Jev decides which agent a user is addressing in a voice session
The Jev inference call adds about 100ms, keeping voice-to-voice latency well under the 1,500ms P95 target, with full code in the author's thread.
I love this Jev + voice demo: Jev decides which of the two agents in the session the user is talking to. The Jev inference call adds about 100ms, so voice-to-voice latency here is still way under our target of 1,500ms P95. (This is using the very fast PhoneLLM model.) Jon's thread has another example of six "characters" in the session. Full code is in his thread, too. (It's worth noting that Claude and Codex can both one-shot complex Pipecat agents like this, now, getting all the low-level hard things like turn detection right, and using abstractions like workers to build sub-agents and handoff.)
This post quotes another post: “Opus 5.5 one-shots a two-character voice bot pipeline”