落地私聊图片询价：识图当口播复用三线文字流程，图上航线三字码保持原样，避免被改写成中文地名。

Co-authored-by: Cursor <cursoragent@cursor.com>

inquiry-agent/agent/llm/http_client.py

@@ -19,6 +19,42 @@ from agent.llm.modes import LlmMode
 logger = logging.getLogger(__name__)
 
 
+def message_plain_text(msg: dict[str, Any]) -> str:
+    """
+    把 assistant message 收成纯文本。
+
+    千问视觉的 content 经常是 [{text: ...}] 数组，不是字符串；
+    思考模式打开时正文可能只在 reasoning_content。两者都漏会让识图变「没看清」。
+    """
+    if not isinstance(msg, dict):
+        return ""
+    text = _content_to_text(msg.get("content"))
+    if text:
+        return text
+    reason = msg.get("reasoning_content")
+    if isinstance(reason, str) and reason.strip():
+        return reason.strip()
+    return ""
+
+
+def _content_to_text(content: Any) -> str:
+    """字符串或千问多模态数组 → 纯文本。"""
+    if isinstance(content, str):
+        return content.strip()
+    if isinstance(content, list):
+        parts: list[str] = []
+        for item in content:
+            if isinstance(item, str) and item.strip():
+                parts.append(item.strip())
+            elif isinstance(item, dict):
+                piece = item.get("text") if "text" in item else item.get("content")
+                got = _content_to_text(piece)
+                if got:
+                    parts.append(got)
+        return "\n".join(parts).strip()
+    return ""
+
+
 @dataclass
 class LlmHttpResult:
     ok: bool
@@ -72,7 +108,7 @@ class LlmHttpClient:
         mode_val = mode.value if isinstance(mode, LlmMode) else str(mode)
         if provider == "qwen":
             if mode_val == LlmMode.VISION.value:
-                return self.settings.qwen_vl_model or "qwen-vl-plus"
+                return self.settings.qwen_vl_model or "qwen3-vl-plus"
             return self.settings.qwen_text_model or "qwen-plus"
         return self.settings.deepseek_model or "deepseek-chat"
 
@@ -115,6 +151,9 @@ class LlmHttpClient:
             "messages": messages,
             "temperature": temperature,
         }
+        if provider == "qwen":
+            # 识图只要可见原文。思考开着时 content 常为空，销售侧就会「认不到」。
+            body["enable_thinking"] = False
         if tools:
             body["tools"] = tools
         if tool_choice is not None:
@@ -147,7 +186,7 @@ class LlmHttpClient:
             tool_calls: list[dict[str, Any]] = []
             if choices:
                 msg = (choices[0].get("message") or {})
-                content = str(msg.get("content") or "")
+                content = message_plain_text(msg if isinstance(msg, dict) else {})
                 raw_calls = msg.get("tool_calls") or []
                 if isinstance(raw_calls, list):
                     tool_calls = [c for c in raw_calls if isinstance(c, dict)]