Private-AI

Running

App Files Files Community

seawolf2357 commited on Jun 24

Commit

4b7d6e9

verified ·

1 Parent(s): 6415ee8

Update app.py

Browse files

Files changed (1) hide show

app.py +384 -322

app.py CHANGED Viewed

@@ -23,6 +23,11 @@ import gradio as gr
 import io
 from scipy import signal
 import wave
 load_dotenv()
@@ -61,6 +66,106 @@ SUPPORTED_LANGUAGES = {
     "ms": "Bahasa Melayu (Malay)"
 }
 # HTML content embedded as a string
 HTML_CONTENT = """<!DOCTYPE html>
 <html lang="ko">
@@ -147,6 +252,60 @@ HTML_CONTENT = """<!DOCTYPE html>
             font-size: 32px;
             letter-spacing: 1px;
         }
         /* Settings section */
         .settings-section {
             background-color: var(--card-bg);
@@ -518,8 +677,14 @@ HTML_CONTENT = """<!DOCTYPE html>
         <div class="main-content">
             <div class="sidebar">
                 <div class="settings-section">
-                    <h3 style="margin: 0 0 15px 0; color: var(--primary-color);">설정(텍스트 채팅에만 적용)</h3>
                     <div class="settings-grid">
                         <div class="setting-item">
                             <span class="setting-label">웹 검색</span>
@@ -528,9 +693,15 @@ HTML_CONTENT = """<!DOCTYPE html>
                             </div>
                         </div>
                         <div class="setting-item">
-                            <span class="setting-label">다국어 번역 채팅</span>
                             <select id="language-select">
-                                <option value="">비활성화</option>
                                 <option value="ko">한국어 (Korean)</option>
                                 <option value="en">English</option>
                                 <option value="es">Español (Spanish)</option>
@@ -565,7 +736,7 @@ HTML_CONTENT = """<!DOCTYPE html>
                     </div>
                     <div class="text-input-section">
                         <label for="system-prompt" class="setting-label">시스템 프롬프트:</label>
-                        <textarea id="system-prompt" placeholder="AI 어시스턴트의 성격, 역할, 행동 방식을 정의하세요...">You are a helpful assistant. Respond in a friendly and professional manner.</textarea>
                     </div>
                 </div>
@@ -594,8 +765,11 @@ HTML_CONTENT = """<!DOCTYPE html>
         let peerConnection;
         let webrtc_id;
         let webSearchEnabled = false;
         let selectedLanguage = "";
-        let systemPrompt = "You are a helpful assistant. Respond in a friendly and professional manner.";
         const audioOutput = document.getElementById('audio-output');
         const startButton = document.getElementById('start-button');
         const sendButton = document.getElementById('send-button');
@@ -603,15 +777,87 @@ HTML_CONTENT = """<!DOCTYPE html>
         const statusDot = document.getElementById('status-dot');
         const statusText = document.getElementById('status-text');
         const searchToggle = document.getElementById('search-toggle');
         const languageSelect = document.getElementById('language-select');
         const systemPromptInput = document.getElementById('system-prompt');
         const textInput = document.getElementById('text-input');
         let audioLevel = 0;
         let animationFrame;
         let audioContext, analyser, audioSource;
         let dataChannel = null;
         let isVoiceActive = false;
         // Web search toggle functionality
         searchToggle.addEventListener('click', () => {
             webSearchEnabled = !webSearchEnabled;
@@ -619,6 +865,14 @@ HTML_CONTENT = """<!DOCTYPE html>
             console.log('Web search enabled:', webSearchEnabled);
         });
         // Language selection
         languageSelect.addEventListener('change', () => {
             selectedLanguage = languageSelect.value;
@@ -627,7 +881,7 @@ HTML_CONTENT = """<!DOCTYPE html>
         // System prompt update
         systemPromptInput.addEventListener('input', () => {
-            systemPrompt = systemPromptInput.value || "You are a helpful assistant. Respond in a friendly and professional manner.";
         });
         // Text input handling
@@ -645,7 +899,7 @@ HTML_CONTENT = """<!DOCTYPE html>
             if (!message) return;
             // Add user message to chat
-            addMessage('user', message);
             textInput.value = '';
             // Show sending indicator
@@ -664,8 +918,10 @@ HTML_CONTENT = """<!DOCTYPE html>
                     body: JSON.stringify({
                         message: message,
                         web_search_enabled: webSearchEnabled,
-                        target_language: selectedLanguage,
-                        system_prompt: systemPrompt
                     })
                 });
@@ -680,10 +936,13 @@ HTML_CONTENT = """<!DOCTYPE html>
                 } else {
                     // Add assistant response
                     let content = data.response;
-                    if (selectedLanguage && data.language) {
                         content += ` <span class="language-info">[${data.language}]</span>`;
                     }
-                    addMessage('assistant', content);
                 }
             } catch (error) {
                 console.error('Error sending text message:', error);
@@ -704,10 +963,11 @@ HTML_CONTENT = """<!DOCTYPE html>
                 sendButton.style.display = 'none';
             } else {
                 statusText.textContent = '연결 대기 중';
-                sendButton.style.display = 'block';  // Show send button even when disconnected for text chat
                 isVoiceActive = false;
             }
         }
         function updateButtonState() {
             const button = document.getElementById('start-button');
             if (peerConnection && (peerConnection.connectionState === 'connecting' || peerConnection.connectionState === 'new')) {
@@ -737,6 +997,7 @@ HTML_CONTENT = """<!DOCTYPE html>
                 updateStatus('disconnected');
             }
         }
         function setupAudioVisualization(stream) {
             audioContext = new (window.AudioContext || window.webkitAudioContext)();
             analyser = audioContext.createAnalyser();
@@ -771,6 +1032,7 @@ HTML_CONTENT = """<!DOCTYPE html>
             updateAudioLevel();
         }
         function showError(message) {
             const toast = document.getElementById('error-toast');
             toast.textContent = message;
@@ -780,6 +1042,7 @@ HTML_CONTENT = """<!DOCTYPE html>
                 toast.style.display = 'none';
             }, 5000);
         }
         async function setupWebRTC() {
             const config = __RTC_CONFIGURATION__;
             peerConnection = new RTCPeerConnection(config);
@@ -792,6 +1055,7 @@ HTML_CONTENT = """<!DOCTYPE html>
                     toast.style.display = 'none';
                 }, 5000);
             }, 5000);
             try {
                 const stream = await navigator.mediaDevices.getUserMedia({
                     audio: true
@@ -834,6 +1098,7 @@ HTML_CONTENT = """<!DOCTYPE html>
                         peerConnection.addEventListener("icegatheringstatechange", checkState);
                     }
                 });
                 peerConnection.addEventListener('connectionstatechange', () => {
                     console.log('connectionstatechange', peerConnection.connectionState);
                     if (peerConnection.connectionState === 'connected') {
@@ -843,15 +1108,8 @@ HTML_CONTENT = """<!DOCTYPE html>
                     }
                     updateButtonState();
                 });
-                webrtc_id = Math.random().toString(36).substring(7);
-                // Log current settings before sending
-                console.log('Sending offer with settings:', {
-                    webrtc_id: webrtc_id,
-                    web_search_enabled: webSearchEnabled,
-                    target_language: selectedLanguage,
-                    system_prompt: systemPrompt
-                });
                 const response = await fetch('/webrtc/offer', {
                     method: 'POST',
@@ -861,10 +1119,13 @@ HTML_CONTENT = """<!DOCTYPE html>
                         type: peerConnection.localDescription.type,
                         webrtc_id: webrtc_id,
                         web_search_enabled: webSearchEnabled,
-                        target_language: selectedLanguage,
-                        system_prompt: systemPrompt
                     })
                 });
                 const serverResponse = await response.json();
                 if (serverResponse.status === 'failed') {
                     showError(serverResponse.meta.error === 'concurrency_limit_reached'
@@ -873,21 +1134,25 @@ HTML_CONTENT = """<!DOCTYPE html>
                     stop();
                     return;
                 }
                 await peerConnection.setRemoteDescription(serverResponse);
                 const eventSource = new EventSource('/outputs?webrtc_id=' + webrtc_id);
-                eventSource.addEventListener("output", (event) => {
                     const eventJson = JSON.parse(event.data);
                     let content = eventJson.content;
-                    if (selectedLanguage && eventJson.language) {
                         content += ` <span class="language-info">[${eventJson.language}]</span>`;
                     }
-                    addMessage("assistant", content);
                 });
                 eventSource.addEventListener("search", (event) => {
                     const eventJson = JSON.parse(event.data);
                     if (eventJson.query) {
-                        addMessage("search-result", `웹 검색 중: "${eventJson.query}"`);
                     }
                 });
             } catch (err) {
@@ -897,7 +1162,8 @@ HTML_CONTENT = """<!DOCTYPE html>
                 stop();
             }
         }
-        function addMessage(role, content) {
             const messageDiv = document.createElement('div');
             messageDiv.classList.add('message', role);
@@ -908,18 +1174,30 @@ HTML_CONTENT = """<!DOCTYPE html>
             }
             chatMessages.appendChild(messageDiv);
             chatMessages.scrollTop = chatMessages.scrollHeight;
         }
         function stop() {
             console.log('[STOP] Stopping connection...');
-            // Cancel animation frame first
             if (animationFrame) {
                 cancelAnimationFrame(animationFrame);
                 animationFrame = null;
             }
-            // Close audio context
             if (audioContext) {
                 audioContext.close();
                 audioContext = null;
@@ -927,17 +1205,14 @@ HTML_CONTENT = """<!DOCTYPE html>
                 audioSource = null;
             }
-            // Close data channel
             if (dataChannel) {
                 dataChannel.close();
                 dataChannel = null;
             }
-            // Close peer connection
             if (peerConnection) {
                 console.log('[STOP] Current connection state:', peerConnection.connectionState);
-                // Stop all transceivers
                 if (peerConnection.getTransceivers) {
                     peerConnection.getTransceivers().forEach(transceiver => {
                         if (transceiver.stop) {
@@ -946,7 +1221,6 @@ HTML_CONTENT = """<!DOCTYPE html>
                     });
                 }
-                // Stop all senders
                 if (peerConnection.getSenders) {
                     peerConnection.getSenders().forEach(sender => {
                         if (sender.track) {
@@ -955,7 +1229,6 @@ HTML_CONTENT = """<!DOCTYPE html>
                     });
                 }
-                // Stop all receivers
                 if (peerConnection.getReceivers) {
                     peerConnection.getReceivers().forEach(receiver => {
                         if (receiver.track) {
@@ -964,28 +1237,22 @@ HTML_CONTENT = """<!DOCTYPE html>
                     });
                 }
-                // Close the connection
                 peerConnection.close();
-                // Clear the reference
                 peerConnection = null;
                 console.log('[STOP] Connection closed');
             }
-            // Reset audio level
             audioLevel = 0;
             isVoiceActive = false;
-            // Update UI
             updateButtonState();
-            // Clear any existing webrtc_id
             if (webrtc_id) {
                 console.log('[STOP] Clearing webrtc_id:', webrtc_id);
                 webrtc_id = null;
             }
         }
         startButton.addEventListener('click', () => {
             console.log('clicked');
             console.log(peerConnection, peerConnection?.connectionState);
@@ -996,11 +1263,6 @@ HTML_CONTENT = """<!DOCTYPE html>
                 stop();
             }
         });
-        // Initialize on page load
-        window.addEventListener('DOMContentLoaded', () => {
-            sendButton.style.display = 'block';
-        });
     </script>
 </body>
@@ -1051,7 +1313,6 @@ class BraveSearchClient:
 # Initialize search client globally
 brave_api_key = os.getenv("BSEARCH_API")
 search_client = BraveSearchClient(brave_api_key) if brave_api_key else None
-print(f"Search client initialized: {search_client is not None}, API key present: {bool(brave_api_key)}")
 # Store connection settings
 connection_settings = {}
@@ -1076,32 +1337,25 @@ def update_chatbot(chatbot: list[dict], response: ResponseAudioTranscriptDoneEve
 async def process_text_chat(message: str, web_search_enabled: bool, target_language: str,
-                          system_prompt: str) -> Dict[str, str]:
     """Process text chat using GPT-4o-mini model"""
     try:
-        # If target language is set, override system prompt completely
-        if target_language:
-            language_name = SUPPORTED_LANGUAGES.get(target_language, target_language)
-            # Create system prompt in target language
-            if target_language == "en":
-                base_instructions = f"You are a helpful assistant. You speak ONLY English. Never use Korean or any other language. {system_prompt}"
-                user_prefix = "Please respond in English: "
-            elif target_language == "ja":
-                base_instructions = f"あなたは親切なアシスタントです。日本語のみを話します。韓国語や他の言語は絶対に使用しません。{system_prompt}"
-                user_prefix = "日本語で答えてください: "
-            elif target_language == "zh":
-                base_instructions = f"你是一个乐于助人的助手。你只说中文。绝不使用韩语或其他语言。{system_prompt}"
-                user_prefix = "请用中文回答: "
-            elif target_language == "es":
-                base_instructions = f"Eres un asistente útil. Solo hablas español. Nunca uses coreano u otros idiomas. {system_prompt}"
-                user_prefix = "Por favor responde en español: "
             else:
-                base_instructions = f"You are a helpful assistant that speaks ONLY {language_name}. {system_prompt}"
-                user_prefix = f"Please respond in {language_name}: "
         else:
-            base_instructions = system_prompt or "You are a helpful assistant."
