app.py

# ================================
# ChatVLD - Hugging Face com FAISS Cache (Corrigido, sem download)
# ================================

import os
from pathlib import Path
import gradio as gr

from langchain_groq import ChatGroq
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.messages import AIMessage, HumanMessage
from langchain_community.document_loaders import PyMuPDFLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_huggingface import HuggingFaceEmbeddings

# ================================
# 1. MODELO GROQ
# ================================
groq_api_key = os.environ.get("GROQ_API_KEY")
if not groq_api_key:
    raise ValueError("❌ Defina a variável de ambiente GROQ_API_KEY nos Secrets do Space.")

model = ChatGroq(model="llama3-70b-8192", temperature=0)

# ================================
# 2. EMBEDDINGS HUGGINGFACE
# ================================
embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")

# ================================
# 3. CARREGAR PDFS LOCAIS
# ================================
pdf_files = ["documento.pdf", "documento2.pdf"]

# Verificar se os PDFs existem
for pdf in pdf_files:
    if not Path(pdf).exists():
        raise FileNotFoundError(f"❌ O arquivo {pdf} não foi encontrado na pasta do Space. "
                                "Suba o PDF manualmente.")

# ================================
# 4. INDEXAÇÃO FAISS (CACHE)
# ================================
index_path = Path("index_faiss")
if index_path.exists():
    print("✅ Carregando índice FAISS do cache...")
    vectorstore = FAISS.load_local(index_path, embeddings, allow_dangerous_deserialization=True)
else:
    print("⏳ Criando índice FAISS com múltiplos PDFs...")
    docs = []
    for pdf in pdf_files:
        loader = PyMuPDFLoader(pdf)
        docs.extend(loader.load())

    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
    texts = text_splitter.split_documents(docs)

    vectorstore = FAISS.from_documents(texts, embeddings)
    vectorstore.save_local(index_path)
    print("✅ Índice FAISS salvo.")

retriever = vectorstore.as_retriever()

# ================================
# 5. PROMPT TEMPLATE
# ================================
prompt = ChatPromptTemplate.from_messages([
    ("system", "Você é um assistente especializado nos documentos fornecidos."),
    MessagesPlaceholder(variable_name="chat_history"),
    ("human", "{input}"),
    ("system", "Se necessário, use também este contexto: {context}")
])

# ================================
# 6. HISTÓRICO
# ================================
chat_history = []

def responder(message, history):
    context_docs = retriever.get_relevant_documents(message)
    context_text = "\n".join([doc.page_content for doc in context_docs[:3]])

    chain = prompt | model
    response = chain.invoke({
        "input": message,
        "chat_history": chat_history,
        "context": context_text
    })

    chat_history.append(HumanMessage(content=message))
    chat_history.append(AIMessage(content=response.content))

    return response.content

# ================================
# 7. CSS PERSONALIZADO
# ================================
custom_css = """
#chatbot_box {
    background-color: white !important; /* Caixa do chatbot na cor original */
    color: black !important;
    border-radius: 12px;
    padding: 10px;
}

#input_box {
    background-color: #1a1a1a !important; /* Preto leve futurista */
    color: white !important;
    border-radius: 12px;
    padding: 10px;
}

button {
    background: linear-gradient(45deg, #4A00E0, #8E2DE2) !important; /* Azul com roxo */
    color: white !important;
    border-radius: 8px !important;
}
"""

# ================================
# 8. INTERFACE GRADIO
# ================================
with gr.Blocks(css=custom_css, theme="soft") as iface:
    with gr.Tabs():
        with gr.TabItem("💬 ChatVLD Futurista"):
            chat_interface = gr.ChatInterface(
                fn=responder,
                title="🤖 ChatVLD Futurista",
                description="Olá! Sou o suporte da impressora Codeline SS5632. Em que posso ajudar?",
                chatbot=gr.Chatbot(elem_id="chatbot_box"),
                textbox=gr.Textbox(placeholder="Digite sua mensagem...", elem_id="input_box")
            )

        with gr.TabItem("🔗 VALID NODE"):
            gr.Markdown("Clique no botão abaixo para acessar o VALID NODE:")
            gr.Markdown("[Ir para VALID NODE](https://172.17.200.97)")

iface.launch()