Capability · Real-Time Voice
Two ways to power a conversation — pick what fits.
Run the standard pipeline (speech-to-text, a language model, text-to-speech) or a single real-time audio-native model — both produce a real conversation, with different trade-offs.
What it actually does
- Standard pipeline: separate speech recognition, language understanding, and voice, each independently configurable
- Real-time option: one audio-native model handles the whole conversation directly
- Switch per script — not a platform-wide, all-or-nothing choice
- Both options support live interruption and human handoff the same way
How it works
The pipeline
One capability, end to end.
Call arrives
Pipeline or real-time model selected per script
Conversation happens
Response delivered
Handoff or booking happens the same either way