-            user_prefix = ""
         messages = [
             {"role": "system", "content": base_instructions}
@@ -1109,7 +1363,6 @@ async def process_text_chat(message: str, web_search_enabled: bool, target_langu
         # Handle web search if enabled
         if web_search_enabled and search_client:
-            # Check if the message requires web search
             search_keywords = ["날씨", "기온", "비", "눈", "뉴스", "소식", "현재", "최근",
                              "오늘", "지금", "가격", "환율", "주가", "weather", "news",
                              "current", "today", "price", "2024", "2025"]
@@ -1117,26 +1370,18 @@ async def process_text_chat(message: str, web_search_enabled: bool, target_langu
             should_search = any(keyword in message.lower() for keyword in search_keywords)
             if should_search:
-                # Perform web search
                 search_results = await search_client.search(message)
                 if search_results:
                     search_context = "웹 검색 결과:\n\n"
                     for i, result in enumerate(search_results[:5], 1):
                         search_context += f"{i}. {result['title']}\n{result['description']}\n\n"
-                    # Add search context in target language if set
-                    if target_language:
-                        search_instruction = f"Use this search information but respond in {SUPPORTED_LANGUAGES.get(target_language, target_language)} only: "
-                    else:
-                        search_instruction = "다음 웹 검색 결과를 참고하여 답변하세요: "
                     messages.append({
                         "role": "system",
-                        "content": search_instruction + "\n\n" + search_context
                     })
-        # Add user message with language prefix
-        messages.append({"role": "user", "content": user_prefix + message})
         # Call GPT-4o-mini
         response = await client.chat.completions.create(
@@ -1148,23 +1393,8 @@ async def process_text_chat(message: str, web_search_enabled: bool, target_langu
         response_text = response.choices[0].message.content
-        # Final check - remove any Korean if target language is not Korean
-        if target_language and target_language != "ko":
-            import re
-            if re.search(r'[가-힣]', response_text):
-                print(f"[TEXT CHAT] WARNING: Korean detected in response for {target_language}")
-                # Try again with stronger prompt
-                messages[-1] = {"role": "user", "content": f"ONLY {SUPPORTED_LANGUAGES.get(target_language, target_language)}, NO KOREAN: {message}"}
-                retry_response = await client.chat.completions.create(
-                    model="gpt-4.1-mini",
-                    messages=messages,
-                    temperature=0.3,
-                    max_tokens=2000
-                )
-                response_text = retry_response.choices[0].message.content
-        print(f"[TEXT CHAT] Target language: {target_language}")
-        print(f"[TEXT CHAT] Response preview: {response_text[:100]}...")
         return {
             "response": response_text,
@@ -1178,7 +1408,8 @@ async def process_text_chat(message: str, web_search_enabled: bool, target_langu
 class OpenAIHandler(AsyncStreamHandler):
     def __init__(self, web_search_enabled: bool = False, target_language: str = "",
-                 system_prompt: str = "", webrtc_id: str = None) -> None:
         super().__init__(
             expected_layout="mono",
             output_sample_rate=SAMPLE_RATE,
@@ -1195,14 +1426,11 @@ class OpenAIHandler(AsyncStreamHandler):
         self.web_search_enabled = web_search_enabled
         self.target_language = target_language
         self.system_prompt = system_prompt
-        print(f"[INIT] Handler created with web_search={web_search_enabled}, "
-              f"target_language={target_language}")
     def copy(self):
-        # Get the most recent settings
         if connection_settings:
-            # Get the most recent webrtc_id
             recent_ids = sorted(connection_settings.keys(),
                               key=lambda k: connection_settings[k].get('timestamp', 0),
                               reverse=True)
@@ -1210,17 +1438,15 @@ class OpenAIHandler(AsyncStreamHandler):
                 recent_id = recent_ids[0]
                 settings = connection_settings[recent_id]
-                # Log the settings being copied
-                print(f"[COPY] Copying settings from {recent_id}:")
                 return OpenAIHandler(
                     web_search_enabled=settings.get('web_search_enabled', False),
                     target_language=settings.get('target_language', ''),
                     system_prompt=settings.get('system_prompt', ''),
-                    webrtc_id=recent_id
                 )
-        print(f"[COPY] No settings found, creating default handler")
         return OpenAIHandler(web_search_enabled=False)
     async def search_web(self, query: str) -> str:
@@ -1233,7 +1459,6 @@ class OpenAIHandler(AsyncStreamHandler):
         if not results:
             return f"'{query}'에 대한 검색 결과를 찾을 수 없습니다."
-        # Format search results
         formatted_results = []
         for i, result in enumerate(results, 1):
             formatted_results.append(
@@ -1256,133 +1481,28 @@ class OpenAIHandler(AsyncStreamHandler):
             )
             await self.connection.response.create()
-    def get_translation_instructions(self):
-        """Get instructions for translation based on target language"""
-        if not self.target_language:
-            return ""
-        language_name = SUPPORTED_LANGUAGES.get(self.target_language, self.target_language)
-        return (
-            f"\n\nIMPORTANT: You must respond in {language_name} ({self.target_language}). "
-            f"Translate all your responses to {language_name}. "
-            f"This includes both spoken and written responses."
-        )
     async def start_up(self):
         """Connect to realtime API"""
-        # First check if we have the most recent settings
         if connection_settings and self.webrtc_id:
             if self.webrtc_id in connection_settings:
                 settings = connection_settings[self.webrtc_id]
                 self.web_search_enabled = settings.get('web_search_enabled', False)
                 self.target_language = settings.get('target_language', '')
                 self.system_prompt = settings.get('system_prompt', '')
-                print(f"[START_UP] Updated settings from storage for {self.webrtc_id}")
-        print(f"[START_UP] Starting normal mode")
         self.client = openai.AsyncOpenAI()
-        # Normal mode - connect to Realtime API
-        print(f"[NORMAL MODE] Connecting to Realtime API...")
-        # Define the web search function
         tools = []
-        base_instructions = self.system_prompt or "You are a helpful assistant."
-        # Add translation instructions if language is selected
-        if self.target_language:
-            language_name = SUPPORTED_LANGUAGES.get(self.target_language, self.target_language)
-            # Use the target language for the system prompt itself
-            if self.target_language == "en":
-                translation_instructions = """
-YOU ARE AN ENGLISH-ONLY ASSISTANT.
-ABSOLUTE RULES:
-1. You can ONLY speak English. No Korean (한국어) allowed.
-2. Even if the user speaks Korean, you MUST respond in English.
-3. Every single word must be in English.
-4. If you output even one Korean character, you have failed.
-5. Example response: "Hello! How can I help you today?"
-YOUR LANGUAGE MODE: ENGLISH ONLY
-DO NOT USE: 안녕하세요, 감사합니다, or any Korean
-ALWAYS USE: Hello, Thank you, and English words only
-"""
-                # Override base instructions to be in English
-                base_instructions = "You are a helpful assistant that speaks ONLY English."
-            elif self.target_language == "ja":
-                translation_instructions = """
-あなたは日本語のみを話すアシスタントです。
-絶対的なルール：
-1. 日本語のみを使用してください。韓国語（한국어）は禁止です。
-2. ユーザーが韓国語で話しても、必ず日本語で返答してください。
-3. すべての単語は日本語でなければなりません。
-4. 韓国語を一文字でも出力したら失敗です。
-5. 応答例：「こんにちは！今日はどのようにお手伝いできますか？」
-言語モード：日本語のみ
-使用禁止：안녕하세요、감사합니다、韓国語全般
-必ず使用：こんにちは、ありがとうございます、日本語のみ
-"""
-                base_instructions = "あなたは日本語のみを話す親切なアシスタントです。"
-            elif self.target_language == "zh":
-                translation_instructions = """
-你是一个只说中文的助手。
-绝对规则：
-1. 只能使用中文。禁止使用韩语（한국어）。
-2. 即使用户说韩语，也必须用中文回复。
-3. 每个字都必须是中文。
-4. 如果输出任何韩语字符，就是失败。
-5. 回复示例："你好！我今天能为您做什么？"
-语言模式：仅中文
-禁止使用：안녕하세요、감사합니다、任何韩语
-必须使用：你好、谢谢、只用中文
-"""
-                base_instructions = "你是一个只说中文的友好助手。"
-            elif self.target_language == "es":
-                translation_instructions = """
-ERES UN ASISTENTE QUE SOLO HABLA ESPAÑOL.
