Every realtime voice API today — OpenAI Realtime, Gemini Live, Hume — assumes one person talking to one AI. Sun, a new Product Hunt launch, is built for the case that breaks: a room with multiple humans and multiple agents all sharing the same audio channel.
## What it actually does
Sun is a voice-first model for collaborative interaction — meetings, group calls, classroom debates, multi-agent workflows. It handles multi-speaker turn-taking, tracks who’s talking, and does agent-aware barge-in instead of crude voice-activity detection. You can even run two AIs against each other on a live audio channel through a single API.
## Why it’s interesting
The numbers it leads with are context and concurrency: roughly 10x the context window of ChatGPT Realtime and Gemini Live, plus native multi-speaker awareness rather than a single-user assumption bolted onto a group setting. There’s a browser playground to try it with no card required, which is the fastest way to tell whether the turn-taking actually holds up with three voices in the room.

Leave a comment