Point an agent at Gemini Live and the caller talks to Google’s duplex speech-to-speech model directly — no separate transcriber or text-to-speech stage. The caller’s audio streams into one bidirectional session that listens and speaks on the same connection, so the agent starts replying sooner and barge-in feels instant. This is a shipped, first-class provider. TeleQuick Voice ships a native Gemini Live adapter that speaks Google’s BidiGenerateContent WebSocket protocol end to end: it opens the session, upsamples the call’s audio to the rate Gemini expects, streams the model’s spoken reply back onto the leg, and wires Gemini’s server-side voice activity detection to barge-in. You turn it on by naming gemini as the provider in your agent config — there is no code to write.
Gemini Live is one of several duplex providers. For the shared model of how a speech-to-speech agent works — the single REALTIME node, the greeting behavior, session budgets — see BYO Speech-to-Speech Models. This page covers the Gemini-specific config, voices, and behavior.

​
Configure a Gemini Live agent

A duplex agent has a single node. Set the entry node to REALTIME, name gemini as the provider, and pick a model and a voice.
1

Name gemini in the agent spec

entry_node: REALTIME
pipeline:
  REALTIME:
    node_type: REALTIME
    provider: gemini
    model: gemini-2.0-flash-exp    # "models/" prefix added for you if omitted
    voice: Aoede                   # Aoede | Charon | Fenrir | Kore | Puck
    instructions: |
      You are a friendly support agent for Acme. Keep replies short and
      speak naturally.
​
provider
string
required
Set to gemini to select the Gemini Live adapter.
​
model
string
A Gemini Live model id, e.g. gemini-2.0-flash-exp. The adapter qualifies it to Gemini’s required models/<id> form for you, so either gemini-2.0-flash-exp or models/gemini-2.0-flash-exp works.
​
voice
string
One of Gemini’s prebuilt voices: Aoede, Charon, Fenrir, Kore, or Puck. Leave it empty to accept Google’s server default. The runtime defaults to Aoede when you omit it.
​
instructions
string
The system prompt. It is sent as Gemini’s systemInstruction in the opening setup message and also seeds the inbound greeting.
2

Supply your Gemini API key

Credentials are stored per tenant and never shared across workspaces. Add your Gemini API key in the agent console at agent.telequick.dev, or via the control-plane API at portal.telequick.dev. The runtime reads it when it opens the session — no key means the provider returns no session and the call falls back to a silent passthrough rather than dropping.
3

Point a number at the agent and call

Save the agent, attach a phone number or trunk to it, and place a call. TeleQuick Voice opens the Gemini Live WebSocket on the first inbound audio frame and streams both directions from there.

​
How the audio flows

You never touch sample rates — the adapter handles the conversions Gemini requires:
1

Caller audio in

The call’s 8 kHz PCM16 is upsampled to the 16 kHz PCM16 Gemini Live expects on input and streamed as realtimeInput.audio frames.
2

Model audio out

Gemini returns 24 kHz PCM16 audio, which the runtime resamples down and paces back onto the call leg (G.711 to the caller on a phone leg, Opus to a browser leg).
3

Barge-in

When the caller talks over the agent, Gemini’s server VAD emits an interrupted signal; the runtime cuts the in-flight reply immediately.

​
Turn-taking

For a REALTIME entry node the runtime defaults to server_vad — Gemini Live does its own endpointing and interruption detection, so you do not add a local turn detector. When Gemini reports an interruption the runtime maps it to barge-in and stops the outgoing audio. If you want to tune how aggressively the agent yields, or to understand the fallback behavior for models without server VAD, see Turn Detection & Barge-In.

​
Greeting on inbound calls

Some carrier SBCs withhold RTP until they hear yours. On an inbound call the runtime prompts Gemini to speak first so the media path opens — the model greets from your instructions. This is on by default for inbound direct-media calls; no extra config is required.

​
Tools mid-conversation

The Gemini Live adapter renders your configured tools into Gemini’s functionDeclarations in the opening setup message, receives toolCall events during a spoken turn, and returns your result as a toolResponse — so the model can call an HTTP or MCP tool and speak the outcome without leaving the call. Configure tools exactly as you would for any agent; see Tool Calling.
Function calling over Gemini Live is newer than the OpenAI Realtime path, which is the most battle-tested realtime adapter today. If your agent leans heavily on complex multi-tool turns, validate the behavior on a staging number before you route production traffic.

​
Session budgets

Every realtime session is bounded so a stuck provider can’t run forever: the handshake is given a few seconds before the runtime gives up and falls back, and sessions are closed at a 15-minute ceiling even if the provider hasn’t. Tune lifetime, idle re-engage, and hangup budgets on Session Lifecycle.

​
Next steps

Speech-to-speech model

The shared duplex-agent model, greeting behavior, and provider matrix.

OpenAI Realtime

The most complete realtime adapter — server VAD, function calling, greeting.

Turn detection

Tune barge-in, endpointing, and backchannel handling.

All runtime knobs

Models, codecs, budgets, and guardrails in one reference.