Update app.py

app.py

@@ -1,34 +1,220 @@
-import streamlit as st
-from openai import OpenAI
-
-# 读取 Hugging Face Secrets（推荐方式）
-client = OpenAI(api_key=st.secrets["OPENAI_API_KEY"])
-
-# 页面配置
-st.set_page_config(page_title="GPT‑5 Agent", page_icon="🤖", layout="wide")
-st.title("🤖 GPT‑5 智能助手")
-st.write("使用 OpenAI GPT‑5 的 Streamlit 聊天应用。")
-
-# 聊天历史记录
-if "chat_history" not in st.session_state:
-    st.session_state.chat_history = []
-
-# 聊天输入框
-user_input = st.chat_input("请输入你的问题...")
-
-if user_input:
-    st.session_state.chat_history.append({"role": "user", "content": user_input})
-    with st.spinner("GPT‑5 正在思考..."):
-        response = client.chat.completions.create(
-            model="gpt-4o-mini",  # 👈 可替换为你想使用的具体 GPT‑5 模型
-            messages=st.session_state.chat_history,
-        )
-        ai_reply = response.choices[0].message.content
-        st.session_state.chat_history.append({"role": "assistant", "content": ai_reply})
-
-# 渲染聊天记录
-for msg in st.session_state.chat_history:
-    if msg["role"] == "user":
-        st.chat_message("user").write(msg["content"])
-    else:
-        st.chat_message("assistant").write(msg["content"])
+# ai_blog_assistant.py
+from typing import TypedDict, Annotated, List
+from langgraph.graph import StateGraph, END
+from langgraph.graph.message import add_messages
+from llm_selector import LLMSelector
+import json
+
+class BlogState(TypedDict):
+    messages: Annotated[list, add_messages]
+    topic: str
+    keywords: List[str]
+    outline: str
+    research: str
+    draft: str
+    final: str
+    seo_score: int
+
+# 使用最快的免费模型
+llm = LLMSelector.get_groq(model="llama-3.1-8b-instant")  # 最快
+quality_llm = LLMSelector.get_groq(model="llama-3.1-70b-versatile")  # 质量
+
+def keyword_analyzer(state: BlogState) -> BlogState:
+    """Agent 1: 关键词分析"""
+    prompt = f"""你是SEO专家。
+
+主题: {state['topic']}
+
+任务: 提取5-8个核心关键词
+要求: 考虑搜索量和竞争度
+
+只输出关键词列表（JSON格式）：
+{{"keywords": ["关键词1", "关键词2", ...]}}"""
+    
+    response = llm.invoke(prompt)
+    
+    try:
+        result = json.loads(response.content)
+        state['keywords'] = result['keywords']
+    except:
+        state['keywords'] = ["人工智能", "技术", "应用"]
+    
+    print(f"✅ [关键词分析] {state['keywords']}")
+    return state
+
+def outline_creator(state: BlogState) -> BlogState:
+    """Agent 2: 创建大纲"""
+    prompt = f"""你是内容策划师。
+
+主题: {state['topic']}
+关键词: {', '.join(state['keywords'])}
+
+任务: 创建文章大纲
+要求:
+1. 3-5个主要章节
+2. 每章节2-3个小点
+3. 逻辑清晰
+
+输出格式:
+# 标题
+## 章节1
+- 要点1
+- 要点2
+...
