update:优化代码

main/manager-api/src/main/java/xiaozhi/modules/agent/service/impl/AgentPluginMappingServiceImpl.java

@@ -1,5 +1,6 @@
 package xiaozhi.modules.agent.service.impl;
 
+import java.util.ArrayList;
 import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
@@ -7,11 +8,11 @@ import java.util.Map;
 import org.apache.commons.lang3.StringUtils;
 import org.springframework.stereotype.Service;
 
-import com.alibaba.druid.support.json.JSONUtils;
 import com.baomidou.mybatisplus.core.conditions.update.UpdateWrapper;
 import com.baomidou.mybatisplus.extension.service.impl.ServiceImpl;
 
 import lombok.RequiredArgsConstructor;
+import xiaozhi.common.utils.JsonUtils;
 import xiaozhi.modules.agent.dao.AgentPluginMappingMapper;
 import xiaozhi.modules.agent.entity.AgentPluginMapping;
 import xiaozhi.modules.agent.service.AgentPluginMappingService;
@@ -35,7 +36,8 @@ public class AgentPluginMappingServiceImpl extends ServiceImpl<AgentPluginMappin
     @Override
     public List<AgentPluginMapping> agentPluginParamsByAgentId(String agentId) {
         List<AgentPluginMapping> list = agentPluginMappingMapper.selectPluginsByAgentId(agentId);
-        int index = 0;
+        Map<String, List<KnowledgeBaseEntity>> knowledgeBaseMap = new HashMap<>();
+        Map<String, ModelConfigEntity> modelConfigMap = new HashMap<>();
         for (int i = list.size() - 1; i >= 0; i--) {
             AgentPluginMapping mapping = list.get(i);
             if (StringUtils.isBlank(mapping.getProviderCode())) {
@@ -51,14 +53,39 @@ public class AgentPluginMappingServiceImpl extends ServiceImpl<AgentPluginMappin
                     list.remove(i);
                     continue;
                 }
-                Map<String, String> paramInfo = new HashMap<>(2);
+                List<KnowledgeBaseEntity> knowledgeBaseList = knowledgeBaseMap.get(modelConfigEntity.getModelCode());
-                paramInfo.put("name", knowledgeBaseEntity.getName());
+                if (knowledgeBaseList == null) {
-                paramInfo.put("description", knowledgeBaseEntity.getDescription());
+                    knowledgeBaseList = new ArrayList<>();
-                mapping.setParamInfo(JSONUtils.toJSONString(paramInfo));
+                }
-                String providerCode = "xzKnowledgeBase_search_from_" + modelConfigEntity.getModelCode() + "_"
+                modelConfigMap.put(modelConfigEntity.getModelCode(), modelConfigEntity);
-                        + index;
+                knowledgeBaseList.add(knowledgeBaseEntity);
-                index++;
+                knowledgeBaseMap.put(modelConfigEntity.getModelCode(), knowledgeBaseList);
-                mapping.setProviderCode(providerCode);
+                list.remove(i);
+            }
+        }
+        if (knowledgeBaseMap.size() > 0) {
+            for (String pluginCode : knowledgeBaseMap.keySet()) {
+                List<KnowledgeBaseEntity> knowledgeBaseList = knowledgeBaseMap.get(pluginCode);
+                if (knowledgeBaseList == null || knowledgeBaseList.isEmpty()) {
+                    continue;
+                }
+
+                AgentPluginMapping agentPluginMapping = new AgentPluginMapping();
+                agentPluginMapping.setAgentId(agentId);
+                agentPluginMapping.setPluginId(pluginCode);
+                agentPluginMapping.setProviderCode("search_from_" + pluginCode);
+                agentPluginMapping.setId(Long.valueOf(list.size() + 1));
+
+                Map<String, Object> paramInfo = new HashMap<>(4);
+                ModelConfigEntity modelConfigEntity = modelConfigMap.get(pluginCode);
+                paramInfo.put("base_url", modelConfigEntity.getConfigJson().getStr("base_url"));
+                paramInfo.put("api_key", modelConfigEntity.getConfigJson().getStr("api_key"));
+                paramInfo.put("dataset_ids",
+                        knowledgeBaseList.stream().map(KnowledgeBaseEntity::getDatasetId).toList());
+                paramInfo.put("description",
+                        String.join(",", knowledgeBaseList.stream().map(KnowledgeBaseEntity::getDescription).toList()));
+                agentPluginMapping.setParamInfo(JsonUtils.toJsonString(paramInfo));
+                list.add(agentPluginMapping);
             }
         }
         return list;

main/xiaozhi-server/config.yaml

@@ -147,7 +147,15 @@ plugins:
       - ".wav"
       - ".p3"
     refresh_time: 300 # 刷新音乐列表的时间间隔，单位为秒
-
+  search_from_ragflow:
+    # 知识库的描述信息，方便大语言模型知道什么时候调用
+    description: "当用户问xxx时，调用本方法，使用知识库中的信息回答问题"
+    # ragflow接口配置
+    base_url: "http://192.168.0.8"
+    # ragflow api访问令牌
+    api_key: "ragflow-xxx"
+    # ragflow知识库id
+    dataset_ids: ["123456789"]
 # 声纹识别配置
 voiceprint:
   # 声纹接口地址
@@ -239,6 +247,7 @@ Intent:
     functions:
       - change_role
       - get_weather
+      # - search_from_ragflow
       # - get_news_from_chinanews
       - get_news_from_newsnow
       # play_music是服务器自带的音乐播放，hass_play_music是通过home assistant控制的独立外部程序音乐播放

main/xiaozhi-server/core/providers/tools/server_plugins/plugin_executor.py

@@ -1,6 +1,6 @@
 """服务端插件工具执行器"""
 
-from typing import Dict, Any, Optional, Tuple
+from typing import Dict, Any
 from ..base import ToolType, ToolDefinition, ToolExecutor
 from plugins_func.register import all_function_registry, Action, ActionResponse
 
@@ -11,53 +11,12 @@ class ServerPluginExecutor(ToolExecutor):
     def __init__(self, conn):
         self.conn = conn
         self.config = conn.config
-        # 存储知识库工具名称到真实插件名称的映射
-        self._knowledge_base_mapping: Dict[str, str] = {}
-
-    def _parse_knowledge_base_config(self, config_name: str) -> Optional[str]:
-        """
-        解析知识库配置名称，提取真实的插件名称
-
-        Args:
-            config_name: 配置名称，格式为 xzKnowledgeBase_<plugin_name>_<index>
-
-        Returns:
-            真实的插件名称，如果不是知识库配置则返回 None
-
-        Example:
-            "xzKnowledgeBase_search_from_ragflow_0" -> "search_from_ragflow"
-            "xzKnowledgeBase_search_from_ragflow_1" -> "search_from_ragflow"
-        """
-        if not config_name.startswith("xzKnowledgeBase_"):
-            return None
-
-        # 移除前缀
-        name_without_prefix = config_name[len("xzKnowledgeBase_"):]
-
-        # 找到最后一个下划线的位置
-        last_underscore_index = name_without_prefix.rfind("_")
-
-        if last_underscore_index == -1:
-            return None
-
-        # 提取真实插件名称（从开头到最后一个下划线之前）
-        real_plugin_name = name_without_prefix[:last_underscore_index]
-
-        return real_plugin_name
 
     async def execute(
         self, conn, tool_name: str, arguments: Dict[str, Any]
     ) -> ActionResponse:
         """执行服务端插件工具"""
-        # 检查是否是知识库工具调用
+        func_item = all_function_registry.get(tool_name)
-        real_plugin_name = self._knowledge_base_mapping.get(tool_name)
-        if real_plugin_name:
-            # 使用真实的插件名称获取函数
-            func_item = all_function_registry.get(real_plugin_name)
-        else:
-            # 普通插件调用
-            func_item = all_function_registry.get(tool_name)
-
         if not func_item:
             return ActionResponse(
                 action=Action.NOTFOUND, response=f"插件函数 {tool_name} 不存在"
@@ -112,51 +71,27 @@ class ServerPluginExecutor(ToolExecutor):
         for func_name in all_required_functions:
             func_item = all_function_registry.get(func_name)
             if func_item:
+                # 从函数注册中获取描述
+                fun_description = (
+                    self.config.get("plugins", {})
+                    .get(func_name, {})
+                    .get("description", "")
+                )
+                if fun_description is not None and len(fun_description) > 0:
+                    if "function" in func_item.description and isinstance(
+                        func_item.description["function"], dict
+                    ):
+                        func_item.description["function"][
+                            "description"
+                        ] = fun_description
                 tools[func_name] = ToolDefinition(
                     name=func_name,
                     description=func_item.description,
                     tool_type=ToolType.SERVER_PLUGIN,
                 )
 
-        # 处理知识库配置
-        plugins_config = self.config.get("plugins", {})
-        for config_name, config_value in plugins_config.items():
-            # 检查是否是知识库配置
-            real_plugin_name = self._parse_knowledge_base_config(config_name)
-            if real_plugin_name:
-                # 获取真实的插件函数
-                func_item = all_function_registry.get(real_plugin_name)
-                if func_item and isinstance(config_value, dict):
-                    # 从配置中获取自定义的 name 和 description
-                    custom_name = config_value.get("name", "")
-                    custom_description = config_value.get("description", "")
-
-                    # 创建动态的工具名称（使用配置名称去掉前缀部分作为工具名）
-                    tool_name = config_name[len("xzKnowledgeBase_"):]
-
-                    # 复制原始函数描述并修改
-                    custom_func_desc = func_item.description.copy()
-                    if "function" in custom_func_desc:
-                        custom_func_desc["function"] = custom_func_desc["function"].copy()
-                        custom_func_desc["function"]["name"] = tool_name
-                        custom_func_desc["function"]["description"] = custom_description
-
-                    # 注册工具
-                    tools[tool_name] = ToolDefinition(
-                        name=tool_name,
-                        description=custom_func_desc,
-                        tool_type=ToolType.SERVER_PLUGIN,
-                    )
-
-                    # 保存映射关系
-                    self._knowledge_base_mapping[tool_name] = real_plugin_name
-
         return tools
 
     def has_tool(self, tool_name: str) -> bool:
         """检查是否有指定的服务端插件工具"""
-        # 检查是否是知识库工具
-        if tool_name in self._knowledge_base_mapping:
-            return True
-        # 检查是否是普通工具
         return tool_name in all_function_registry

main/xiaozhi-server/plugins_func/functions/search_from_ragflow.py

@@ -1,5 +1,11 @@
+import requests
+import sys
+from config.logger import setup_logging
 from plugins_func.register import register_function, ToolType, ActionResponse, Action
 
+TAG = __name__
+logger = setup_logging()
+
 # 定义基础的函数描述模板
 SEARCH_FROM_RAGFLOW_FUNCTION_DESC = {
     "type": "function",
@@ -8,26 +14,85 @@ SEARCH_FROM_RAGFLOW_FUNCTION_DESC = {
         "description": "从知识库中查询信息",
         "parameters": {
             "type": "object",
-            "properties": {"query": {"type": "string", "description": "查询的关键词"}},
+            "properties": {"question": {"type": "string", "description": "查询的问题"}},
-            "required": ["query"],
+            "required": ["question"],
         },
     },
 }
 
 
 @register_function(
-    "search_from_ragflow", SEARCH_FROM_RAGFLOW_FUNCTION_DESC, ToolType.WAIT
+    "search_from_ragflow", SEARCH_FROM_RAGFLOW_FUNCTION_DESC, ToolType.SYSTEM_CTL
 )
-def search_from_ragflow(query=None):
+def search_from_ragflow(conn, question=None):
-    """
+    # 确保字符串参数正确处理编码
-    用于从ragflow知识库中查询信息
+    if question and isinstance(question, str):
-    """
+        # 确保问题参数是UTF-8编码的字符串
-    # TODO 从ragflow知识库中查询信息
+        pass
-    if query and "医生" in query:
-        response_text = "医院有张山、里斯、王五3名全科医生，其中王五医生是主要擅长眼科"
-    elif query and "科室" in query:
-        response_text = "医院眼科、麻醉科"
     else:
-        response_text = "暂无相关信息"
+        question = str(question) if question is not None else ""
 
-    return ActionResponse(Action.REQLLM, response_text, None)
+    base_url = conn.config["plugins"]["search_from_ragflow"].get("base_url", "")
+    api_key = conn.config["plugins"]["search_from_ragflow"].get("api_key", "")
+    dataset_ids = conn.config["plugins"]["search_from_ragflow"].get("dataset_ids", [])
+
+    url = base_url + "/api/v1/retrieval"
+    headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
+
+    # 确保payload中的字符串都是UTF-8编码
+    payload = {"question": question, "dataset_ids": dataset_ids}
+
+    try:
+        # 使用ensure_ascii=False确保JSON序列化时正确处理中文
+        response = requests.post(
+            url,
+            json=payload,
+            headers=headers,
+            timeout=5,
+            verify=False,
+        )
+
+        # 显式设置响应的编码为utf-8
+        response.encoding = "utf-8"
+
+        response.raise_for_status()
+
+        # 先获取文本内容，然后手动处理JSON解码
+        response_text = response.text
+        import json
+
+        result = json.loads(response_text)
+
+        if result.get("code") != 0:
+            error_detail = response.get("error", {}).get("detail", "")
+            # 安全地记录错误信息
+            logger.bind(tag=TAG).error(
+                "从RAGflow获取信息失败，原因：%s", str(error_detail)
+            )
+            return ActionResponse(Action.RESPONSE, None, "RAG接口返回异常")
+
+        chunks = result.get("data", {}).get("chunks", [])
+        contents = []
+        for chunk in chunks:
+            content = chunk.get("content", "")
+            if content:
+                # 安全地处理内容字符串
+                if isinstance(content, str):
+                    contents.append(content)
+                elif isinstance(content, bytes):
+                    contents.append(content.decode("utf-8", errors="replace"))
+                else:
+                    contents.append(str(content))
+
+        # 构建适合大模型的上下文内容（每段前加编号，段间两个换行）
+        context_text = "\n\n".join(
+            f"{i+1}. {c.strip()}" for i, c in enumerate(contents[:5])
+        )
+        if not context_text:
+            context_text = "根据知识库查询结果，没有相关信息。"
+        return ActionResponse(Action.REQLLM, context_text, None)
+
+    except Exception as e:
+        # 使用安全的方式记录异常，避免编码问题
+        logger.bind(tag=TAG).error("从RAGflow获取信息失败，原因：%s", str(e))
+        return ActionResponse(Action.RESPONSE, None, "RAG接口返回异常")