feat: low-latency debounce, context logs, and stable draw/upload config

qingjian_cs/app/auto_draw.py

@@ -26,7 +26,7 @@ async def auto_draw_preview(
     """
     try:
         logger.info("[作图] 开始 customer=%s image=%s", customer_id, image_url)
-        from services.service_gemini import GeminiExtractV2Service  # type: ignore
+        from services.service_gemini_stable import GeminiExtractStableService  # type: ignore
         from services.service_tuhui_upload import upload_to_tuhui  # type: ignore
     except Exception as e:
         logger.error("[作图] 依赖加载失败: %s", e)
@@ -56,7 +56,7 @@ async def auto_draw_preview(
         logger.info("[作图] 原图下载完成 size=%s", len(resp.content))
 
         logger.info("[作图] Gemini 生成中")
-        service = GeminiExtractV2Service()
+        service = GeminiExtractStableService()
         ok_extract, msg_extract, _ = await service.extract_pattern(
             input_path=input_path,
             output_path=output_path,

qingjian_cs/app/client.py

@@ -14,6 +14,7 @@ from .config import (
     AUTO_DRAW_ENABLED,
     AUTO_QUOTE_WAIT_SECONDS,
     DECISION_TIMEOUT_SECONDS,
+    IMAGE_MESSAGE_DEBOUNCE_SECONDS,
     MAX_CONCURRENT_TURNS,
     MESSAGE_DEBOUNCE_SECONDS,
     QINGJIAN_WS_URI,
@@ -133,7 +134,7 @@ class QingjianClient:
 
     def _debounce_seconds(self, msg: str) -> float:
         if extract_image_urls(msg):
-            return 2.5
+            return float(IMAGE_MESSAGE_DEBOUNCE_SECONDS)
         return float(MESSAGE_DEBOUNCE_SECONDS)
 
     async def send_message(self, message: dict) -> None:
@@ -271,7 +272,7 @@ class QingjianClient:
     def _fallback_reply(self, action: str) -> str:
         if action == "transfer":
             return "我先给你转人工处理。"
-        return "收到，我先处理一下。"
+        return "我先看看"
 
     def _is_outbound_echo(self, data: dict, msg: str) -> bool:
         """
@@ -338,6 +339,20 @@ class QingjianClient:
             "last_reply": self.last_reply_key.get(key, ""),
             "recent_dialogue": recent_dialogue[-12:],
         }
+        try:
+            rd = context.get("recent_dialogue", []) or []
+            rd_preview = " | ".join(
+                [f"{str(x.get('role',''))}:{str(x.get('text',''))[:32]}" for x in rd[-8:] if isinstance(x, dict)]
+            )
+            self.logger.info(
+                "[AI上下文] customer=%s msg=%s pending_images=%s recent=%s",
+                context["customer_id"],
+                str(merged_msg or "")[:120],
+                context["pending_images"],
+                rd_preview,
+            )
+        except Exception:
+            pass
 
         activity_event(self.logger, "agent_process_start", trace_id=trace_id, customer_id=context["customer_id"], acc_id=context["acc_id"], intent=context["intent"])        
         route, decision, state = await self.orchestrator.decide(context)

qingjian_cs/app/config.py

@@ -12,9 +12,10 @@ OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "").strip()
 OPENAI_BASE_URL = os.getenv("OPENAI_BASE_URL", "https://ark.cn-beijing.volces.com/api/v3").strip()
 OPENAI_MODEL_NAME = os.getenv("OPENAI_MODEL_NAME", "doubao-seed-2-0-pro-260215").strip()
 
-MESSAGE_DEBOUNCE_SECONDS = int(os.getenv("MESSAGE_DEBOUNCE_SECONDS", "6"))
+MESSAGE_DEBOUNCE_SECONDS = float(os.getenv("MESSAGE_DEBOUNCE_SECONDS", "0.8"))
+IMAGE_MESSAGE_DEBOUNCE_SECONDS = float(os.getenv("IMAGE_MESSAGE_DEBOUNCE_SECONDS", "1.2"))
 AUTO_QUOTE_WAIT_SECONDS = int(os.getenv("AUTO_QUOTE_WAIT_SECONDS", "18"))
-AGENT_MAX_ITERS = int(os.getenv("AGENT_MAX_ITERS", "3"))
+AGENT_MAX_ITERS = int(os.getenv("AGENT_MAX_ITERS", "1"))
 FAST_ROUTE_ENABLED = os.getenv("FAST_ROUTE_ENABLED", "1").strip() in {"1", "true", "True", "yes", "on"}
 SHORT_REPLY_MAX_CHARS = int(os.getenv("SHORT_REPLY_MAX_CHARS", "18"))