Skip to main content
OpenAI Realtime is a speech-to-speech model. It listens, thinks, and speaks in one connection, replacing the separate voice and transcription steps. Because nothing converts speech to text and back on the way through, replies come faster and the model hears how something was said, not just what was said.

​
What you need

The key is an ordinary OpenAI key — the same kind used for the standard OpenAI connection. Get it from platform.openai.com/api-keys, and see that page for how project scoping works and why production should use a service-account key.
Realtime is a separate connection even though it uses the same kind of key. Connecting OpenAI does not connect OpenAI Realtime, and you can point them at different projects if you want the usage billed apart.

​
Adding the connection

  1. Go to Connections and choose OpenAI Realtime.
  2. Paste the API key.
  3. Save.

​
What you can select afterwards

Turn on Realtime in Model Configurations and choose OpenAI Realtime as the provider. Model and voice — pick from the dropdowns. OpenAI’s realtime guide covers what each voice sounds like. Language — English (US, GB, AU), Arabic (SA, EG, AE, IQ), Portuguese (PT, BR), German, French, Spanish, Italian, Japanese, Korean, and Chinese.

​
How a speech-to-speech call differs

A speech-to-speech model replaces the transcriber and the voice provider. When a workflow is set to realtime, those two settings no longer apply.
  • The voice comes from the model. You cannot pair a realtime model with a voice from another provider.
  • You still need a language model configured. It runs the work the realtime model does not — extracting variables and running QA analysis after the call. Without one, those features silently produce nothing.
  • Context compaction does not apply and is ignored if enabled.

​
Realtime or standard OpenAI

Standard OpenAI lets you pair its language model with a different voice or transcriber. Realtime is one model doing everything, so that flexibility is gone — you take its voices and its behaviour together. Choose realtime when responsiveness matters most. Choose standard when you want a specific voice, a specific model, or providers mixed across the pipeline.

​
Troubleshooting

Confirm the project the key belongs to has billing enabled and access to the realtime model. Realtime is not available on every account tier.
Both need an LLM configured alongside the realtime provider. Check the LLM tab in Model Configurations.