Google Live API adds decisions triggers with ~100-150ms server latency for live voice agents

pbbakkum · x · 2026-10-23

A Google engineer demoed a new Live API feature: you can set decisions model triggers in the API, letting the model make multiple lightweight decisions per chunk of user-input transcription — e.g. classifying input or clicking a UI element. This is much faster than a function call and pairs well with live voice conversations.

The shared example config fires three decisions per transcript delta on the session.inputtranscript.delta event: expression (neutral/happy/sad/concerned), intensity (low/medium/high), and gesture (none/nod/shake/lean) — useful for driving real-time avatar behavior. Server-side latency is roughly 100-150ms. Docs are available via the Live API documentation.

Related event: Google Live API Adds Decision Triggers for Faster Responses(2 posts)→

Original post →

More from coding & agent

coding & agent channel →