-REGLAS ABSOLUTAS:
-1. Solo puedes hablar español. No se permite coreano (한국어).
-2. Incluso si el usuario habla coreano, DEBES responder en español.
-3. Cada palabra debe estar en español.
-4. Si produces aunque sea un carácter coreano, has fallado.
-5. Respuesta ejemplo: "¡Hola! ¿Cómo puedo ayudarte hoy?"
-MODO DE IDIOMA: SOLO ESPAÑOL
-NO USAR: 안녕하세요, 감사합니다, o cualquier coreano
-SIEMPRE USAR: Hola, Gracias, y solo palabras en español
-"""
-                base_instructions = "Eres un asistente útil que habla SOLO español."
-            else:
-                translation_instructions = f"""
-YOU MUST ONLY SPEAK {language_name.upper()}.
-RULES:
-1. Output only in {language_name}
-2. Never use Korean
-3. Always respond in {language_name}
-"""
-                base_instructions = f"You are a helpful assistant that speaks ONLY {language_name}."
-        else:
-            translation_instructions = ""
         if self.web_search_enabled and self.search_client:
             tools = [{
                 "type": "function",
                 "function": {
                     "name": "web_search",
-                    "description": "Search the web for current information. Use this for weather, news, prices, current events, or any time-sensitive topics.",
                     "parameters": {
                         "type": "object",
                         "properties": {
@@ -1395,94 +1515,29 @@ RULES:
                     }
                 }
             }]
-            print("Web search function added to tools")
-            search_instructions = (
-                "\n\nYou have web search capabilities. "
-                "IMPORTANT: You MUST use the web_search function for ANY of these topics:\n"
-                "- Weather (날씨, 기온, 비, 눈)\n"
-                "- News (뉴스, 소식)\n"
-                "- Current events (현재, 최근, 오늘, 지금)\n"
-                "- Prices (가격, 환율, 주가)\n"
-                "- Sports scores or results\n"
-                "- Any question about 2024 or 2025\n"
-                "- Any time-sensitive information\n\n"
-                "When in doubt, USE web_search. It's better to search and provide accurate information "
-                "than to guess or use outdated information."
-            )
-            # Combine all instructions
-            if translation_instructions:
-                # Translation instructions already include base_instructions
-                instructions = translation_instructions + search_instructions
-            else:
-                instructions = base_instructions + search_instructions
-        else:
-            # No web search
-            if translation_instructions:
-                instructions = translation_instructions
-            else:
-                instructions = base_instructions
-        print(f"[NORMAL MODE] Base instructions: {base_instructions[:100]}...")
-        print(f"[NORMAL MODE] Translation instructions: {translation_instructions[:200] if translation_instructions else 'None'}...")
-        print(f"[NORMAL MODE] Combined instructions length: {len(instructions)}")
-        print(f"[NORMAL MODE] Target language: {self.target_language}")
         async with self.client.beta.realtime.connect(
             model="gpt-4o-mini-realtime-preview-2024-12-17"
         ) as conn:
-            # Update session with tools
             session_update = {
                 "turn_detection": {"type": "server_vad"},
-                "instructions": instructions,
                 "tools": tools,
                 "tool_choice": "auto" if tools else "none",
                 "temperature": 0.7,
                 "max_response_output_tokens": 4096,
                 "modalities": ["text", "audio"],
-                "voice": "alloy"  # Default voice
             }
-            # Use appropriate voice for the language
-            if self.target_language:
-                # Force language through multiple mechanisms
-                # 1. Use voice that's known to work well with the language
-                voice_map = {
-                    "en": "nova",      # Nova has clearer English
-                    "es": "nova",      # Nova works for Spanish
-                    "fr": "shimmer",   # Shimmer for French
-                    "de": "echo",      # Echo for German
-                    "ja": "alloy",     # Alloy can do Japanese
-                    "zh": "alloy",     # Alloy can do Chinese
-                    "ko": "nova",      # Nova for Korean
-                }
-                session_update["voice"] = voice_map.get(self.target_language, "nova")
-                # 2. Add language to modalities (experimental)
-                session_update["modalities"] = ["text", "audio"]
-                # 3. Set output format
-                session_update["output_audio_format"] = "pcm16"
-                # 4. Add language hint to the system (if supported by API)
-                if self.target_language in ["en", "es", "fr", "de", "ja", "zh"]:
-                    session_update["language"] = self.target_language  # Try setting language directly
-                print(f"[TRANSLATION MODE] Session update: {json.dumps(session_update, indent=2)}")
             await conn.session.update(session=session_update)
             self.connection = conn
-            print(f"Connected with tools: {len(tools)} functions, voice: {session_update.get('voice', 'default')}")
             async for event in self.connection:
-                # Debug logging for function calls
-                if event.type.startswith("response.function_call"):
-                    print(f"Function event: {event.type}")
                 if event.type == "response.audio_transcript.done":
-                    print(f"[RESPONSE] Transcript: {event.transcript[:100]}...")
-                    print(f"[RESPONSE] Expected language: {self.target_language}")
                     output_data = {
                         "event": event,
@@ -1502,7 +1557,6 @@ RULES:
                 # Handle function calls
                 elif event.type == "response.function_call_arguments.start":
-                    print(f"Function call started")
                     self.function_call_in_progress = True
                     self.current_function_args = ""
                     self.current_call_id = getattr(event, 'call_id', None)
@@ -1513,22 +1567,17 @@ RULES:
                 elif event.type == "response.function_call_arguments.done":
                     if self.function_call_in_progress:
-                        print(f"Function call done, args: {self.current_function_args}")
                         try:
                             args = json.loads(self.current_function_args)
                             query = args.get("query", "")
-                            # Emit search event to client
                             await self.output_queue.put(AdditionalOutputs({
                                 "type": "search",
                                 "query": query
                             }))
-                            # Perform the search
                             search_results = await self.search_web(query)
-                            print(f"Search results length: {len(search_results)}")
-                            # Send function result back to the model
                             if self.connection and self.current_call_id:
                                 await self.connection.conversation.item.create(
                                     item={
@@ -1547,9 +1596,7 @@ RULES:
                             self.current_call_id = None
     async def receive(self, frame: tuple[int, np.ndarray]) -> None:
-        # Normal mode - use Realtime API
         if not self.connection:
-            print(f"[RECEIVE] No connection in normal mode, skipping")
             return
         try:
             _, array = frame
@@ -1560,10 +1607,8 @@ RULES:
             print(f"Error in receive: {e}")
     async def emit(self) -> tuple[int, np.ndarray] | AdditionalOutputs | None:
-        # Normal mode
         item = await wait_for_item(self.output_queue)
-        # Check if it's a dict with text message
         if isinstance(item, dict) and item.get('type') == 'text_message':
             await self.process_text_message(item['content'])
             return None
@@ -1571,13 +1616,9 @@ RULES:
         return item
     async def shutdown(self) -> None:
-        print(f"[SHUTDOWN] Called")
-        # Normal mode - close Realtime API connection
         if self.connection:
             await self.connection.close()
             self.connection = None
-            print("[NORMAL MODE] Connection closed")
 # Create initial handler instance
@@ -1588,7 +1629,7 @@ chatbot = gr.Chatbot(type="messages")
 # Create stream with handler instance
 stream = Stream(
-    handler,  # Pass instance, not factory
     mode="send-receive",
     modality="audio",
     additional_inputs=[chatbot],
@@ -1601,6 +1642,11 @@ stream = Stream(
 app = FastAPI()
 # Mount stream
 stream.mount(app)
@@ -1614,22 +1660,18 @@ async def custom_offer(request: Request):
     web_search_enabled = body.get("web_search_enabled", False)
     target_language = body.get("target_language", "")
     system_prompt = body.get("system_prompt", "")
-    print(f"[OFFER] Received offer with webrtc_id: {webrtc_id}")
