Hello guys, Please clarify, is it possible to connect OpenAI Voice via API to our Assistant ? We need to resolve this issue, whether it is possible, since we haven't quite figured it out yet.

Yes, it’s possible to integrate OpenAI Voice with your Assistant via API, but the setup can be tricky depending on your assistant’s current architecture. I can guide you through the exact configuration to get it running smoothly and avoid common pitfalls. Can you share whether you’re looking for real-time streaming or pre-recorded responses? @Phill

In real time @Jarrah

Got it, real-time makes sense. Streaming OpenAI Voice in real-time can be tricky but fully doable, and the key is handling latency and audio chunking properly so the conversation feels natural. I can assist you on the exact setup that works reliably in production. If you want, we can continue this privately and I can map out the full solution for your assistant. @Phill

Yes, please. Please write to me.

Done ✅

Yes, it is absolutely possible to connect OpenAI Voice to your Vapi Assistant! There are two approaches depending on your needs: Option 1: OpenAI as a Voice/TTS Provider You can configure OpenAI as your text-to-speech (TTS) provider in your assistant's voice settings:

{
  "voice": {
    "provider": "openai",
    "voiceId": "alloy"  // Available voices: alloy, echo, shimmer
  }
}

How to set this up in the Dashboard: 1. Navigate to Assistants in your Vapi Dashboard 2. Select your assistant or create a new one 3. In the Voice section, choose

OpenAI

as the provider 4. Select your preferred voice (alloy, echo, shimmer) 5. Click Publish to save Option 2: OpenAI Realtime API (Speech-to-Speech) For ultra-low latency conversations, Vapi supports OpenAI's Realtime API, which processes audio input and output natively (speech-to-speech) without the traditional STT → LLM → TTS pipeline:

{
  "model": {
    "provider": "openai",
    "model": "gpt-realtime-2025-08-28",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      }
    ]
  },
  "voice": {
    "provider": "openai",
    "voiceId": "alloy"  // Supported: alloy, echo, shimmer, marin, cedar
  }
}

Available Realtime Models: •

gpt-realtime-2025-08-28

(Production-ready) •

gpt-4o-realtime-preview-2024-12-17

(Preview) •

gpt-4o-mini-realtime-preview-2024-12-17

(Preview, cost-optimized) Note: The realtime models have some limitations - Knowledge Bases are not currently supported, and certain voices (ash, ballad, coral, fable, onyx, nova) are not compatible. Documentation References: • OpenAI Voice Provider: [https://docs.vapi.ai/providers/voice/openai](https://docs.vapi.ai/providers/voice/openai) • OpenAI Realtime Integration: [https://docs.vapi.ai/openai-realtime](https://docs.vapi.ai/openai-realtime) Let me know if you need help configuring either option!