ-law at 8kHz. Deepgram expects PCM at 16kHz. Misconfigure the transcode and the call sounds fine on your end — the caller hears static. No error. No log.
Most teams run two separate pipelines — one for Twilio phone calls, one for browser WebRTC. HZRelay gives both the same session interface: same events, same SDK, one config line to switch.
OpenAI streams tokens. Your web UI needs them. So does your mobile app. So does your analytics pipeline. HZRelay fans one source stream to N subscribers — zero copy, backpressure handled.
Stripe sends `payment_intent.succeeded`. Twilio sends `CallStatus=completed`. GitHub sends `push`. Three shapes, three retry policies, three failure modes. One HZRelay session normalizes all of them.
STT: 298ms. LLM first token: 370ms. TTS first audio: 75ms. Transport: 19ms. Total: 762ms — and that's good. Here's how to measure every stage and which provider swaps actually move the needle.
New adapters, latency guides, and SDK releases — direct to your inbox. No spam.