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"))
 

qingjian_cs/services/service_gemini_stable.py

@@ -0,0 +1,122 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import asyncio
+import base64
+import logging
+import os
+from typing import Any
+
+import aiohttp
+
+logger = logging.getLogger(__name__)
+
+GEMINI_API_KEY = "sk-YOUR_NEW_KEY"
+GEMINI_API_URL = "https://api.laozhang.ai/v1beta/models"
+GEMINI_MODEL = "gemini-3.1-flash-image-preview"
+GEMINI_IMAGE_SIZE = "4K"
+GEMINI_MAX_RETRIES = 3
+
+DEFAULT_PROMPT = (
+    "提取印花图案，去褶皱并补齐缺失区域，生成完整清晰的平面图。"
+    "严格保持原图元素位置、颜色和细节，不要改风格。"
+)
+
+
+class GeminiExtractStableService:
+    def image_to_base64(self, image_path: str) -> str | None:
+        try:
+            if not os.path.exists(image_path):
+                logger.error("文件不存在: %s", image_path)
+                return None
+            with open(image_path, "rb") as f:
+                return base64.b64encode(f.read()).decode("utf-8")
+        except Exception as e:
+            logger.error("Base64转换失败: %s", e)
+            return None
+
+    async def extract_pattern(
+        self,
+        input_path: str,
+        output_path: str,
+        custom_prompt: str | None = None,
+        aspect_ratio: str = "1:1",
+    ) -> tuple[bool, str, dict[str, Any]]:
+        img64 = self.image_to_base64(input_path)
+        if not img64:
+            return False, "图片编码失败", {}
+
+        prompt = custom_prompt or DEFAULT_PROMPT
+        api_url = f"{GEMINI_API_URL}/{GEMINI_MODEL}:generateContent"
+        valid_ratios = {"1:1", "9:16", "16:9", "3:4", "4:3", "3:2", "2:3", "5:4", "4:5"}
+        image_config: dict[str, Any] = {"imageSize": GEMINI_IMAGE_SIZE}
+        if aspect_ratio in valid_ratios:
+            image_config["aspectRatio"] = aspect_ratio
+
+        headers = {
+            "Authorization": f"Bearer {GEMINI_API_KEY}",
+            "Content-Type": "application/json",
+        }
+        payload = {
+            "contents": [
+                {
+                    "parts": [
+                        {"text": prompt},
+                    ]
+                }
+            ],
+            "generationConfig": {
+                "responseModalities": ["IMAGE"],
+                "imageConfig": image_config,
+            },
+        }
+
+        logger.info("[稳定作图] model=%s ratio=%s size=%s", GEMINI_MODEL, image_config.get("aspectRatio", "1:1"), GEMINI_IMAGE_SIZE)
+        timeout = aiohttp.ClientTimeout(total=180, connect=30)
+        for i in range(1, GEMINI_MAX_RETRIES + 1):
+            try:
+                logger.info("[稳定作图] 请求中 第%s/%s次", i, GEMINI_MAX_RETRIES)
+                async with aiohttp.ClientSession(timeout=timeout) as session:
+                    async with session.post(api_url, headers=headers, json=payload) as resp:
+                        txt = await resp.text()
+                        if resp.status != 200:
+                            logger.error("[稳定作图] 请求失败 status=%s body=%s", resp.status, txt[:240])
+                            if i < GEMINI_MAX_RETRIES:
+                                await asyncio.sleep(i)
+                                continue
+                            return False, f"API失败:{resp.status}", {}
+                        result = await resp.json()
+
+                try:
+                    parts = result["candidates"][0]["content"]["parts"]
+                    b64 = ""
+                    for p in parts:
+                        if "inlineData" in p and isinstance(p["inlineData"], dict):
+                            b64 = str(p["inlineData"].get("data", "") or "")
+                            if b64:
+                                break
+                    if not b64:
+                        logger.error("[稳定作图] 未找到 inlineData")
+                        return False, "未找到图片数据", {}
+                    image_data = base64.b64decode(b64)
+                except Exception as e:
+                    logger.error("[稳定作图] 解析返回失败: %s", e)
+                    return False, f"解析失败:{e}", {}
+
+                out_dir = os.path.dirname(output_path)
+                if out_dir:
+                    os.makedirs(out_dir, exist_ok=True)
+                with open(output_path, "wb") as f:
+                    f.write(image_data)
+                logger.info("[稳定作图] 输出成功: %s", output_path)
+                return True, "ok", {"output_path": output_path, "file_size": len(image_data), "api_used": "gemini_stable"}
+            except Exception as e:
+                logger.error("[稳定作图] 异常: %s", e)
+                if i < GEMINI_MAX_RETRIES:
+                    await asyncio.sleep(i)
+                    continue
+                return False, f"异常:{e}", {}
+        return False, "未知失败", {}
+
+    async def cleanup(self) -> None:
+        return None

qingjian_cs/services/service_tuhui_upload.py

@@ -7,7 +7,7 @@ from urllib.parse import urljoin
 import requests
 
 
-TUHUI_BASE_URL = os.getenv("TUHUI_BASE_URL", "http://127.0.0.1:8002").strip()
+TUHUI_BASE_URL = os.getenv("TUHUI_BASE_URL", "https://tuhui.cloud").strip()
 TUHUI_PHONE = os.getenv("TUHUI_PHONE", "17520145271").strip()
 TUHUI_PASSWORD = os.getenv("TUHUI_PASSWORD", "zuowei1216").strip()
 TUHUI_DEFAULT_PRICE = int(os.getenv("TUHUI_DEFAULT_PRICE", "20"))
@@ -104,4 +104,3 @@ async def upload_to_tuhui(
         price,
         category,
     )
-