diff --git "a/, i.get(script,)[:80])\n        for v in w.values(): walk(v)\n    elif isinstance(w, list):\n        for v in w: walk(v)\nwalk(d)\n\\\"\"" "b/, i.get(script,)[:80])\n        for v in w.values(): walk(v)\n    elif isinstance(w, list):\n        for v in w: walk(v)\nwalk(d)\n\\\"\""
new file mode 100644
index 0000000..e69de29
diff --git a/pipeline_service/agent_loop_v2.py b/pipeline_service/agent_loop_v2.py
index 7bb768f..8c16617 100644
--- a/pipeline_service/agent_loop_v2.py
+++ b/pipeline_service/agent_loop_v2.py
@@ -198,7 +198,14 @@ class AgentExecutor:
                 await self._maybe_compress()
 
             # 5b. LLM 调用（native function calling，返回 dict）
-            resp = await self._call_llm()
+            # 模型不可用等配置错误必须真实报给用户（error 事件），
+            # 绝不能让异常打断流——前端会停在"思考中..."干等。
+            try:
+                resp = await self._call_llm()
+            except Exception as e:
+                logger.error(f"run: LLM 调用失败 turn={turn+1}: {e}")
+                yield json.dumps({"type": "error", "message": str(e)}, ensure_ascii=False) + "\n"
+                return
 
             # 5c. 原生 function calling：优先处理 tool_calls
             native_calls = (resp.get("tool_calls") or []) if isinstance(resp, dict) else []
diff --git a/pipeline_service/llm_bridge.py b/pipeline_service/llm_bridge.py
index f333856..53ed366 100644
--- a/pipeline_service/llm_bridge.py
+++ b/pipeline_service/llm_bridge.py
@@ -50,10 +50,14 @@ async def _get_model_config(model_name: str = None, org_id: str = None) -> dict:
         dbname = "pipeline"
         async with db.sqlorContext(dbname) as sor:
             if model_name:
-                sql = "SELECT api_base, api_key, model_id FROM llm WHERE name=${name}$ AND status='active'"
+                # name/model_id 双匹配：与 gateway._resolve_and_persist_model 语义一致——
+                # 调用方传 llm.name 或 API 模型名（model_id）都能解析，避免语义漂移。
+                sql = ("SELECT api_base, api_key, model_id FROM llm "
+                       "WHERE status='active' AND (name=${name}$ OR model_id=${name}$)")
                 params = {"name": model_name}
                 if org_id:
-                    sql += " AND org_id=${org}$"
+                    # 与 gateway 一致：本机构模型 + 系统级共享（org_id 为空/0）
+                    sql += " AND (org_id=${org}$ OR org_id='' OR org_id='0')"
                     params["org"] = org_id
                 sql += " LIMIT 1"
                 recs = await sor.sqlExe(sql, params)
@@ -80,6 +84,24 @@ async def _get_model_config(model_name: str = None, org_id: str = None) -> dict:
     return {}
 
 
+def _no_llm_error(model_name=None, org_id=None) -> ValueError:
+    """模型不可用的错误必须真实可行动：写明模型名与原因，前端原样展示、运维据此处理。
+
+    禁止笼统的 'No LLM API configured'——用户看到它只会卡住干等。
+    """
+    if model_name:
+        return ValueError(
+            f"模型「{model_name}」不可用：llm 表中未找到可用配置"
+            f"（name/model_id 不匹配、status 非 active"
+            + (f"，或不属于当前机构 org={org_id}" if org_id else "")
+            + "）。请在模型下拉里改选可用模型，或联系管理员在模型管理中补齐配置。"
+        )
+    return ValueError(
+        "没有可用模型：当前会话未指定模型，且 llm 表无启用模型。"
+        "请在模型下拉中选择模型，或联系管理员配置模型。"
+    )
+
+
 # 瞬时错误重试：超时/连接错误/限流/服务端5xx 均重试；配置错误(无key)、鉴权/参数4xx 不重试
 _RETRYABLE_STATUS = (429, 500, 502, 503, 504)
 _LLM_MAX_ATTEMPTS = 3
@@ -168,7 +190,7 @@ async def llm_call(prompt: str, model: str = None, temperature: float = 0.7, org
         model_id = model or os.environ.get("LLM_MODEL", "gpt-4o-mini")
 
     if not api_key:
-        raise ValueError("No LLM API configured. Please add a model in the llm table or set LLM_API_KEY env var.")
+        raise _no_llm_error(model, org_id)
 
     import aiohttp
 
@@ -207,7 +229,7 @@ async def llm_call_msgs(messages: list, model: str = None, temperature: float =
         model_id = model or os.environ.get("LLM_MODEL", "gpt-4o-mini")
 
     if not api_key:
-        raise ValueError("No LLM API configured")
+        raise _no_llm_error(model, org_id)
 
     headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
     payload = {"model": model_id, "messages": messages, "temperature": temperature}
@@ -237,7 +259,7 @@ async def llm_call_msgs_native(messages: list, tools: list = None, model: str =
         model_id = model or os.environ.get("LLM_MODEL", "gpt-4o-mini")
 
     if not api_key:
-        raise ValueError("No LLM API configured")
+        raise _no_llm_error(model, org_id)
 
     headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
     payload = {"model": model_id, "messages": messages, "temperature": temperature}