-    print(f"[OFFER] web_search_enabled: {web_search_enabled}")
-    print(f"[OFFER] target_language: {target_language}")
-    # Store settings with timestamp
     if webrtc_id:
         connection_settings[webrtc_id] = {
             'web_search_enabled': web_search_enabled,
             'target_language': target_language,
             'system_prompt': system_prompt,
             'timestamp': asyncio.get_event_loop().time()
         }
-        print(f"[OFFER] Stored settings for {webrtc_id}:")
-        print(f"[OFFER] {connection_settings[webrtc_id]}")
     # Remove our custom route temporarily
     custom_route = None
@@ -1638,16 +1680,11 @@ async def custom_offer(request: Request):
             custom_route = app.routes.pop(i)
             break
-    # Forward to stream's offer handler
-    print(f"[OFFER] Forwarding to stream.offer()")
     response = await stream.offer(body)
-    # Re-add our custom route
     if custom_route:
         app.routes.insert(0, custom_route)
-    print(f"[OFFER] Response status: {response.get('status', 'unknown') if isinstance(response, dict) else 'OK'}")
     return response
@@ -1660,12 +1697,14 @@ async def chat_text(request: Request):
         web_search_enabled = body.get("web_search_enabled", False)
         target_language = body.get("target_language", "")
         system_prompt = body.get("system_prompt", "")
         if not message:
             return {"error": "메시지가 비어있습니다."}
-        # Process text chat
-        result = await process_text_chat(message, web_search_enabled, target_language, system_prompt)
         return result
@@ -1674,16 +1713,41 @@ async def chat_text(request: Request):
         return {"error": "채팅 처리 중 오류가 발생했습니다."}
 @app.post("/text_message/{webrtc_id}")
 async def receive_text_message(webrtc_id: str, request: Request):
     """Receive text message from client"""
     body = await request.json()
     message = body.get("content", "")
-    # Find the handler for this connection
     if webrtc_id in stream.handlers:
         handler = stream.handlers[webrtc_id]
-        # Queue the text message for processing
         await handler.output_queue.put({
             'type': 'text_message',
             'content': message
@@ -1698,10 +1762,8 @@ async def outputs(webrtc_id: str):
     async def output_stream():
         async for output in stream.output_stream(webrtc_id):
             if hasattr(output, 'args') and output.args:
-                # Check if it's a search event
                 if isinstance(output.args[0], dict) and output.args[0].get('type') == 'search':
                     yield f"event: search\ndata: {json.dumps(output.args[0])}\n\n"
-                # Regular transcript event with language info
                 elif isinstance(output.args[0], dict) and 'event' in output.args[0]:
                     event_data = output.args[0]
                     if 'event' in event_data and hasattr(event_data['event'], 'transcript'):

 import io
 from scipy import signal
 import wave
+from datetime import datetime
+import sqlite3
+import aiosqlite
+from langdetect import detect, LangDetectException
+import uuid
 load_dotenv()
     "ms": "Bahasa Melayu (Malay)"
 }
+# Database setup
+DB_PATH = "chat_history.db"
+async def init_db():
+    """Initialize the SQLite database"""
+    async with aiosqlite.connect(DB_PATH) as db:
+        await db.execute("""
+            CREATE TABLE IF NOT EXISTS conversations (
+                id TEXT PRIMARY KEY,
+                title TEXT NOT NULL,
+                created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+                updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+            )
+        """)
+        await db.execute("""
+            CREATE TABLE IF NOT EXISTS messages (
+                id INTEGER PRIMARY KEY AUTOINCREMENT,
+                conversation_id TEXT NOT NULL,
+                role TEXT NOT NULL,
+                content TEXT NOT NULL,
+                language TEXT,
+                created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+                FOREIGN KEY (conversation_id) REFERENCES conversations (id)
+            )
+        """)
+        await db.commit()
+async def save_message(conversation_id: str, role: str, content: str, language: str = None):
+    """Save a message to the database"""
+    async with aiosqlite.connect(DB_PATH) as db:
+        # Check if conversation exists
+        cursor = await db.execute(
+            "SELECT id FROM conversations WHERE id = ?",
+            (conversation_id,)
+        )
+        exists = await cursor.fetchone()
+        if not exists:
+            # Create new conversation
+            title = content[:50] + "..." if len(content) > 50 else content
+            await db.execute(
+                "INSERT INTO conversations (id, title) VALUES (?, ?)",
+                (conversation_id, title)
+            )
+        else:
+            # Update conversation timestamp
+            await db.execute(
+                "UPDATE conversations SET updated_at = CURRENT_TIMESTAMP WHERE id = ?",
+                (conversation_id,)
+            )
+        # Insert message
+        await db.execute(
+            "INSERT INTO messages (conversation_id, role, content, language) VALUES (?, ?, ?, ?)",
+            (conversation_id, role, content, language)
+        )
+        await db.commit()
+async def get_conversations():
+    """Get all conversations"""
+    async with aiosqlite.connect(DB_PATH) as db:
+        db.row_factory = aiosqlite.Row
+        cursor = await db.execute(
+            "SELECT * FROM conversations ORDER BY updated_at DESC"
+        )
+        conversations = await cursor.fetchall()
+        return [dict(conv) for conv in conversations]
+async def get_conversation_messages(conversation_id: str):
+    """Get all messages for a conversation"""
+    async with aiosqlite.connect(DB_PATH) as db:
+        db.row_factory = aiosqlite.Row
+        cursor = await db.execute(
+            "SELECT * FROM messages WHERE conversation_id = ? ORDER BY created_at",
+            (conversation_id,)
+        )
+        messages = await cursor.fetchall()
+        return [dict(msg) for msg in messages]
+def detect_language(text: str) -> str:
+    """Detect the language of the input text"""
+    try:
+        lang = detect(text)
+        # Map detected language to our supported languages
+        if lang == 'ko':
+            return 'ko'
+        elif lang == 'en':
+            return 'en'
+        elif lang in SUPPORTED_LANGUAGES:
+            return lang
+        else:
+            # Default to Korean if unsupported language
+            return 'ko'
+    except LangDetectException:
+        # Default to Korean if detection fails
+        return 'ko'
 # HTML content embedded as a string
 HTML_CONTENT = """<!DOCTYPE html>
 <html lang="ko">
             font-size: 32px;
             letter-spacing: 1px;
         }
+        /* History section */
+        .history-section {
+            background-color: var(--card-bg);
+            border-radius: 12px;
+            padding: 20px;
+            border: 1px solid var(--border-color);
+            overflow-y: auto;
+            flex-shrink: 0;
+            max-height: 300px;
+        }
+        .history-item {
+            padding: 10px;
+            margin-bottom: 8px;
+            background-color: var(--dark-bg);
+            border-radius: 6px;
+            cursor: pointer;
+            transition: all 0.2s;
+            display: flex;
+            justify-content: space-between;
+            align-items: center;
+        }
+        .history-item:hover {
+            background-color: var(--hover-color);
+            transform: translateX(5px);
+        }
+        .history-item-title {
+            font-size: 14px;
+            color: var(--text-color);
+            overflow: hidden;
+            text-overflow: ellipsis;
+            white-space: nowrap;
+            flex-grow: 1;
+        }
+        .history-item-date {
+            font-size: 12px;
+            color: #888;
+            margin-left: 10px;
+        }
+        .new-chat-button {
+            width: 100%;
+            background: linear-gradient(135deg, #2ecc71, #27ae60);
+            color: white;
+            border: none;
+            padding: 10px 20px;
+            font-size: 14px;
+            border-radius: 6px;
+            cursor: pointer;
+            margin-bottom: 10px;
+            transition: all 0.3s;
+        }
+        .new-chat-button:hover {
+            background: linear-gradient(135deg, #27ae60, #229954);
+            transform: translateY(-2px);
+        }
         /* Settings section */
         .settings-section {
             background-color: var(--card-bg);
         <div class="main-content">
             <div class="sidebar">
+                <div class="history-section">
+                    <h3 style="margin: 0 0 15px 0; color: var(--primary-color);">대화 기록</h3>
+                    <button class="new-chat-button" onclick="startNewChat()">+ 새 대화</button>
+                    <div id="history-list"></div>
+                </div>
                 <div class="settings-section">
+                    <h3 style="margin: 0 0 15px 0; color: var(--primary-color);">설정</h3>
                     <div class="settings-grid">
                         <div class="setting-item">
                             <span class="setting-label">웹 검색</span>
                             </div>
                         </div>
                         <div class="setting-item">
+                            <span class="setting-label">자동 언어 감지</span>
+                            <div id="auto-lang-toggle" class="toggle-switch active">
+                                <div class="toggle-slider"></div>
+                            </div>
+                        </div>
+                        <div class="setting-item">
+                            <span class="setting-label">번역 언어</span>
                             <select id="language-select">
+                                <option value="">자동 감지</option>
                                 <option value="ko">한국어 (Korean)</option>
                                 <option value="en">English</option>
                                 <option value="es">Español (Spanish)</option>
                     </div>
                     <div class="text-input-section">
                         <label for="system-prompt" class="setting-label">시스템 프롬프트:</label>
+                        <textarea id="system-prompt" placeholder="AI 어시스턴트의 성격, 역할, 행동 방식을 정의하세요...">당신은 친절하고 도움이 되는 AI 어시스턴트입니다. 사용자의 요청에 정확하고 유용한 답변을 제공합니다.</textarea>
                     </div>
                 </div>
         let peerConnection;
         let webrtc_id;
         let webSearchEnabled = false;
+        let autoLanguageDetection = true;
         let selectedLanguage = "";
+        let systemPrompt = "당신은 친절하고 도움이 되는 AI 어시스턴트입니다. 사용자의 요청에 정확하고 유용한 답변을 제공합니다.";
+        let currentConversationId = null;
         const audioOutput = document.getElementById('audio-output');
         const startButton = document.getElementById('start-button');
         const sendButton = document.getElementById('send-button');
         const statusDot = document.getElementById('status-dot');
         const statusText = document.getElementById('status-text');
         const searchToggle = document.getElementById('search-toggle');
+        const autoLangToggle = document.getElementById('auto-lang-toggle');
         const languageSelect = document.getElementById('language-select');
         const systemPromptInput = document.getElementById('system-prompt');
         const textInput = document.getElementById('text-input');
+        const historyList = document.getElementById('history-list');
         let audioLevel = 0;
         let animationFrame;
         let audioContext, analyser, audioSource;
         let dataChannel = null;
         let isVoiceActive = false;
+        // Initialize
+        window.addEventListener('DOMContentLoaded', async () => {
+            sendButton.style.display = 'block';
+            await loadHistory();
+            startNewChat();
+        });
+        // Start new chat
+        function startNewChat() {
+            currentConversationId = generateUUID();
+            chatMessages.innerHTML = '';
+            console.log('Started new conversation:', currentConversationId);
+        }
+        // Generate UUID
+        function generateUUID() {
+            return 'xxxx-xxxx-4xxx-yxxx-xxxx'.replace(/[xy]/g, function(c) {
+                const r = Math.random() * 16 | 0;
+                const v = c === 'x' ? r : (r & 0x3 | 0x8);
+                return v.toString(16);
+            });
+        }
+        // Load conversation history
+        async function loadHistory() {
+            try {
+                const response = await fetch('/conversations');
+                const conversations = await response.json();
+                historyList.innerHTML = '';
+                conversations.forEach(conv => {
+                    const item = document.createElement('div');
+                    item.className = 'history-item';
+                    item.onclick = () => loadConversation(conv.id);
+                    const title = document.createElement('div');
+                    title.className = 'history-item-title';
+                    title.textContent = conv.title;
+                    const date = document.createElement('div');
+                    date.className = 'history-item-date';
+                    date.textContent = new Date(conv.updated_at).toLocaleDateString('ko-KR');
+                    item.appendChild(title);
+                    item.appendChild(date);
+                    historyList.appendChild(item);
+                });
+            } catch (error) {
+                console.error('Failed to load history:', error);
+            }
+        }
+        // Load a specific conversation
+        async function loadConversation(conversationId) {
+            try {
+                const response = await fetch(`/conversation/${conversationId}`);
+                const messages = await response.json();
+                currentConversationId = conversationId;
+                chatMessages.innerHTML = '';
+                messages.forEach(msg => {
+                    addMessage(msg.role, msg.content, false);
+                });
+            } catch (error) {
+                console.error('Failed to load conversation:', error);
+            }
+        }
         // Web search toggle functionality
         searchToggle.addEventListener('click', () => {
             webSearchEnabled = !webSearchEnabled;
             console.log('Web search enabled:', webSearchEnabled);
         });
+        // Auto language detection toggle
+        autoLangToggle.addEventListener('click', () => {
+            autoLanguageDetection = !autoLanguageDetection;
+            autoLangToggle.classList.toggle('active', autoLanguageDetection);
+            languageSelect.disabled = autoLanguageDetection;
+            console.log('Auto language detection:', autoLanguageDetection);
+        });
         // Language selection
         languageSelect.addEventListener('change', () => {
             selectedLanguage = languageSelect.value;
         // System prompt update
         systemPromptInput.addEventListener('input', () => {
+            systemPrompt = systemPromptInput.value || "당신은 친절하고 도움이 되는 AI 어시스턴트입니다.";
         });
         // Text input handling
             if (!message) return;
             // Add user message to chat
+            addMessage('user', message, true);
             textInput.value = '';
             // Show sending indicator
                     body: JSON.stringify({
                         message: message,
                         web_search_enabled: webSearchEnabled,
+                        target_language: autoLanguageDetection ? '' : selectedLanguage,
+                        system_prompt: systemPrompt,
+                        conversation_id: currentConversationId,
+                        auto_detect: autoLanguageDetection
                     })
                 });
                 } else {
                     // Add assistant response
                     let content = data.response;
+                    if (data.language) {
                         content += ` <span class="language-info">[${data.language}]</span>`;
                     }
+                    addMessage('assistant', content, true);
+                    // Refresh history
+                    await loadHistory();
                 }
             } catch (error) {
                 console.error('Error sending text message:', error);
                 sendButton.style.display = 'none';
             } else {
                 statusText.textContent = '연결 대기 중';
+                sendButton.style.display = 'block';
                 isVoiceActive = false;
             }
         }
         function updateButtonState() {
             const button = document.getElementById('start-button');
             if (peerConnection && (peerConnection.connectionState === 'connecting' || peerConnection.connectionState === 'new')) {
                 updateStatus('disconnected');
             }
         }
         function setupAudioVisualization(stream) {
             audioContext = new (window.AudioContext || window.webkitAudioContext)();
             analyser = audioContext.createAnalyser();
             updateAudioLevel();
         }
         function showError(message) {
             const toast = document.getElementById('error-toast');
             toast.textContent = message;
                 toast.style.display = 'none';
             }, 5000);
         }
         async function setupWebRTC() {
             const config = __RTC_CONFIGURATION__;
             peerConnection = new RTCPeerConnection(config);
                     toast.style.display = 'none';
                 }, 5000);
             }, 5000);
             try {
                 const stream = await navigator.mediaDevices.getUserMedia({
                     audio: true
                         peerConnection.addEventListener("icegatheringstatechange", checkState);
                     }
                 });
                 peerConnection.addEventListener('connectionstatechange', () => {
                     console.log('connectionstatechange', peerConnection.connectionState);
                     if (peerConnection.connectionState === 'connected') {
