app.py

import openai
import gradio as gr
import fitz  # PyMuPDF
from openai import OpenAI
import traceback

# 全域變數
api_key = ""
selected_model = "gpt-4"
summary_text = ""
client = None
pdf_text = ""

def set_api_key(user_api_key):
    """設定 OpenAI API Key 並初始化客戶端"""
    global api_key, client
    try:
        api_key = user_api_key.strip()
        if not api_key:
            return "❌ API Key 不能為空"

        client = OpenAI(api_key=api_key)

        # 測試 API Key 是否有效
        client.chat.completions.create(
            model="gpt-4",
            messages=[{"role": "user", "content": "你好"}],
            max_tokens=5
        )

        return "✅ API Key 已設定並驗證成功"
    except Exception as e:
        return f"❌ API Key 設定失敗: {str(e)}"

def set_model(model_name):
    """設定選擇的模型"""
    global selected_model
    selected_model = model_name
    return f"✅ 模型已選擇：{model_name}"

def extract_pdf_text(file_path):
    """從 PDF 文件中提取文字"""
    try:
        doc = fitz.open(file_path)
        text = ""
        for page_num, page in enumerate(doc):
            page_text = page.get_text()
            if page_text.strip():
                text += f"\n--- 第 {page_num + 1} 頁 ---\n{page_text}"
        doc.close()
        return text
    except Exception as e:
        return f"❌ PDF 解析錯誤: {str(e)}"

def generate_summary(pdf_file):
    """從 PDF 內容生成摘要"""
    global summary_text, pdf_text

    if not client:
        return "❌ 請先設定 OpenAI API Key"

    if not pdf_file:
        return "❌ 請先上傳 PDF 文件"

    try:
        pdf_text = extract_pdf_text(pdf_file.name)
        if not pdf_text.strip():
            return "⚠️ 無法解析 PDF 文字，可能為純圖片 PDF 或空白文件。"

        pdf_text_truncated = pdf_text[:8000]  # 簡單限制長度
        response = client.chat.completions.create(
            model=selected_model,
            messages=[
                {"role": "system", "content": "請將以下 PDF 內容整理為條列式摘要，並用繁體中文回答："},
                {"role": "user", "content": pdf_text_truncated}
            ],
            temperature=0.3
        )

        summary_text = response.choices[0].message.content
        return summary_text

    except Exception as e:
        print(f"錯誤詳情: {traceback.format_exc()}")
        return f"❌ 摘要生成失敗: {str(e)}"

def ask_question(user_question):
    """基於 PDF 內容回答問題"""
    if not client:
        return "❌ 請先設定 OpenAI API Key"

    if not summary_text and not pdf_text:
        return "❌ 請先生成 PDF 摘要"

    if not user_question.strip():
        return "❌ 請輸入問題"

    try:
        context = f"PDF 摘要:\n{summary_text}\n\n原始內容（部分）:\n{pdf_text[:2000]}"
        response = client.chat.completions.create(
            model=selected_model,
            messages=[
                {"role": "system", "content": f"根據以下 PDF 內容回答用戶的問題，請用繁體中文回答並客觀精準：\n{context}"},
                {"role": "user", "content": user_question}
            ],
            temperature=0.2
        )
        return response.choices[0].message.content

    except Exception as e:
        print(f"錯誤詳情: {traceback.format_exc()}")
        return f"❌ 問答生成失敗: {str(e)}"

def clear_all():
    """清除所有資料"""
    global summary_text, pdf_text
    summary_text = ""
    pdf_text = ""
    return "", "", ""

# Gradio 介面
with gr.Blocks(title="PDF 摘要助手") as demo:
    gr.Markdown("# 📄 PDF 摘要 & 問答助手 (修正版)")

    with gr.Tab("🔧 設定"):
        api_key_input = gr.Textbox(label="🔑 輸入 OpenAI API Key", type="password")
        api_key_status = gr.Textbox(label="狀態", interactive=False, value="等待設定 API Key...")
        api_key_btn = gr.Button("確認設定 API Key")
        api_key_btn.click(set_api_key, inputs=api_key_input, outputs=api_key_status)

        model_choice = gr.Radio(["gpt-4", "gpt-4.1", "gpt-4.5"], label="選擇 AI 模型", value="gpt-4")
        model_status = gr.Textbox(label="模型狀態", interactive=False, value="✅ 已選擇：gpt-4")
        model_choice.change(set_model, inputs=model_choice, outputs=model_status)

    with gr.Tab("📄 PDF 摘要"):
        pdf_upload = gr.File(label="選擇 PDF 文件", file_types=[".pdf"])
        summary_btn = gr.Button("生成摘要")
        summary_output = gr.Textbox(label="PDF 摘要", lines=10)
        summary_btn.click(generate_summary, inputs=pdf_upload, outputs=summary_output)

    with gr.Tab("❓ 問答"):
        question_input = gr.Textbox(label="請輸入您的問題", lines=2)
        question_btn = gr.Button("送出問題")
        answer_output = gr.Textbox(label="AI 回答", lines=10)
        question_btn.click(ask_question, inputs=question_input, outputs=answer_output)
        question_input.submit(ask_question, inputs=question_input, outputs=answer_output)

    clear_btn = gr.Button("🗑️ 清除所有資料")
    clear_btn.click(clear_all, outputs=[summary_output, question_input, answer_output])

if __name__ == "__main__":
    demo.launch(show_error=True)