Turn detection: built in, or build it yourself.
The hard realtime part is knowing when the caller is done and handing the model the right context. Omni packages turn detection inside the agent, so you don't need a separate turn-detection product.
Built into Omni
Included- Turn detection (endpointing) the moment speech ends
- Grounded KB context retrieved inline
- Included in the Omni voice-agent stack
- No separate product to price and integrate
Build it yourself
Your time + infra- Tune your own VAD / silence thresholds per accent and line
- Stand up + host a vector DB and retrieval service
- Glue transcription, endpointing, and retrieval yourself
- Own the latency, scaling, and on-call
Use Omni for the whole loop, or Cue to keep your own voice
Omni ($0.05/min) gives you the whole agent end-to-end - hearing, turn-taking, grounding, thinking, and speaking in one hop. Want to keep your own LLM and the TTS that sounds best for your language or brand? Cue ($0.015/min) hands you telephony-grade turn detection and inline knowledge-base grounding, and lets you bring the rest, your LLM and your voice.
Choose how you want to start.
Use an instant sandbox key for the API, or open Agents Beta and build without code. Sandbox usage never bills. Live keys require available credit.
Agents is live in beta. Sandbox keys have daily limits and never touch billing.