`GeminiLiveAgentAdapter`

Connects to Google's Gemini Live API over WebSocket. Native VAD and barge-in are built-in.

Constructor

python

import os
import scenario
 
adapter = scenario.GeminiLiveAgentAdapter(
    model="gemini-2.5-flash-native-audio-latest",
    voice="Algieba",
    system_instruction="You are a helpful assistant.",
    api_key=os.environ.get("GEMINI_API_KEY"),
)

If api_key / apiKey is not passed, it falls back to GEMINI_API_KEY (and, in TypeScript, GOOGLE_API_KEY) in the environment.

Capabilities

streaming_transcripts	native_vad	dtmf	interruption	input_formats	output_formats
✅	✅	❌	✅	`pcm16/16000`	`pcm16/24000`

Worked examples

Python: gemini_live.py — basic Gemini Live scenario; gemini_live_interruption.py — barge-in handling.
TypeScript: gemini-live.test.ts — Gemini Live native-audio happy path.

Known-fixed quirks

Truncated agent reply on turn 2+ — fixed in commit 760a464 (PR #355). See Troubleshooting → Historical fixes.