                     }
                     updateButtonState();
                 });
+                webrtc_id = Math.random().toString(36).substring(7);
                 const response = await fetch('/webrtc/offer', {
                     method: 'POST',
                         type: peerConnection.localDescription.type,
                         webrtc_id: webrtc_id,
                         web_search_enabled: webSearchEnabled,
+                        target_language: autoLanguageDetection ? '' : selectedLanguage,
+                        system_prompt: systemPrompt,
+                        conversation_id: currentConversationId,
+                        auto_detect: autoLanguageDetection
                     })
                 });
                 const serverResponse = await response.json();
                 if (serverResponse.status === 'failed') {
                     showError(serverResponse.meta.error === 'concurrency_limit_reached'
                     stop();
                     return;
                 }
                 await peerConnection.setRemoteDescription(serverResponse);
                 const eventSource = new EventSource('/outputs?webrtc_id=' + webrtc_id);
+                eventSource.addEventListener("output", async (event) => {
                     const eventJson = JSON.parse(event.data);
                     let content = eventJson.content;
+                    if (eventJson.language) {
                         content += ` <span class="language-info">[${eventJson.language}]</span>`;
                     }
+                    addMessage("assistant", content, true);
+                    // Refresh history after receiving a response
+                    await loadHistory();
                 });
                 eventSource.addEventListener("search", (event) => {
                     const eventJson = JSON.parse(event.data);
                     if (eventJson.query) {
+                        addMessage("search-result", `웹 검색 중: "${eventJson.query}"`, true);
                     }
                 });
             } catch (err) {
                 stop();
             }
         }
+        function addMessage(role, content, save = false) {
             const messageDiv = document.createElement('div');
             messageDiv.classList.add('message', role);
             }
             chatMessages.appendChild(messageDiv);
             chatMessages.scrollTop = chatMessages.scrollHeight;
+            // Save to database if needed
+            if (save && currentConversationId && role !== 'search-result') {
+                fetch('/message', {
+                    method: 'POST',
+                    headers: { 'Content-Type': 'application/json' },
+                    body: JSON.stringify({
+                        conversation_id: currentConversationId,
+                        role: role,
+                        content: content.replace(/<[^>]*>/g, ''), // Remove HTML tags
+                        language: ''
+                    })
+                }).catch(error => console.error('Failed to save message:', error));
+            }
         }
         function stop() {
             console.log('[STOP] Stopping connection...');
             if (animationFrame) {
                 cancelAnimationFrame(animationFrame);
                 animationFrame = null;
             }
             if (audioContext) {
                 audioContext.close();
                 audioContext = null;
                 audioSource = null;
             }
             if (dataChannel) {
                 dataChannel.close();
                 dataChannel = null;
             }
             if (peerConnection) {
                 console.log('[STOP] Current connection state:', peerConnection.connectionState);
                 if (peerConnection.getTransceivers) {
                     peerConnection.getTransceivers().forEach(transceiver => {
                         if (transceiver.stop) {
                     });
                 }
                 if (peerConnection.getSenders) {
                     peerConnection.getSenders().forEach(sender => {
                         if (sender.track) {
                     });
                 }
                 if (peerConnection.getReceivers) {
                     peerConnection.getReceivers().forEach(receiver => {
                         if (receiver.track) {
                     });
                 }
                 peerConnection.close();
                 peerConnection = null;
                 console.log('[STOP] Connection closed');
             }
             audioLevel = 0;
             isVoiceActive = false;
             updateButtonState();
             if (webrtc_id) {
                 console.log('[STOP] Clearing webrtc_id:', webrtc_id);
                 webrtc_id = null;
             }
         }
         startButton.addEventListener('click', () => {
             console.log('clicked');
             console.log(peerConnection, peerConnection?.connectionState);
                 stop();
             }
         });
     </script>
 </body>
 # Initialize search client globally
 brave_api_key = os.getenv("BSEARCH_API")
 search_client = BraveSearchClient(brave_api_key) if brave_api_key else None
 # Store connection settings
 connection_settings = {}
 async def process_text_chat(message: str, web_search_enabled: bool, target_language: str,
+                          system_prompt: str, conversation_id: str, auto_detect: bool) -> Dict[str, str]:
     """Process text chat using GPT-4o-mini model"""
     try:
+        # Auto-detect language if enabled
+        if auto_detect:
+            detected_lang = detect_language(message)
+            if detected_lang in ['en', 'ko']:
+                target_language = detected_lang
             else:
+                target_language = 'ko'  # Default to Korean
+        # Save user message
+        await save_message(conversation_id, "user", message, target_language)
+        # Prepare system prompt based on language
+        if target_language == 'en':
+            base_instructions = f"You are a helpful assistant. {system_prompt}"
         else:
+            base_instructions = system_prompt or "당신은 친절하고 도움이 되는 AI 어시스턴트입니다."
         messages = [
             {"role": "system", "content": base_instructions}
         # Handle web search if enabled
         if web_search_enabled and search_client:
             search_keywords = ["날씨", "기온", "비", "눈", "뉴스", "소식", "현재", "최근",
                              "오늘", "지금", "가격", "환율", "주가", "weather", "news",
                              "current", "today", "price", "2024", "2025"]
             should_search = any(keyword in message.lower() for keyword in search_keywords)
             if should_search:
                 search_results = await search_client.search(message)
                 if search_results:
                     search_context = "웹 검색 결과:\n\n"
                     for i, result in enumerate(search_results[:5], 1):
                         search_context += f"{i}. {result['title']}\n{result['description']}\n\n"
                     messages.append({
                         "role": "system",
+                        "content": "다음 웹 검색 결과를 참고하여 답변하세요: \n\n" + search_context
                     })
+        messages.append({"role": "user", "content": message})
         # Call GPT-4o-mini
         response = await client.chat.completions.create(
         response_text = response.choices[0].message.content
+        # Save assistant response
+        await save_message(conversation_id, "assistant", response_text, target_language)
         return {
             "response": response_text,
 class OpenAIHandler(AsyncStreamHandler):
     def __init__(self, web_search_enabled: bool = False, target_language: str = "",
+                 system_prompt: str = "", webrtc_id: str = None, conversation_id: str = None,
+                 auto_detect: bool = True) -> None:
         super().__init__(
             expected_layout="mono",
             output_sample_rate=SAMPLE_RATE,
         self.web_search_enabled = web_search_enabled
         self.target_language = target_language
         self.system_prompt = system_prompt
+        self.conversation_id = conversation_id
+        self.auto_detect = auto_detect
     def copy(self):
         if connection_settings:
             recent_ids = sorted(connection_settings.keys(),
                               key=lambda k: connection_settings[k].get('timestamp', 0),
                               reverse=True)
                 recent_id = recent_ids[0]
                 settings = connection_settings[recent_id]
                 return OpenAIHandler(
                     web_search_enabled=settings.get('web_search_enabled', False),
                     target_language=settings.get('target_language', ''),
                     system_prompt=settings.get('system_prompt', ''),
+                    webrtc_id=recent_id,
+                    conversation_id=settings.get('conversation_id'),
+                    auto_detect=settings.get('auto_detect', True)
                 )
         return OpenAIHandler(web_search_enabled=False)
     async def search_web(self, query: str) -> str:
         if not results:
             return f"'{query}'에 대한 검색 결과를 찾을 수 없습니다."
