app.py

# app.py
"""
AKIRA V16 — OLLAMA + FLASK (HF SPACE + DOCKER)
- Inicia ollama serve em subprocess
- Flask roda na porta 7860 (HF)
- Ollama roda em 11434 (interno)
- /api/chat → Akira com gírias de Luanda
"""
import subprocess
import time
import requests
import sys
from flask import Flask, request, jsonify
from loguru import logger

app = Flask(__name__)
OLLAMA_URL = "http://localhost:11434"

# === INICIA OLLAMA EM BACKGROUND ===
def start_ollama():
    logger.info("Iniciando ollama serve em background...")
    subprocess.Popen(
        ["ollama", "serve"],
        stdout=subprocess.DEVNULL,
        stderr=subprocess.DEVNULL
    )

# === VERIFICA OLLAMA ===
def check_ollama():
    try:
        resp = requests.get(f"{OLLAMA_URL}/api/tags", timeout=10)
        return resp.status_code == 200
    except:
        return False

# === AGUARDA OLLAMA ===
def wait_for_ollama():
    logger.info("Aguardando Ollama (11434)...")
    for i in range(60):
        if check_ollama():
            logger.success("Ollama Qwen 2.5 7B PRONTO!")
            return True
        time.sleep(2)
    logger.critical("Ollama não subiu!")
    return False

# === RAIZ ===
@app.route("/")
def index():
    return '''
    <div style="font-family: monospace; text-align: center; margin: 50px; background: #000; color: #0f0; padding: 40px;">
        <h1>AKIRA V16 OLLAMA</h1>
        <p><strong>Qwen 2.5 7B (4GB)</strong></p>
        <p><strong>ROTA:</strong> <code>POST /api/chat</code></p>
        <pre>
curl -X POST https://akra35567-ollama.hf.space/api/chat \\
  -d '{"mensagem": "Epá, Akira! Tud bem?"}'
        </pre>
        <p>Health: <a href="/health">/health</a></p>
        <p>Luanda, Angola — 11/11/2025 21:25 WAT</p>
    </div>
    ''', 200

# === HEALTHCHECK ===
@app.route("/health")
def health():
    if check_ollama():
        return jsonify({"status": "OK", "model": "qwen2.5:7b"}), 200
    return jsonify({"status": "INICIANDO..."}), 503

# === AKIRA CHAT ===
@app.route("/api/chat", methods=['POST'])
def chat():
    if not check_ollama():
        return jsonify({'resposta': 'Epá, tô acordando... espera!'}), 503

    data = request.get_json() or {}
    msg = data.get('mensagem', '').strip()
    if not msg:
        return jsonify({'error': 'mensagem obrigatória'}), 400

    prompt = f"""Você é Akira, angolana de Luanda, mulher, fala com gírias (epá, puto, kandando, musseque), direta, sem censura.

Usuário: {msg}
Akira:"""

    payload = {
        "model": "qwen2.5:7b",
        "prompt": prompt,
        "stream": False,
        "options": {
            "temperature": 0.85,
            "num_predict": 256
        }
    }

    try:
        resp = requests.post(f"{OLLAMA_URL}/api/generate", json=payload, timeout=30)
        if resp.status_code == 200:
            resposta = resp.json().get("response", "").strip()
            if not resposta:
                resposta = "Epá, tô aqui, puto!"
            return jsonify({'resposta': resposta})
    except Exception as e:
        return jsonify({'resposta': 'Epá, tô off!'}), 500

# === INÍCIO ===
if __name__ == "__main__":
    logger.remove()
    logger.add(sys.stderr, format="<green>{time:HH:mm:ss}</green> → <level>{message}</level>")
    logger.info("AKIRA V16 — Iniciando...")

    start_ollama()
    if not wait_for_ollama():
        sys.exit(1)

    logger.info("Flask rodando na porta 7860")
    app.run(host="0.0.0.0", port=7860, debug=False)