Who is the call for, and on which path?
Start from language, setting, and the purpose of the conversation. Browser and mobile voice run today; telephony is built and key-gated.
INTUNELABS VOICE PIPELINE
A realtime pipeline tuned to Korean conversation.
A Korean-first pipeline that listens and answers out loud. It waits through a backchannel, tells barge-in from a nod, and steps a capability down instead of dropping the call. We orchestrate vendor speech and language models; the conversation layer is ours.
Request a demo
TURN-TAKING
A pause is not the end of the turn.
A brief “yes” or a pause is not treated as the end of a turn. Hear when the agent listens and answers in a demo call. We do not publish latency benchmarks.
See the call flow in Simcheong
CONFIGURED PER SERVICE
Voice and role are set in the console.
Console is where the team sets the agent’s voice, role, and greeting. The pipeline runs that configuration the same way in the browser and on mobile.
Explore Console
DEGRADE, DON'T DROP
When a path is uncertain, the call continues.
Risky paths are off by default. When a capability is uncertain, the pipeline steps it down, and speech synthesis has a pinned failover. Silence is watched so a quiet call is not left hanging. We do not publish latency benchmarks.
SETTLED BEFORE DEPLOY
The job of the call, and where a person steps in.
What does the team review?
Role bounds, situations that need a person, and data handling — including retention — are scoped per deployment. Risky paths are off by default.
Walk the pipeline on a demo call.
We’ll review turn-taking, reliability, and which transports belong in this deployment. We do not train the speech or language models, and per-call timings are instrumented for operations — not published as benchmarks.