Update app.py

app.py

@@ -1,24 +1,19 @@
 """
-Multi-Agent RAG-Enhanced LLM System
+Multi-Agent RAG-Enhanced LLM System for Hugging Face Spaces
 감독자(Supervisor) -> 창의성 생성자(Creative) -> 비평자(Critic) -> 감독자(Final) 
 4단계 파이프라인을 통한 고품질 답변 생성 시스템
 """
 
 import os
 import json
-import asyncio
 import time
 from typing import Optional, List, Dict, Any, Tuple
-from contextlib import asynccontextmanager
 from datetime import datetime
 from enum import Enum
 
 import requests
-import uvicorn
-from fastapi import FastAPI, HTTPException
-from fastapi.middleware.cors import CORSMiddleware
-from pydantic import BaseModel, Field
 import gradio as gr
+from pydantic import BaseModel, Field
 from dotenv import load_dotenv
 
 # 환경변수 로드
@@ -42,16 +37,6 @@ class Message(BaseModel):
     content: str
 
 
-class ChatRequest(BaseModel):
-    messages: List[Message]
-    model: str = "accounts/fireworks/models/qwen3-235b-a22b-instruct-2507"
-    max_tokens: int = Field(default=4096, ge=1, le=8192)
-    temperature: float = Field(default=0.6, ge=0, le=2)
-    top_p: float = Field(default=1.0, ge=0, le=1)
-    top_k: int = Field(default=40, ge=1, le=100)
-    use_search: bool = Field(default=True)
-
-
 class AgentResponse(BaseModel):
     role: AgentRole
     content: str
@@ -285,7 +270,7 @@ URL: {result.get('url', 'N/A')}
         
         return "\n".join(formatted)
     
-    async def process_with_agents(
+    def process_with_agents(
         self, 
         query: str, 
         search_results: List[Dict],
@@ -419,27 +404,39 @@ URL: {result.get('url', 'N/A')}
 # Gradio UI
 # ============================================================================
 
-def create_gradio_interface(multi_agent_system: MultiAgentSystem, search_client: BraveSearchClient):
+def create_gradio_interface():
     """Gradio 인터페이스 생성"""
     
-    async def process_query(
+    # 클라이언트 초기화
+    try:
+        llm_client = FireworksClient()
+        search_client = BraveSearchClient()
+        multi_agent_system = MultiAgentSystem(llm_client, search_client)
+        system_ready = True
+    except Exception as e:
+        print(f"⚠️ System initialization error: {e}")
+        multi_agent_system = None
+        search_client = None
+        system_ready = False
+    
+    def process_query(
         message: str,
-        history: List[List[str]],
+        history: List[Dict],
         use_search: bool,
         show_agent_thoughts: bool,
         search_count: int,
         temperature: float,
         max_tokens: int
-    ):
+    ) -> Tuple[List[Dict], str, str]:
         """쿼리 처리 함수"""
         
-        if not message:
-            return "", history, "", ""
+        if not message or not system_ready:
+            return history, "", ""
         
         try:
             # 검색 수행
             search_results = []
-            if use_search and search_client.api_key:
+            if use_search and search_client and search_client.api_key:
                 search_results = search_client.search(message, count=search_count)
             
             # 설정
@@ -449,7 +446,7 @@ def create_gradio_interface(multi_agent_system: MultiAgentSystem, search_client:
             }
             
             # 멀티 에이전트 처리
-            response = await multi_agent_system.process_with_agents(
+            response = multi_agent_system.process_with_agents(
                 query=message,
                 search_results=search_results,
                 config=config
@@ -490,15 +487,17 @@ def create_gradio_interface(multi_agent_system: MultiAgentSystem, search_client:
             final_answer = response.final_answer
             final_answer += f"\n\n---\n⏱️ *처리 시간: {response.processing_time:.2f}초*"
             
-            # 히스토리 업데이트
-            history.append([message, final_answer])
+            # 히스토리 업데이트 (OpenAI 형식)
+            history.append({"role": "user", "content": message})
+            history.append({"role": "assistant", "content": final_answer})
             
-            return "", history, agent_thoughts, search_display
+            return history, agent_thoughts, search_display
             
         except Exception as e:
             error_msg = f"❌ 오류 발생: {str(e)}"
-            history.append([message, error_msg])
-            return "", history, "", ""
+            history.append({"role": "user", "content": message})
+            history.append({"role": "assistant", "content": error_msg})
+            return history, "", ""
     
     # Gradio 인터페이스
     with gr.Blocks(
@@ -509,9 +508,6 @@ def create_gradio_interface(multi_agent_system: MultiAgentSystem, search_client:
             max-width: 1400px !important;
             margin: auto !important;
         }
-        #chatbot {
-            height: 600px !important;
-        }
         """
     ) as demo:
         gr.Markdown("""
@@ -521,13 +517,20 @@ def create_gradio_interface(multi_agent_system: MultiAgentSystem, search_client:
         **처리 과정:** 감독자(구조화) → 창의성 생성자(창의적 답변) → 비평자(검증) → 최종 감독자(종합)
         """)
         
+        if not system_ready:
+            gr.Markdown("""
+            ⚠️ **시스템 초기화 실패**: API 키를 확인해주세요.
+            - FIREWORKS_API_KEY 필요
+            - BRAVE_SEARCH_API_KEY (선택사항)
+            """)
+        
         with gr.Row():
             # 메인 채팅 영역
             with gr.Column(scale=3):
                 chatbot = gr.Chatbot(
                     height=500,
                     label="💬 대화",
-                    elem_id="chatbot"
+                    type="messages"  # OpenAI 스타일 메시지 형식
                 )
                 
                 msg = gr.Textbox(
@@ -614,18 +617,26 @@ def create_gradio_interface(multi_agent_system: MultiAgentSystem, search_client:
             process_query,
             inputs=[msg, chatbot, use_search, show_agent_thoughts, 
                    search_count, temperature, max_tokens],
-            outputs=[msg, chatbot, agent_thoughts, search_sources]
+            outputs=[chatbot, agent_thoughts, search_sources]
+        ).then(
+            lambda: "",
+            None,
+            msg
         )
         
         msg.submit(
             process_query,
             inputs=[msg, chatbot, use_search, show_agent_thoughts,
                    search_count, temperature, max_tokens],
-            outputs=[msg, chatbot, agent_thoughts, search_sources]
+            outputs=[chatbot, agent_thoughts, search_sources]
+        ).then(
+            lambda: "",
+            None,
+            msg
         )
         
         clear.click(
-            lambda: (None, None, None),
+            lambda: ([], None, None),
             None,
             [chatbot, agent_thoughts, search_sources]
         )
@@ -633,113 +644,6 @@ def create_gradio_interface(multi_agent_system: MultiAgentSystem, search_client:
     return demo
 
 
-# ============================================================================
-# FastAPI 앱
-# ============================================================================
-
-@asynccontextmanager
-async def lifespan(app: FastAPI):
-    """앱 생명주기 관리"""
-    port = int(os.getenv("PORT", 7860))
-    print("\n" + "="*60)
-    print("🚀 Multi-Agent RAG System Starting...")
-    print(f"📍 Port: {port}")
-    print("="*60)
-    yield
-    print("\n👋 Shutting down...")
-
-
-app = FastAPI(
-    title="Multi-Agent RAG System API",
-    description="4-Stage Agent Collaboration System with RAG",
-    version="3.0.0",
-    lifespan=lifespan
-)
-
-# CORS 설정
-app.add_middleware(
-    CORSMiddleware,
-    allow_origins=["*"],
-    allow_credentials=True,
-    allow_methods=["*"],
-    allow_headers=["*"]
-)
-
-# 클라이언트 초기화
-try:
-    llm_client = FireworksClient()
-    search_client = BraveSearchClient()
-    multi_agent_system = MultiAgentSystem(llm_client, search_client)
-except Exception as e:
-    print(f"⚠️ Initialization error: {e}")
-    llm_client = None
-    search_client = None
-    multi_agent_system = None
-
-
-# API 엔드포인트
-@app.get("/")
-async def root():
-    """루트 엔드포인트"""
-    port = int(os.getenv("PORT", 7860))
-    return {
-        "name": "Multi-Agent RAG System",
-        "version": "3.0.0",
-        "status": "running",
-        "ui": f"http://localhost:{port}/ui",
-        "docs": f"http://localhost:{port}/docs"
-    }
-
-
-@app.post("/api/chat")
-async def chat_endpoint(request: ChatRequest):
-    """멀티 에이전트 채팅 API"""
-    if not multi_agent_system:
-        raise HTTPException(status_code=500, detail="System not initialized")
-    
-    try:
-        # 검색 수행
-        search_results = []
-        if request.use_search and search_client.api_key:
-            last_message = request.messages[-1].content if request.messages else ""
-            search_results = search_client.search(last_message, count=5)
-        
-        # 멀티 에이전트 처리
-        response = await multi_agent_system.process_with_agents(
-            query=request.messages[-1].content,
-            search_results=search_results,
-            config={
-                "temperature": request.temperature,
-                "max_tokens": request.max_tokens
-            }
-        )
-        
-        return response
-        
-    except Exception as e:
-        raise HTTPException(status_code=500, detail=str(e))
-
-
-@app.get("/health")
-async def health_check():
-    """헬스 체크"""
-    return {
-        "status": "healthy",
-        "timestamp": datetime.now().isoformat(),
-        "services": {
-            "llm": "ready" if llm_client else "not configured",
-            "search": "ready" if search_client and search_client.api_key else "not configured",
-            "multi_agent": "ready" if multi_agent_system else "not configured"
-        }
-    }
-
-
-# Gradio 마운트
-if multi_agent_system:
-    gradio_app = create_gradio_interface(multi_agent_system, search_client)
-    app = gr.mount_gradio_app(app, gradio_app, path="/ui")
-
-
 # ============================================================================
 # 메인 실행
 # ============================================================================
@@ -757,53 +661,21 @@ if __name__ == "__main__":
     # API 키 확인
     if not os.getenv("FIREWORKS_API_KEY"):
         print("\n⚠️  FIREWORKS_API_KEY가 설정되지 않았습니다.")
-        key = input("Fireworks API Key 입력: ").strip()
-        if key:
-            os.environ["FIREWORKS_API_KEY"] = key
-            llm_client = FireworksClient(key)
+        print("Hugging Face Spaces Settings에서 설정해주세요.")
     
     if not os.getenv("BRAVE_SEARCH_API_KEY"):
         print("\n⚠️  BRAVE_SEARCH_API_KEY가 설정되지 않았습니다.")
-        print("   (선택사항: 검색 기능을 사용하려면 입력)")
-        key = input("Brave Search API Key 입력 (Enter=건너뛰기): ").strip()
-        if key:
-            os.environ["BRAVE_SEARCH_API_KEY"] = key
-            search_client = BraveSearchClient(key)
-    
-    # 시스템 재초기화
-    if llm_client:
-        multi_agent_system = MultiAgentSystem(llm_client, search_client)
-        gradio_app = create_gradio_interface(multi_agent_system, search_client)
-        app = gr.mount_gradio_app(app, gradio_app, path="/ui")
+        print("검색 기능이 비활성화됩니다.")
     
-    print("\n" + "="*60)
-    print("✅ 시스템 준비 완료!")
-    print("="*60)
+    # Gradio 앱 실행
+    demo = create_gradio_interface()
     
     # Hugging Face Spaces 환경 확인
     is_hf_spaces = os.getenv("SPACE_ID") is not None
-    port = int(os.getenv("PORT", 7860))
     
     if is_hf_spaces:
         print("\n🤗 Hugging Face Spaces에서 실행 중...")
-        print(f"   포트: {port}")
+        demo.launch(server_name="0.0.0.0", server_port=7860)
     else:
-        print("\n📍 접속 주소:")
-        print(f"   🎨 Gradio UI: http://localhost:{port}/ui")
-        print(f"   📚 API Docs: http://localhost:{port}/docs")
-        print(f"   🔧 Chat API: POST http://localhost:{port}/api/chat")
-    
-    print("\n💡 Ctrl+C를 눌러 종료")
-    print("="*60 + "\n")
-    
-    # Hugging Face Spaces 환경 감지
-    port = int(os.getenv("PORT", 7860))
-    host = "0.0.0.0"
-    
-    uvicorn.run(
-        app,
-        host=host,
-        port=port,
-        reload=False,
-        log_level="info"
-    )
+        print("\n💻 로컬 환경에서 실행 중...")
+        demo.launch(server_name="0.0.0.0", server_port=7860, share=False)