AetherGate

Live voice models for AI receptionists and calling agents

VOICE FOR BUSINESS

Build a receptionist people can talk to.

Explore flagship models for phone agents, appointment booking and customer support. Choose a conversational model for the whole dialogue, or build a pipeline with separate listening, reasoning and speaking models.

11 models · 6 providers · Documentation reviewed 2026-10-17. Provider access and availability vary.

Integration status: native provider APIs

Native provider integration: live audio connects directly to the provider. AetherGate does not proxy voice sessions, issue voice session tokens, meter their audio charges or enforce its gateway spending limits on these connections. Use a provider credential on your server; an AetherGate app key will not authenticate to these APIs.

The text reasoning stage of a modular pipeline can use AetherGate's chat API. Native voice charges stay on your provider account and are not included in dashboard savings or usage totals.

Showing 5 models

OpenAI · Live conversation

Native API

GPT-Live 1

gpt-live-1

Natural receptionist conversations

Full-duplex conversation: listen and speak at the same time, with backend delegation for business logic.

  • Simultaneous listening & speaking
  • Interruptions
  • Backend delegation

Provider billing

$0.05 / session minute, billed per second. Backend reasoning and tools are extra.

Current provider prices
Provider setup guide
Integration steps & model configuration
  1. Create a Live session on your backend with model gpt-live-1.
  2. Follow the provider connection guide for WebRTC or your telephony integration.
  3. Connect backend delegation to your scheduling, CRM or human handoff service.

Manage your OpenAI API key · Keep it on your server in OPENAI_API_KEY.

Configuration reference; implement the provider SDK and audio lifecycle from the setup guide. This does not create a connection or place a call.

{
  "provider": "OpenAI",
  "model": "gpt-live-1",
  "role": "conversation",
  "endpoint": "https://api.openai.com/v1/live/sessions",
  "transports": [
    "WebRTC",
    "WebSocket",
    "SIP"
  ],
  "credential_environment_variable": "OPENAI_API_KEY",
  "integration": "native_provider",
  "aethergate_proxy_supported": false,
  "setup_url": "https://developers.openai.com/api/docs/guides/voice-webrtc"
}
Model documentation

OpenAI · Live conversation

Native API

GPT-Realtime 2.1

gpt-realtime-2.1

Receptionists that book and take actions

Speech-to-speech reasoning and tool use for multi-step calls, appointment booking and customer support.

  • Tool calling
  • Interruptions
  • Configurable reasoning

Provider billing

$32 input / $64 output per million audio tokens. Text, caching and tools have separate rates.

Current provider prices
Provider setup guide
Integration steps & model configuration
  1. Connect your backend to the Realtime API with this model, or use the provider's WebRTC/SIP flow.
  2. Configure the voice, turn detection and tools your receptionist needs.
  3. Validate tool arguments on your backend and confirm bookings before writing to your calendar.

Manage your OpenAI API key · Keep it on your server in OPENAI_API_KEY.

Configuration reference; implement the provider SDK and audio lifecycle from the setup guide. This does not create a connection or place a call.

{
  "provider": "OpenAI",
  "model": "gpt-realtime-2.1",
  "role": "conversation",
  "endpoint": "wss://api.openai.com/v1/realtime",
  "transports": [
    "WebRTC",
    "WebSocket",
    "SIP"
  ],
  "credential_environment_variable": "OPENAI_API_KEY",
  "integration": "native_provider",
  "aethergate_proxy_supported": false,
  "setup_url": "https://developers.openai.com/api/docs/guides/realtime"
}
Model documentation

OpenAI · Live conversation

Native API

GPT-Realtime 2.1 Mini

gpt-realtime-2.1-mini

High-volume call routing and FAQs

A smaller realtime reasoning model for shorter, repeatable calls with tool access.

  • Tool calling
  • Interruptions
  • Lower-cost audio tier

Provider billing

$10 input / $20 output per million audio tokens. Text, caching and tools have separate rates.

Current provider prices
Provider setup guide
Integration steps & model configuration
  1. Use the Realtime API with model gpt-realtime-2.1-mini.
  2. Test your FAQs, names and booking workflow against your actual call audio.
  3. Define a human transfer path for calls the model cannot resolve.

Manage your OpenAI API key · Keep it on your server in OPENAI_API_KEY.

Configuration reference; implement the provider SDK and audio lifecycle from the setup guide. This does not create a connection or place a call.

{
  "provider": "OpenAI",
  "model": "gpt-realtime-2.1-mini",
  "role": "conversation",
  "endpoint": "wss://api.openai.com/v1/realtime",
  "transports": [
    "WebRTC",
    "WebSocket",
    "SIP"
  ],
  "credential_environment_variable": "OPENAI_API_KEY",
  "integration": "native_provider",
  "aethergate_proxy_supported": false,
  "setup_url": "https://developers.openai.com/api/docs/guides/realtime"
}
Model documentation

Google · Live conversation

Native API

Gemini 3.8 Live

gemini-3.8-live

Multilingual voice and visual assistants

Native audio conversations with interruptions, tool use and optional camera or screen context.

  • Multilingual
  • Interruptions
  • Tool calling
  • Visual input

Provider billing

Audio and other input/output usage at Google Live API rates. See the current provider price sheet.

Current provider prices
Provider setup guide
Integration steps & model configuration
  1. Use Google GenAI SDK live.connect with model gemini-3.8-live and AUDIO responses.
  2. Send microphone audio in the format required by the Live API; configure tools on your server.
  3. For direct browser connections, mint short-lived provider tokens on your backend.

Manage your Google API key · Keep it on your server in GEMINI_API_KEY.

Configuration reference; implement the provider SDK and audio lifecycle from the setup guide. This does not create a connection or place a call.

{
  "provider": "Google",
  "model": "gemini-3.8-live",
  "role": "conversation",
  "endpoint": "wss://generativelanguage.googleapis.com/ws/google.ai.generativelanguage.v1beta.GenerativeService.BidiGenerateContent",
  "transports": [
    "WebSocket"
  ],
  "credential_environment_variable": "GEMINI_API_KEY",
  "integration": "native_provider",
  "aethergate_proxy_supported": false,
  "setup_url": "https://ai.google.dev/gemini-api/docs/live-api/get-started-sdk"
}
Model documentation

xAI · Live conversation

Native API

Grok Voice 3.0

grok-voice-3.0

Phone agents and tool-enabled support

Conversational voice with reasoning, server-side tools and phone-compatible audio formats.

  • Tool calling
  • Interruptions
  • Phone audio codecs

Provider billing

Voice session pricing; optional tool charges are separate. Check xAI's current rates.

Current provider prices
Provider setup guide
Integration steps & model configuration
  1. Open the provider Realtime connection with model=grok-voice-3.0.
  2. Set the voice, instructions, turn detection and business tools with session.update.
  3. Follow the provider telephony guide to bridge a phone call, or use ephemeral tokens for clients.

Manage your xAI API key · Keep it on your server in XAI_API_KEY.

Configuration reference; implement the provider SDK and audio lifecycle from the setup guide. This does not create a connection or place a call.

{
  "provider": "xAI",
  "model": "grok-voice-3.0",
  "role": "conversation",
  "endpoint": "wss://api.x.ai/v1/realtime",
  "transports": [
    "WebSocket",
    "WebRTC",
    "SIP"
  ],
  "credential_environment_variable": "XAI_API_KEY",
  "integration": "native_provider",
  "aethergate_proxy_supported": false,
  "setup_url": "https://docs.x.ai/developers/model-capabilities/audio/speech-to-speech"
}
Model documentation

Two practical ways to build your calling agent

One conversational model

Start with GPT-Live, GPT-Realtime, Gemini Live or Grok Voice. Your phone service streams the call to the provider; your backend handles booking, lookups and handoff. A model alone does not provision a phone number or connect to the telephone network.

Choose each stage

Use Flux or Scribe to listen, an AetherGate text model to decide what to say, and Sonic, ElevenLabs or Flux TTS to speak. Your application coordinates the streams and interruptions. Only the text requests routed through AetherGate appear in its spend totals.

Text gateway quickstart →