Skip to main content
Gemini Live is a speech-to-speech model. It listens to the caller and speaks back directly, without a separate transcriber or voice provider in between. This is a different connection from Gemini, even though both use an AI Studio API key. Connecting one does not connect the other.

​
What you need

​
Getting your API key

  1. Go to Google AI Studio.
  2. Sign in and create an API key.
  3. Pick a Google Cloud project when prompted, then copy the key.
The same key works for both Gemini and Gemini Live. You still add it to each connection separately.
If you need Google Cloud billing, IAM controls, or data kept in a specific region, use Vertex AI Gemini Live instead. It runs the same family of models through your own Google Cloud project, authenticated by a service account rather than an API key.

​
Adding the connection

  1. Go to Connections and choose Gemini Live.
  2. Paste the API key.
  3. Save.

​
What you can select afterwards

Choose your model on Model Configurations. Model — Google’s native-audio Live model. See the Live API models. Voice — selected by name; Puck is the default. Google’s voice list describes how each one sounds. Language — English (US, GB, AU, IN), Spanish, French, German, Portuguese (BR), Japanese, Korean, Mandarin, Hindi, four Arabic variants, and seven Indian languages, among others.

​
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 model hears the caller and speaks back itself. A few consequences worth knowing before switching a live workflow over:
  • The voice comes from the model. You cannot pair a Gemini Live model with a voice from another provider.
  • You still get a transcript. Both sides of the call are transcribed for the run record, so tracing and post-call analysis work as usual.
  • Moving between nodes reconnects the session. Gemini Live cannot change its instructions mid-session, so a node transition restarts it. The reconnect waits until the agent has finished speaking, so the caller does not hear it cut off.
  • 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. See Realtime.

​
Troubleshooting

Confirm you copied it from Google AI Studio rather than the Google Cloud console. A service-account JSON is a different credential and belongs on Vertex AI Gemini Live.
Gemini and Gemini Live are separate connections. Add a Gemini Live connection using the same API key.
The workflow is not set to realtime. Switch it over on Model Configurations, then pick the Gemini Live model — the transcriber and voice settings stop applying at that point.