feat:新增投顾agent和nl2sqlagent

tool/llm.py

@@ -198,18 +198,41 @@ class LLMClient:
         LLM_EMBED_DIMENSIONS），MRL 模型（qwen3-embedding 等）会截断并重新归一化到目标维度。
         """
         backend = self.primary
-        if not backend.embed_model:
+        embed_model = (
+            getattr(self.cfg, "api_embed_model", "")
+            if backend.name == "api"
+            else backend.embed_model
+        ) or backend.embed_model
+        if not embed_model:
             raise LLMFailError(f"backend={backend.name} 未配置 embed 模型")
-        url = f"{backend.base_url}/embeddings"
-        payload = {
-            "model": backend.embed_model,
-            "input": texts,
-            # "dimensions": self.cfg.embed_dimensions,
-        }
+        is_dashscope_compatible = "/compatible-mode/" in backend.base_url.lower()
+        if is_dashscope_compatible:
+            base_url = backend.base_url.rstrip("/")
+            marker = "/compatible-mode/v1"
+            if base_url.lower().endswith(marker):
+                base_url = base_url[: -len(marker)]
+            url = f"{base_url}/api/v1/services/embeddings/multimodal-embedding/multimodal-embedding"
+            payload = {
+                "model": embed_model,
+                "input": {"contents": [{"text": value} for value in texts]},
+            }
+        else:
+            url = f"{backend.base_url}/embeddings"
+            payload = {
+                "model": embed_model,
+                "input": texts,
+                # "dimensions": self.cfg.embed_dimensions,
+            }
         async with backend.client(self.cfg.timeout) as client:
             r = await client.post(url, headers=backend.headers, json=payload)
             r.raise_for_status()
         data = r.json()
+        if is_dashscope_compatible:
+            embeddings = data["output"]["embeddings"]
+            return [
+                item["embedding"]
+                for item in sorted(embeddings, key=lambda item: item["index"])
+            ]
         return [item["embedding"] for item in data["data"]]
 
     async def embed_one(self, text: str) -> list[float]: