# --- Google AI Studio (Gemini Live API + research LLM) ---
# Get a key at https://aistudio.google.com/apikey
GOOGLE_API_KEY=your-google-ai-studio-key
# Use the AI Studio backend (not Vertex AI) for this demo.
GOOGLE_GENAI_USE_VERTEXAI=FALSE

# --- Tavily (web search used by the deep research agent) ---
# Get a key at https://app.tavily.com
TAVILY_API_KEY=your-tavily-key

# --- LangSmith tracing (optional but recommended) ---
# One unified trace per conversation: the voice session is the root span and the
# deep research agent nests inside the deep_research tool span. Get a key at
# https://smith.langchain.com
LANGSMITH_TRACING=true
LANGSMITH_API_KEY=your-langsmith-key
LANGSMITH_PROJECT=adk-realtime-deepagents-voice

# --- Optional overrides ---
# Live voice model. Latest native-audio model as of 2026.
# LIVE_MODEL=gemini-3.1-flash-live-preview
# Model the deep research agent reasons with.
# RESEARCH_MODEL=google_genai:gemini-2.5-flash
# When the model announces finished background research:
#   WHEN_IDLE (default) = wait for a natural pause, then announce
#   INTERRUPT           = speak up as soon as results land
#   SILENT              = fold into context only
# RESEARCH_SCHEDULING=WHEN_IDLE