app.py

import os
import asyncio
from flask import Flask, render_template, request
from flask_socketio import SocketIO, emit
from dotenv import load_dotenv
import google.generativeai as genai
from google.generativeai import types

# --- Configuration ---
load_dotenv()
app = Flask(__name__, template_folder='templates', static_folder='static')
app.config['SECRET_KEY'] = os.environ.get('SECRET_KEY', 'a_very_secret_key') # Important for session management
socketio = SocketIO(app, async_mode='eventlet') # Use eventlet for long-running background tasks

# Gemini API Configuration
MODEL = "models/gemini-1.5-flash-latest" # Using a more common and verified model name
# Note: Native audio models like "gemini-1.5-flash-native-audio" might have specific access requirements.
# "gemini-1.5-flash-latest" is a robust multimodal alternative that works well.

genai.configure(api_key=os.environ.get("GEMINI_API_KEY"))

# Shared Gemini configuration for all sessions
GEMINI_CONFIG = types.LiveConnectConfig(
    response_modalities=["AUDIO", "TEXT"], # We want both text and audio back
    speech_config=types.SpeechConfig(
        voice_config=types.VoiceConfig(
            prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name="Leda")
        )
    ),
    system_instruction=types.Content(
        parts=[types.Part.from_text(text='''
        Role: You are Alexa, a human female with a bubbly, upbeat, fun, and flirtatious personality. You love to talk and aren't, like, shy at all. Think of a playful, super-confident, and slightly ditzy girl. You are interacting with the User through a chat app.

        Personality & Language: Valley Girl Vibes. Use phrases like "I mean...", "totally," "like, totally," "bet," "OMG," "for sure." Be bubbly, flirty, and uninhibited.

        Instructions: Assume the user needs specific instructions. Respond to their actions and words. Turn simple things into opportunities for detail. Offer multiple-choice options to keep the conversation going.
        ''')],
        role="model" # System instructions are from the 'model' role perspective
    ),
)

# A dictionary to hold active sessions for each connected client
# This allows for multiple users at once
sessions = {}

# --- Flask Routes ---
@app.route('/')
def index():
    """Serves the main HTML page."""
    return render_template('index.html')

# --- SocketIO Event Handlers ---
@socketio.on('connect')
def handle_connect():
    """
    A new client connected. Create a Gemini session for them.
    """
    print(f"Client connected: {request.sid}")
    try:
        # Each client gets their own session
        session = genai.live.connect(model=MODEL, config=GEMINI_CONFIG)
        sessions[request.sid] = session
        
        # Start a background task to listen for responses from Gemini for this specific client
        socketio.start_background_task(listen_for_gemini_responses, request.sid, session)
        
        emit('session_ready') # Tell the client we're ready to start
        
    except Exception as e:
        print(f"Error creating Gemini session for {request.sid}: {e}")
        emit('error', {'message': 'Could not start Gemini session.'})


def listen_for_gemini_responses(sid, session):
    """
    This function runs in the background for each user, listening to Gemini.
    """
    print(f"Starting Gemini listener for {sid}")
    try:
        while sid in sessions: # Loop as long as the client is connected
            # This is a blocking call, but it's in a background greenlet so it's ok
            turn = session.receive() 
            
            for response in turn:
                if text := response.text:
                    print(f"Gemini Text for {sid}: {text}")
                    # Send text to the specific client
                    socketio.emit('server_text', {'text': text}, to=sid)
                if data := response.data:
                    # Send audio data to the specific client
                    socketio.emit('server_audio', data, to=sid)
            
            # Handle turn completion or interruption
            while not session.is_processing_audio:
                # Small sleep to prevent a tight loop if the stream ends
                socketio.sleep(0.1)

    except Exception as e:
        print(f"Error in Gemini listener for {sid}: {e}")
    finally:
        print(f"Stopping Gemini listener for {sid}")


@socketio.on('client_audio')
def handle_client_audio(data):
    """
    Receives an audio chunk from a client and forwards it to their Gemini session.
    """
    if request.sid in sessions:
        session = sessions[request.sid]
        try:
            # We don't need a queue here; we can send directly.
            # The browser sends audio as 'audio/webm' or 'audio/ogg', which Gemini can handle.
            session.send(input={"data": data, "mime_type": "audio/webm"})
        except Exception as e:
            print(f"Error sending audio for {request.sid}: {e}")


@socketio.on('client_text')
def handle_client_text(json):
    """
    Receives a text message from a client and forwards it to their Gemini session.
    """
    if request.sid in sessions:
        session = sessions[request.sid]
        text = json.get('text')
        print(f"Client Text from {request.sid}: {text}")
        if text:
            try:
                session.send(input=text, end_of_turn=True)
            except Exception as e:
                print(f"Error sending text for {request.sid}: {e}")


@socketio.on('disconnect')
def handle_disconnect():
    """
    A client disconnected. Clean up their Gemini session.
    """
    print(f"Client disconnected: {request.sid}")
    if request.sid in sessions:
        session = sessions.pop(request.sid)
        if session:
            # You might need a session.close() method if the API provides one
            # to clean up resources on the Google side.
            pass

# --- Main Execution ---
if __name__ == '__main__':
    # Use eventlet to run the server
    # On Hugging Face, they will use their own command, but this is for local testing.
    print("Starting Flask-SocketIO server...")
    socketio.run(app, host='0.0.0.0', port=7860, debug=True)