Blog
Notes on building voice AI.
The pipeline, the providers, the tradeoffs. Short reads from the team that ships Vocily AI.
- 2026-05-283 min readFoundations2026-05-28 · 3 min
How a voice AI agent actually works
Before you pick providers or tune prompts, it helps to know what is happening inside a voice agent on every turn. This is the picture every team should have in their head.
- 2026-05-212 min readStack2026-05-21 · 2 min
Text to speech (TTS): what to pick and why
The voice your agent uses is the first thing a customer judges. TTS choice affects naturalness, latency, language coverage, and cost — all four at once.
- 2026-05-142 min readStack2026-05-14 · 2 min
Speech to text (STT): codemix, accents, and the streaming game
Indian customers code-switch mid-sentence. Most STT engines do not. The provider you pick shapes how forgiving your agent feels.
- 2026-05-072 min readStack2026-05-07 · 2 min
Telephony: getting the call to the agent in one piece
Voice AI without telephony is a web demo. Real product means routing real calls, managing real numbers, and handing off to real humans without dropping the line.