Skip to main content
Voice is the original and most-used channel. A call moves through four hops:
  1. Inbound audio streams from the telephony provider over WebSocket.
  2. The transcriber converts audio to text in near real time.
  3. The LLM receives the transcript plus conversation history and generates a response, optionally calling tools.
  4. The synthesizer converts the response to audio, streamed back to the caller.

​
Making an outbound call

Place an outbound call to a contact from the dashboard or trigger one as part of a campaign.

​
Receiving inbound calls

Point your telephony number at Agentic OS — see Receive Incoming Calls.

​
Latency

Each hop above adds latency. Typical end-to-end latency on a well-tuned agent is 600–900ms. To reduce it:
  • Use a streaming-capable LLM configuration
  • Choose a low-latency transcriber model
  • Tune the synthesizer’s buffering to start playback early
  • Keep response length short for faster generation