Realtime Voice

Where voice fits in ResponseCX — and where the full Voice platform documentation lives.

Realtime Voice

ResponseCX agents can answer the phone. An agent with a voice_model set is a voice agent: the same policies, settings, and functions, delivered over a live call instead of chat — built on Twilio for telephony and the OpenAI Realtime API for speech.

This page is orientation. The Voice platform — API, SDKs, webhooks, supervision, operations — is documented in StateSet Voice.

What a voice interaction looks like

1. Verify the caller — email entry plus a one-time password, persisted so a returning caller skips it.
2. Load context — order history and preferences, so “where’s my order?” needs no order number.
3. Act — modify an order, manage a subscription, take a payment — through the same functions a chat agent uses.
4. Escalate — a human supervisor can monitor, whisper, barge, or take over an in-flight call.

Configure a voice agent

curl -X PUT https://api.stateset.com/v1/agents/{id} \
  -H "Authorization: Bearer $STATESET_API_KEY" \
  -H 'content-type: application/json' \
  -d '{ "voice_model": "gpt-realtime", "activated": "true" }'

Number routing, call placement, transcripts, and webhooks are on the Voice API — the agent record only selects the model.

A voice agent acts in real time with no review step between utterance and effect. Wire consequential actions — refunds, cancellations — through the decision gate before going live, not after the first bad call. Voice is the surface where “the model decided” is most expensive.

Watch a live call

Supervision is part of the platform, not an afterthought:

# List active sessions, then monitor one
curl https://api.stateset.com/v1/voice/sessions?status=active \
  -H "Authorization: Bearer $STATESET_API_KEY"

session.action webhook events record every supervisor act — monitor, whisper, barge, escalate — so the audit trail of who intervened on a call is part of the call record. Details in Voice webhooks.

Related

• StateSet Voice — the platform
• Voice quickstart — a number answering in minutes
• Voice webhooks — call lifecycle events, signed
• Voice MCP server — drive it from an agent