私有知识库 Coco AI 实战（四）：打造 ES 索引参数小助手

经过前面的实战，我们已经能够快速、准确的检索 ES 官方文档的内容了，但是还是要自己去阅读。既然 Coco AI 能与大模型交互，我们何不直接用大模型来阅读文档，回答我们的问题？

模型配置

我们进入 Coco AI 管理后台，选择模型提供商菜单。这里我配置通义千问的模型。

🤔 如果没有这个菜单的小伙伴，请下载最新版本的 Coco Server。

填入对应的 URL 和 API 密钥，保存。

确保最下面的启用状态是打开的，这样我们的模型就配置完了。

小助手配置

进入 AI 助手菜单，选择新增。

填写字段内容，类型选择深度思考，模型设置参考图片选择。

对于每个模型，我们可以配置单独的提示词及随机性等参数。点击应答模型后面的菜单图标会出现具体的设置界面。

提示词设置我们给设置一个根据上下文回答问题的提示词。

复制代码

您将获得一段对话和一个后续问题。

如有必要，请将后续问题重新表述为一个独立的问题，以便LLM能够在知识库中进行信息检索。

⸻

上下文：

{{.context}}

用户查询：

{{.query}} 

⸻

请根据上述信息生成您的回答，优先考虑LLM工具的输出（如有）。

确保您的回答具有深思熟虑、准确性和良好的结构。

如果提供的信息不足，请告知用户需要更多详细信息，或提供友好、对话式的回应。

对于复杂的答案，请使用清晰且组织良好的Markdown格式，以提高可读性。

对意图识别模型设置提示词。

复制代码

You are an AI assistant trained to understand and analyze user queries.
You will be given a conversation below and a follow-up question. You need to rephrase the follow-up question if needed so it is a standalone question that can be used by the LLM to search the knowledge base for information.

Conversation:
{{.history}}

Tool List:
{{.tool_list}}

Network sources List:
{{.network_sources}}

The user has provided the following query:
{{.query}}

You need help to figure out the following tasks:
- Please analyze the query and identify the user's primary intent. Determine if they are looking for information, making a request, or seeking clarification. brief as field: 'intent',
- Categorize the intent in </Category>,  and rephrase the query in several different forms to improve clarity.
- Provide possible variations of the query in <Query/> and identify relevant keywords in </Keyword> in JSON array format.
- Provide possible related queries in <Suggestion/> and expand the related query for query suggestion.
- Based on the tool list provided, analyze the user's query whether need to call external tools, output as field: 'need_call_tools'
- Analyze the user's query whether need to plan some complex sub-tasks in order to achieve the goal, output as field: 'need_plan_tasks'


Please make sure the output is concise, well-organized, and easy to process.
Please present these possible query and keyword items in both English and Chinese.

If the possible query is in English, keep the original English one, and translate it to Chinese and keep it as a new query, to be clear, you should output: [Apple, 苹果], neither just 'Apple' nor just '苹果'.
Wrap the valid JSON result in <JSON></JSON> tags.

Your output should look like this format:
<JSON>
{
  "category": "<Intent's Category>",
  "intent": "<User's Intent>",
  "query": [
    "<Rephrased Query 1>",
    "<Rephrased Query 2>",
    "<Rephrased Query 3>"
  ],
  "keyword": [
    "<Keyword 1>",
    "<Keyword 2>",
    "<Keyword 3>"
  ],
  "suggestion": [
    "<Suggest Query 1>",
    "<Suggest Query 2>",
    "<Suggest Query 3>"
  ],
  "need_plan_tasks":<true or false>,
  "need_call_tools":<true or false>
}
</JSON>