+
+直接给出大纲："""
+    
+    response = llm.invoke(prompt)
+    state['outline'] = response.content
+    print(f"✅ [大纲创建] 完成")
+    return state
+
+def researcher(state: BlogState) -> BlogState:
+    """Agent 3: 深度研究"""
+    prompt = f"""你是资深研究员。
+
+主题: {state['topic']}
+大纲: {state['outline']}
+
+任务: 为每个章节收集资料
+要求:
+1. 数据支撑
+2. 案例分析
+3. 最新趋势
+
+输出研究报告："""
+    
+    response = quality_llm.invoke(prompt)  # 用更好的模型
+    state['research'] = response.content
+    print(f"✅ [研究] 完成")
+    return state
+
+def writer(state: BlogState) -> BlogState:
+    """Agent 4: 撰写文章"""
+    prompt = f"""你是专业博客作家。
+
+主题: {state['topic']}
+关键词: {', '.join(state['keywords'])}
+大纲: {state['outline']}
+研究资料: {state['research']}
+
+任务: 撰写1500字博客文章
+要求:
+1. 标题吸引人（包含主关键词）
+2. 自然融入关键词
+3. 段落简短（2-3句）
+4. 使用项目符号
+5. 包含行动号召(CTA)
+
+撰写文章："""
+    
+    response = quality_llm.invoke(prompt)
+    state['draft'] = response.content
+    print(f"✅ [写作] 完成")
+    return state
+
+def seo_optimizer(state: BlogState) -> BlogState:
+    """Agent 5: SEO优化"""
+    prompt = f"""你是SEO优化专家。
+
+文章:
+{state['draft']}
+
+关键词: {', '.join(state['keywords'])}
+
+任务: 优化SEO
+1. 检查关键词密度 (2-3%)
+2. 优化标题和副标题
+3. 添加meta描述
+4. 改进可读性
+
+输出:
+1. 优化后的文章
+2. SEO评分 (0-100)
+3. 改进建议
+
+格式:
+=== 优化文章 ===
+[文章内容]
+
+=== SEO评分 ===
+评分: XX分
+建议: ...
+
+给出结果："""
+    
+    response = quality_llm.invoke(prompt)
+    content = response.content
+    
+    # 提取评分
+    if "评分:" in content:
+        try:
+            score_line = [line for line in content.split('\n') if '评分:' in line][0]
+            score = int(''.join(filter(str.isdigit, score_line)))
+            state['seo_score'] = min(score, 100)
+        except:
+            state['seo_score'] = 75
+    
+    state['final'] = content
+    print(f"✅ [SEO优化] 评分: {state['seo_score']}")
+    return state
+
+# 构建工作流
+def create_blog_workflow():
+    workflow = StateGraph(BlogState)
+    
+    workflow.add_node("keywords", keyword_analyzer)
+    workflow.add_node("outline", outline_creator)
+    workflow.add_node("research", researcher)
+    workflow.add_node("write", writer)
+    workflow.add_node("seo", seo_optimizer)
+    
+    workflow.set_entry_point("keywords")
+    workflow.add_edge("keywords", "outline")
+    workflow.add_edge("outline", "research")
+    workflow.add_edge("research", "write")
+    workflow.add_edge("write", "seo")
+    workflow.add_edge("seo", END)
+    
+    return workflow.compile()
+
+# 运行
+def generate_blog(topic: str):
+    app = create_blog_workflow()
+    
+    initial_state = {
+        "messages": [],
+        "topic": topic,
+        "keywords": [],
+        "outline": "",
+        "research": "",
+        "draft": "",
+        "final": "",
+        "seo_score": 0
+    }
+    
+    print(f"\n{'='*70}")
+    print(f"🚀 开始生成博客: {topic}")
+    print(f"{'='*70}\n")
+    
+    result = app.invoke(initial_state)
+    
+    print(f"\n{'='*70}")
+    print(f"📝 最终文章 (SEO评分: {result['seo_score']})")
+    print(f"{'='*70}\n")
+    print(result['final'])
+    
+    # 保存文章
+    filename = f"blog_{topic[:20].replace(' ', '_')}.md"
+    with open(filename, 'w', encoding='utf-8') as f:
+        f.write(f"# {topic}\n\n")
+        f.write(f"**关键词**: {', '.join(result['keywords'])}\n\n")
+        f.write(f"**SEO评分**: {result['seo_score']}/100\n\n")
+        f.write("---\n\n")
+        f.write(result['final'])
+    
+    print(f"\n💾 文章已保存到: {filename}")
+    return result
+
+if __name__ == "__main__":
+    generate_blog("ChatGPT在企业中的5个创新应用")