examples/voice_agent_webrtc_langgraph/docker-compose.yml

name: voice-agents-webrtc-langgraph

services:
  python-app:
    build:
      context: ../../
      dockerfile: examples/voice_agent_webrtc_langgraph/Dockerfile
    ports:
      - "9000:9000"   # UI static server
      - "7860:7860"   # Pipeline API/WebSocket
      - "2024:2024"   # LangGraph dev (optional external access)
    volumes:
      - ./audio_dumps:/app/examples/voice_agent_webrtc_langgraph/audio_dumps
      # - /home/fciannella/src/ace-controller/examples/voice_agent_webrtc_langgraph:/app/examples/voice_agent_webrtc_langgraph/audio_prompt.wav
      # - /home/fciannella/src/ace-controller-langgraph-agents:/langgraph-agents
    env_file:
      - .env
    environment:
      - NVIDIA_API_KEY=${NVIDIA_API_KEY}
      - USE_LANGGRAPH=${USE_LANGGRAPH}
      - LANGGRAPH_BASE_URL=${LANGGRAPH_BASE_URL}
      - LANGGRAPH_ASSISTANT=${LANGGRAPH_ASSISTANT}
      - USER_EMAIL=${USER_EMAIL}
      - LANGGRAPH_STREAM_MODE=${LANGGRAPH_STREAM_MODE}
      - LANGGRAPH_DEBUG_STREAM=${LANGGRAPH_DEBUG_STREAM}
      - LANGGRAPH_PORT=${LANGGRAPH_PORT}

      - RIVA_ASR_LANGUAGE=en-US
      - RIVA_TTS_LANGUAGE=en-US
      - RIVA_TTS_VOICE_ID=Magpie-ZeroShot.Female-1
      - ZERO_SHOT_AUDIO_PROMPT=/app/examples/voice_agent_webrtc_langgraph/audio_prompt.wav # set this only if using a zero-shot TTS model with a custom audio prompt
      - ENABLE_SPECULATIVE_SPEECH=true # set to false to disable speculative speech processing

    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "curl -f http://localhost:7860/get_prompt || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 60s
    logging:
      driver: "json-file"
      options:
        max-size: "50m"
        max-file: "5"

volumes:
  nim_cache:
  riva_data: