feat: AI assistant phase 2 - configurable prompt, action operations, FAQ matching, NVIDIA provider

- Admin-configurable AI prompt/quick questions from system_configs DB - GET /api/v1/ai/quick-questions endpoint for fetching quick questions - Local FAQ matching for instant responses (avoid AI calls for common Qs) - AI action extraction: "add customer" intent detected, structured data returned - Frontend action confirmation card with editable fields, calls customer API on confirm - NVIDIA provider (stepfun-ai/step-3.5-flash) for faster chat vs deepseek-v4-flash - Fixed httpx client timeout preventing backend hangs - Added log_usage calls for auth events (register/login/guest/wechat) - Admin tabs (users/stats/logs/config) fully functional with real backend - AiAssistant component added to all tabbar pages
2026-05-20 09:39:22 +08:00
parent 4755cc75ba
commit f8a23855d2
20 changed files with 744 additions and 5 deletions
@@ -0,0 +1,50 @@
+from typing import Dict, Any, Optional, List
+from app.ai.providers.openai import OpenAIProvider, SYSTEM_PROMPTS
+import logging
+import time
+import httpx
+
+logger = logging.getLogger(__name__)
+
+
+class NvidiaProvider(OpenAIProvider):
+    def __init__(self, api_key: str, model: str = "stepfun-ai/step-3.5-flash", base_url: str = "https://integrate.api.nvidia.com/v1"):
+        super().__init__(
+            api_key=api_key,
+            model=model,
+            base_url=base_url,
+            http_client=httpx.AsyncClient(timeout=httpx.Timeout(60.0)),
+        )
+        self._name = f"nvidia-{model}"
+
+    async def chat(self, message: str, history: list = None, system_prompt: str = None) -> Dict[str, Any]:
+        t0 = time.time()
+
+        system = system_prompt or SYSTEM_PROMPTS["chat"]
+        messages = [{"role": "system", "content": system}]
+        if history:
+            for h in history[-10:]:
+                messages.append(h)
+        messages.append({"role": "user", "content": message})
+        t1 = time.time()
+
+        kwargs = {
+            "model": self.model,
+            "messages": messages,
+            "max_tokens": 300,
+            "temperature": 0.3,
+        }
+        resp = await self.client.chat.completions.create(**kwargs)
+        t2 = time.time()
+
+        content = resp.choices[0].message.content or ""
+        if not content and hasattr(resp.choices[0].message, "reasoning"):
+            content = resp.choices[0].message.reasoning
+        t3 = time.time()
+
+        logger.info(
+            f"NVIDIA timing: build_msgs={t1-t0:.1f}s api_call={t2-t1:.1f}s process={t3-t2:.1f}s "
+            f"chars_in={sum(len(m.get('content','')) for m in messages)} chars_out={len(content)}"
+        )
+
+        return {"reply": content, "provider": self.name, "model": self.model}