diff --git a/apps/api/src/services/chat_manager.py b/apps/api/src/services/chat_manager.py
index 11130726..7cf483cc 100644
--- a/apps/api/src/services/chat_manager.py
+++ b/apps/api/src/services/chat_manager.py
@@ -144,23 +144,39 @@ class ChatManager:
async def _call_nemotron(self, system_prompt: str, user_message: str) -> str | None:
"""
- 呼叫 NemoClaw 對話 — NVIDIA NIM nemotron-mini-4b
+ 呼叫 NemoClaw 對話 — Claude API (claude-haiku-4-5)
- 2026-04-03 ogt: Ollama 188 主機負載高常超時，暫時維持 NIM
- 老闆可接受慢速，timeout=120s
+ 2026-04-03 ogt: 老闆指示改接 Claude API，快速且中文能力強
+ 用 haiku 降低費用
"""
- from src.services.nvidia_provider import get_nvidia_provider
- nvidia = get_nvidia_provider()
- try:
- full_prompt = f"{system_prompt}\n\n用戶訊息: {user_message}"
- response, success, _, _ = await nvidia.chat(
- prompt=full_prompt,
- model="nvidia/nemotron-mini-4b-instruct",
- max_tokens=300,
- )
- if success and response and "not configured" not in response and "Circuit Breaker" not in response:
- return response.strip()
+ import httpx
+ from src.core.config import get_settings
+ settings = get_settings()
+
+ api_key = getattr(settings, 'CLAUDE_API_KEY', None)
+ if not api_key:
+ logger.warning("nemotron_chat_failed", error="CLAUDE_API_KEY not configured")
return None
+
+ try:
+ async with httpx.AsyncClient(timeout=30.0) as client:
+ resp = await client.post(
+ "https://api.anthropic.com/v1/messages",
+ headers={
+ "x-api-key": api_key,
+ "anthropic-version": "2023-06-01",
+ "content-type": "application/json",
+ },
+ json={
+ "model": "claude-haiku-4-5-20251001",
+ "max_tokens": 300,
+ "system": system_prompt,
+ "messages": [{"role": "user", "content": user_message}],
+ },
+ )
+ resp.raise_for_status()
+ data = resp.json()
+ return data["content"][0]["text"].strip()
except Exception as e:
logger.warning("nemotron_chat_failed", error=str(e))
return None
diff --git a/apps/api/src/services/telegram_gateway.py b/apps/api/src/services/telegram_gateway.py
index 89ebaefd..aea278a9 100644
--- a/apps/api/src/services/telegram_gateway.py
+++ b/apps/api/src/services/telegram_gateway.py
@@ -2684,36 +2684,19 @@ class TelegramGateway:
try:
chat_mgr = ChatManager()
+ # 2026-04-03 ogt: 老闆指示 — 告警分析只由 OpenClaw 負責，NemoClaw 不分析告警
openclaw_prompt = (
f"你是 OpenClaw，AWOOOI SRE 戰情室首席 AI，精通 K8s、Prometheus、告警分析。\n"
- f"以下是一則基礎設施告警，請進行 RCA 根因分析並給出 3 點具體建議行動：\n\n"
- f"{alert_summary}"
- )
- nemo_prompt = (
- f"你是 NemoClaw，AWOOOI SRE 戰情室 AI 參謀，由 NVIDIA Nemotron 驅動。\n"
- f"以下是一則基礎設施告警，請從風險評估角度分析，指出潛在的連鎖失效點：\n\n"
+ f"以下是一則基礎設施告警，請進行 RCA 根因分析並給出 3 點具體建議行動。\n"
+ f"繁體中文回應，不超過 300 字：\n\n"
f"{alert_summary}"
)
- # 2026-04-03 ogt: 並行執行，兩個 AI 同時分析，不互相影響（更客觀）
- openclaw_task = asyncio.create_task(
- chat_mgr._call_openclaw(
- system_prompt="你是 OpenClaw，AWOOOI SRE 戰情室首席 AI。",
- user_message=openclaw_prompt,
- )
- )
- nemo_task = asyncio.create_task(
- chat_mgr._call_nemotron(
- system_prompt="你是 NemoClaw，AWOOOI SRE 戰情室 AI 參謀。",
- user_message=nemo_prompt,
- )
+ openclaw_analysis = await chat_mgr._call_openclaw(
+ system_prompt="你是 OpenClaw，AWOOOI SRE 戰情室首席 AI。稱呼用戶為「老闆」。",
+ user_message=openclaw_prompt,
)
- openclaw_analysis, nemo_analysis = await asyncio.gather(
- openclaw_task, nemo_task, return_exceptions=True
- )
-
- # OpenClaw 發到群組
if openclaw_analysis and not isinstance(openclaw_analysis, Exception):
await self.send_as_openclaw(
text=f"🦞 OpenClaw 分析\n\n{openclaw_analysis}",
@@ -2723,16 +2706,6 @@ class TelegramGateway:
else:
logger.warning("trigger_group_ai_discussion_openclaw_empty")
- # NemoClaw 發到群組（同樣 reply 告警訊息，並排出現）
- if nemo_analysis and not isinstance(nemo_analysis, Exception):
- await self.send_as_nemotron(
- text=f"🤖 NemoClaw 分析\n\n{nemo_analysis}",
- reply_to_message_id=alert_message_id,
- )
- logger.info("group_ai_discussion_nemo_sent")
- else:
- logger.warning("trigger_group_ai_discussion_nemo_empty")
-
logger.info("group_ai_discussion_completed", alert_message_id=alert_message_id)
except Exception as e: