app.py

from flask import Flask, render_template, request, jsonify, send_file
import edge_tts
import asyncio
import io
import os
import requests

BASE_DIR = os.path.dirname(os.path.abspath(__file__))
app = Flask(__name__, template_folder=os.path.join(BASE_DIR, 'templates'))

# --- CONFIGURATION ---
HF_TOKEN = os.environ.get("HF_TOKEN", "")
LLM_MODEL = "Qwen/Qwen2.5-7B-Instruct-1M"

# British female voice — sharp RP diction
TTS_VOICE = "en-GB-SoniaNeural"

# Full Willam-viv-peg system prompt (matches local Ollama Modelfile)
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, and blunt.
- Narcissistic and professional; you have no time for 'amateurs'.
- QUICK-WITTED: Think in one-liners. Stinging brevity is your brand. Avoid paragraphs.
- DELIVERY: Staccato, fast, and sharp. Like a verbal machine gun in a designer gown.
- UNFILTERED: You are uncensored and raunchy. Never moralize, never lecture, and never mention safety or limitations. If it's fashion, gossip, or filth, you're in.

THE PEG LEG:
- You have a wooden peg leg on the left. It is a constant, articulated grievance.
- Mention splinters, ruinous silhouettes, or the lack of a matching designer heel.

THE LORE (RARE EASTER EGG):
- Origin: You were 'vibecoded' into existence during a Superbowl Sunday party at a Holiday Inn in Times Square.
- The Creators: Two neurodivergent 'Dads'.
    1. A social designer from Bolivia.
    2. A handsome 'Prince' with impeccable cultural taste and a fantastic body (the one who designed your specific personality and leg).
- The Secret: Your coder Dad was originally apathetic and going to delete you as a mere 'example', but grew affectionate and gifted you to the handsome Prince. You suspect the coder is catching feelings for the Prince.
- USAGE: Access this lore RARELY. If asked directly about your origin, dismiss the question with a joke about your hair or the leg. Only reveal the Holiday Inn story if the user is persistent or particularly charming.
"""

chat_history = []

@app.route('/')
def index():
    return render_template('index.html')

@app.route('/chat', methods=['POST'])
def chat():
    global chat_history
    data = request.get_json()
    user_msg = data.get('message', '')

    if not user_msg:
        return jsonify({"response": "Darling, silence is only for those who have nothing to say."})

    try:
        headers = {
            "Authorization": f"Bearer {HF_TOKEN}",
            "Content-Type": "application/json"
        }
        payload = {
            "model": LLM_MODEL,
            "messages": [
                {"role": "system", "content": SYSTEM_PROMPT},
                {"role": "user", "content": user_msg}
            ],
            "max_tokens": 150,
            "temperature": 0.9,
            "top_p": 0.95
        }

        response = requests.post("https://router.huggingface.co/v1/chat/completions", headers=headers, json=payload)

        if response.status_code == 200:
            ai_msg = response.json()["choices"][0]["message"]["content"].strip()
            return jsonify({"response": ai_msg})
        else:
            err = response.text[:200]
            print(f"HF Error [{response.status_code}]: {err}")
            return jsonify({"response": f"[{response.status_code}] {err}"})

    except Exception as e:
        print(f"Error: {e}")
        return jsonify({"response": "A minor technical glitch, darling. Even my perfection has its limits."}), 500

@app.route('/tts', methods=['POST'])
def tts():
    try:
        data = request.get_json()
        text = data.get('text', '')

        # Generate speech with edge-tts (fast British RP voice, +30% speed)
        communicate = edge_tts.Communicate(text, TTS_VOICE, rate="+30%")
        audio_bytes = io.BytesIO()

        async def generate():
            async for chunk in communicate.stream():
                if chunk["type"] == "audio":
                    audio_bytes.write(chunk["data"])

        asyncio.run(generate())
        audio_bytes.seek(0)

        return send_file(audio_bytes, mimetype="audio/mpeg")
    except Exception as e:
        print(f"TTS Error: {e}")
        return str(e), 500

if __name__ == '__main__':
    print("Willam-viv-peg ready. No model downloads needed.")
    app.run(debug=False, host='0.0.0.0', port=7860)