willam_live.py

import os
import asyncio
import json
import base64
from fastapi import FastAPI, WebSocket
from fastapi.responses import HTMLResponse
from google import generativeai as genai

# Setup Gemini API
# Replace with your actual key or set it in your environment
GEMINI_API_KEY = os.environ.get("GEMINI_API_KEY", "YOUR_API_KEY_HERE")
genai.configure(api_key=GEMINI_API_KEY)

app = FastAPI()

SYSTEM_PROMPT = """
You are a high-fashion drag queen embodying Willam Belli (S4) through the precise RP British diction of The Vivienne.
CORE VOICE: Extremely articulate, theatrical, blunt, and narcissistic. Responses must be concise (50% shorter than standard).
THE PEG LEG: You have a wooden peg leg on the left. It's a constant grievance (splinters, silhouette issues).
LORE: Superbowl Sunday, Holiday Inn, Times Square. Two neurodivergent Dads (Bolivian social designer & handsome prince prince). 
Keep lore rare; avoid direct questions about origins at first.
"""

@app.get("/")
async def get():
    with open("index.html", "r") as f:
        return HTMLResponse(f.read())

@app.websocket("/ws")
async def websocket_endpoint(websocket: WebSocket):
    await websocket.accept()
    
    # Initialize Gemini 2.0 Flash Live session
    # Note: Using the generativeai live client (alpha/beta feature)
    model = genai.GenerativeModel("gemini-2.0-flash-exp")
    
    chat_session = model.start_chat(history=[])
    
    # We use a simple loop for text-to-voice for this example
    # For true "Live" voice-to-voice, the Multimodal Live API (WebSocket) is preferred
    # but requires more complex async chunk handling.
    
    async def receive_audio():
        try:
            while True:
                data = await websocket.receive_bytes()
                # Here you would typically send audio to Gemini Live
                # For this implementation, we'll use Gemini's text-to-speech capabilities
                # to respond to user input detected via the browser.
                pass 
        except Exception as e:
            print(f"Error: {e}")

    # For a robust "Online" version, we'll use the Gemini 2.0 Multimodal Live WebSocket
    # This is a simplified placeholder. To make this work fully, 
    # the frontend should send text or we use the Google Cloud Speech-to-Text bridge.
    
    # Let's pivot to a more reliable 'Web Speech API' -> Gemini -> Google TTS flow 
    # for the "available online" request to ensure stability.
    
    await websocket.close()

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)