         formatted_results = []
         for i, result in enumerate(results, 1):
             formatted_results.append(
             )
             await self.connection.response.create()
     async def start_up(self):
         """Connect to realtime API"""
         if connection_settings and self.webrtc_id:
             if self.webrtc_id in connection_settings:
                 settings = connection_settings[self.webrtc_id]
                 self.web_search_enabled = settings.get('web_search_enabled', False)
                 self.target_language = settings.get('target_language', '')
                 self.system_prompt = settings.get('system_prompt', '')
+                self.conversation_id = settings.get('conversation_id')
+                self.auto_detect = settings.get('auto_detect', True)
         self.client = openai.AsyncOpenAI()
         tools = []
+        base_instructions = self.system_prompt or "당신은 친절하고 도움이 되는 AI 어시스턴트입니다."
         if self.web_search_enabled and self.search_client:
             tools = [{
                 "type": "function",
                 "function": {
                     "name": "web_search",
+                    "description": "Search the web for current information.",
                     "parameters": {
                         "type": "object",
                         "properties": {
                     }
                 }
             }]
         async with self.client.beta.realtime.connect(
             model="gpt-4o-mini-realtime-preview-2024-12-17"
         ) as conn:
             session_update = {
                 "turn_detection": {"type": "server_vad"},
+                "instructions": base_instructions,
                 "tools": tools,
                 "tool_choice": "auto" if tools else "none",
                 "temperature": 0.7,
                 "max_response_output_tokens": 4096,
                 "modalities": ["text", "audio"],
+                "voice": "nova"
             }
             await conn.session.update(session=session_update)
             self.connection = conn
             async for event in self.connection:
                 if event.type == "response.audio_transcript.done":
+                    # Save the transcript
+                    if self.conversation_id:
+                        await save_message(self.conversation_id, "assistant", event.transcript)
                     output_data = {
                         "event": event,
                 # Handle function calls
                 elif event.type == "response.function_call_arguments.start":
                     self.function_call_in_progress = True
                     self.current_function_args = ""
                     self.current_call_id = getattr(event, 'call_id', None)
                 elif event.type == "response.function_call_arguments.done":
                     if self.function_call_in_progress:
                         try:
                             args = json.loads(self.current_function_args)
                             query = args.get("query", "")
                             await self.output_queue.put(AdditionalOutputs({
                                 "type": "search",
                                 "query": query
                             }))
                             search_results = await self.search_web(query)
                             if self.connection and self.current_call_id:
                                 await self.connection.conversation.item.create(
                                     item={
                             self.current_call_id = None
     async def receive(self, frame: tuple[int, np.ndarray]) -> None:
         if not self.connection:
             return
         try:
             _, array = frame
             print(f"Error in receive: {e}")
     async def emit(self) -> tuple[int, np.ndarray] | AdditionalOutputs | None:
         item = await wait_for_item(self.output_queue)
         if isinstance(item, dict) and item.get('type') == 'text_message':
             await self.process_text_message(item['content'])
             return None
         return item
     async def shutdown(self) -> None:
         if self.connection:
             await self.connection.close()
             self.connection = None
 # Create initial handler instance
 # Create stream with handler instance
 stream = Stream(
+    handler,
     mode="send-receive",
     modality="audio",
     additional_inputs=[chatbot],
 app = FastAPI()
+# Initialize database on startup
+@app.on_event("startup")
+async def startup_event():
+    await init_db()
 # Mount stream
 stream.mount(app)
     web_search_enabled = body.get("web_search_enabled", False)
     target_language = body.get("target_language", "")
     system_prompt = body.get("system_prompt", "")
+    conversation_id = body.get("conversation_id")
+    auto_detect = body.get("auto_detect", True)
     if webrtc_id:
         connection_settings[webrtc_id] = {
             'web_search_enabled': web_search_enabled,
             'target_language': target_language,
             'system_prompt': system_prompt,
+            'conversation_id': conversation_id,
+            'auto_detect': auto_detect,
             'timestamp': asyncio.get_event_loop().time()
         }
     # Remove our custom route temporarily
     custom_route = None
             custom_route = app.routes.pop(i)
             break
     response = await stream.offer(body)
     if custom_route:
         app.routes.insert(0, custom_route)
     return response
         web_search_enabled = body.get("web_search_enabled", False)
         target_language = body.get("target_language", "")
         system_prompt = body.get("system_prompt", "")
+        conversation_id = body.get("conversation_id")
+        auto_detect = body.get("auto_detect", True)
         if not message:
             return {"error": "메시지가 비어있습니다."}
+        result = await process_text_chat(message, web_search_enabled, target_language,
+                                       system_prompt, conversation_id, auto_detect)
         return result
         return {"error": "채팅 처리 중 오류가 발생했습니다."}
+@app.get("/conversations")
+async def get_conversations_endpoint():
+    """Get all conversations"""
+    conversations = await get_conversations()
+    return conversations
+@app.get("/conversation/{conversation_id}")
+async def get_conversation_endpoint(conversation_id: str):
+    """Get messages for a specific conversation"""
+    messages = await get_conversation_messages(conversation_id)
+    return messages
+@app.post("/message")
+async def save_message_endpoint(request: Request):
+    """Save a message"""
+    body = await request.json()
+    await save_message(
+        body["conversation_id"],
+        body["role"],
+        body["content"],
+        body.get("language")
+    )
+    return {"status": "ok"}
 @app.post("/text_message/{webrtc_id}")
 async def receive_text_message(webrtc_id: str, request: Request):
     """Receive text message from client"""
     body = await request.json()
     message = body.get("content", "")
     if webrtc_id in stream.handlers:
         handler = stream.handlers[webrtc_id]
         await handler.output_queue.put({
             'type': 'text_message',
             'content': message
     async def output_stream():
         async for output in stream.output_stream(webrtc_id):
             if hasattr(output, 'args') and output.args:
                 if isinstance(output.args[0], dict) and output.args[0].get('type') == 'search':
                     yield f"event: search\ndata: {json.dumps(output.args[0])}\n\n"
                 elif isinstance(output.args[0], dict) and 'event' in output.args[0]:
                     event_data = output.args[0]
                     if 'event' in event_data and hasattr(event_data['event'], 'transcript'):