feat(telegram): enable interactive chat with Nemo-4B context

apps/api/src/services/chat_manager.py

@@ -0,0 +1,117 @@
+"""
+AWOOOI Chat Manager - 統帥對話核心
+===================================
+Phase 21.5: 實作 Telegram 互動對話功能
+
+職責:
+1. 整合系統上下文 (K3s 狀態, 最近告警, 目前時間)
+2. 決定對話風格 (OpenClaw 專業風 vs Nemo 參謀風)
+3. 調用 LLM (Nemo-4B / Gemini) 產出回應
+4. 遵守 SOUL.md Nothing.tech 純淨美學
+
+2026-03-31 ogt: 初版建立
+"""
+
+import structlog
+from datetime import datetime
+from src.utils.timezone import now_taipei
+from src.services.nvidia_provider import get_nvidia_provider
+from src.repositories.k8s_repository import get_k8s_repository
+from src.repositories.incident_repository import get_incident_repository
+
+logger = structlog.get_logger(__name__)
+
+class ChatManager:
+    """
+    AWOOOI 對話管理器 - 系統的大腦與聲帶
+    """
+
+    def __init__(self):
+        self.nvidia = get_nvidia_provider()
+        self.k8s = get_k8s_repository()
+        self.incidents = get_incident_repository()
+
+    async def get_system_context(self) -> str:
+        """
+        收集系統即時上下文，供 LLM 參考
+        """
+        now = now_taipei()
+        
+        # 1. K3s 狀態
+        k8s_status = await self.k8s.get_pod_status_summary(namespace="awoooi-prod")
+        cluster_info = f"Cluster: {k8s_status['running']}/{k8s_status['total']} Pods Running"
+        if k8s_status['problem_pods']:
+            cluster_info += f", {len(k8s_status['problem_pods'])} anomalies detected."
+
+        # 2. 最近告警 (取 3 筆)
+        active_incidents = await self.incidents.get_active()
+        incident_summary = "None"
+        if active_incidents:
+            lines = []
+            for inc in active_incidents[:3]:
+                lines.append(f"- {inc.incident_id}: {inc.status.value} (Severity: {inc.severity.value})")
+            incident_summary = "\n".join(lines)
+
+        context = f"""
+## Current System Context (Taipei Time: {now.strftime('%Y-%m-%d %H:%M:%S')})
+- Environment: AWOOOI Production (K3s)
+- {cluster_info}
+- Active Incidents:
+{incident_summary}
+"""
+        return context
+
+    async def generate_response(
+        self, 
+        user_id: int, 
+        username: str, 
+        message_text: str
+    ) -> str:
+        """
+        根據統帥訊息產生回覆
+        """
+        system_context = await self.get_system_context()
+        
+        # 判定是否在跟 Nemo 對話
+        is_asking_nemo = "nemo" in message_text.lower()
+        
+        role_description = "You are OpenClaw, the AI operations assistant for AWOOOI platform."
+        if is_asking_nemo:
+            role_description = "You are Nemo-4B, the elite AI tactical advisor for AWOOOI. Address the user as 'Supreme Commander' (統帥)."
+
+        system_prompt = f"""{role_description}
+{system_context}
+
+## Guidelines:
+1. Keep responses extremely concise and professional (Nothing.tech aesthetic).
+2. For status queries, provide precise data.
+3. For general chat, be supportive but focused on operations.
+4. Language: Preferred Traditional Chinese (繁體中文).
+5. No emojis except for functional ones (🚨, ✅, 📊).
+"""
+
+        try:
+            # 優先使用 NVIDIA Nemo-4B
+            response, provider, success, tokens, cost = await self.nvidia.chat(
+                prompt=f"{system_prompt}\n\nCommander's Message: {message_text}",
+                model="nvidia/nemotron-mini-4b-instruct",
+                max_tokens=1024
+            )
+            
+            if not success:
+                return "🛸 抱歉統帥，Nemo 參謀暫時離線。請稍後再試。"
+
+            return response.strip()
+
+        except Exception as e:
+            logger.exception("chat_generation_error", error=str(e))
+            return "⚠️ 通訊鏈路異常，無法聯繫 AI 腦區。"
+
+# Singleton
+_chat_manager = None
+
+def get_chat_manager() -> ChatManager:
+    global _chat_manager
+    if _chat_manager is None:
+        _chat_manager = ChatManager()
+    return _chat_manager