Voice calls with agents now have a redesigned in-call interface, and the biggest addition is a live transcript — the call is transcribed as it happens, showing both your speech and the agent's replies while the conversation is still going.
The new call interface with real-time transcription in Agentic OS.
What Changed
Live transcript during the call. Both sides of the conversation stream into the call interface as text while you speak, so you can follow along, catch a name or a number the agent said, and confirm what was heard — without waiting for the call to end.
A cleaner in-call surface. The redesigned interface keeps the call controls, the agent identity, and the running transcript in one view, so a call reads like a conversation rather than an opaque audio session.
Transcripts persist after the call. When the call ends, it lands in Analytics → Transcripts with the rest of the agent's conversations — same topic chips, sentiment scoring, model attribution, and per-conversation cost, so a call is reviewable the same way a chat is.
How It Fits the Existing Call Features
This builds on the voice-call stack already documented on the platform. Calls are real-time WebRTC sessions powered by LiveKit, started from the waveform icon in the chat composer, and configured per agent under Configurations → Voice → Voice call.
That configuration still governs everything around the new interface: Live conversation vs. Step-by-step call style, spoken language, the AI provider powering the call (OpenAI, Google, or Anthropic), the call voice, and whether screen sharing is allowed. Screen sharing sessions continue to use live conversation.
Documentation
- Agent Settings: Voice Call — call style, spoken language, AI provider, call voice, and screen sharing
- Chat — starting a voice call from the chat composer
- Analytics: Transcripts — reviewing conversation transcripts, topics, sentiment, and cost