New Live API feature -- you can set decisions model triggers in the API, for example classify user input or click a UI element. This is much faster than a function call and pairs really nicely with live voice conversations. See voice demo about halfway through this video:

Here's some example config, in this case we're configuring Live API to make 3 decisions for every chunk of user-input transcription. The server-side latency is like 100-150ms.

See documentation here for Live API

And blog post, which has an interactive voice decisions demo

let's go @charlierguo! love the decisions voice demo

@romainhuet Nice

@daidaijiayu woahh this solves one of my biggest problems - thank you!

@daidaijiayu yall are cooking!!

dope

Jev is 20x cheaper. Can we at least make this close to Jev if you can’t use your sub on this

faster than function calls for voice triggers is actually a game changer. no more awkward pauses mid-conversation.

The source-event IDs in the docs are a nice detail. A character smiling at something the speaker has already corrected would feel stranger than a small delay.