diff --git a/docSite/assets/imgs/versatile_assistant_1.png b/docSite/assets/imgs/versatile_assistant_1.png new file mode 100644 index 000000000..a58918e43 Binary files /dev/null and b/docSite/assets/imgs/versatile_assistant_1.png differ diff --git a/docSite/assets/imgs/versatile_assistant_2.png b/docSite/assets/imgs/versatile_assistant_2.png new file mode 100644 index 000000000..74f893594 Binary files /dev/null and b/docSite/assets/imgs/versatile_assistant_2.png differ diff --git a/docSite/assets/imgs/versatile_assistant_3.jpg b/docSite/assets/imgs/versatile_assistant_3.jpg new file mode 100644 index 000000000..b2735d771 Binary files /dev/null and b/docSite/assets/imgs/versatile_assistant_3.jpg differ diff --git a/docSite/assets/imgs/versatile_assistant_4.jpg b/docSite/assets/imgs/versatile_assistant_4.jpg new file mode 100644 index 000000000..26bec45d8 Binary files /dev/null and b/docSite/assets/imgs/versatile_assistant_4.jpg differ diff --git a/docSite/assets/imgs/versatile_assistant_5.jpg b/docSite/assets/imgs/versatile_assistant_5.jpg new file mode 100644 index 000000000..1d1a3666a Binary files /dev/null and b/docSite/assets/imgs/versatile_assistant_5.jpg differ diff --git a/docSite/assets/imgs/versatile_assistant_6.jpg b/docSite/assets/imgs/versatile_assistant_6.jpg new file mode 100644 index 000000000..29396514f Binary files /dev/null and b/docSite/assets/imgs/versatile_assistant_6.jpg differ diff --git a/docSite/assets/imgs/versatile_assistant_7.jpg b/docSite/assets/imgs/versatile_assistant_7.jpg new file mode 100644 index 000000000..97a0ff6b1 Binary files /dev/null and b/docSite/assets/imgs/versatile_assistant_7.jpg differ diff --git a/docSite/content/docs/workflow/examples/versatile_assistant.md b/docSite/content/docs/workflow/examples/versatile_assistant.md new file mode 100644 index 000000000..4d3643595 --- /dev/null +++ b/docSite/content/docs/workflow/examples/versatile_assistant.md @@ -0,0 +1,1535 @@ +--- +title: '全能助手' +description: '赋予联网功能，将用户问题进行分类，细分后对接对应 API 获取信息，经 GPT 整理后返回' +icon: 'search' +draft: false +toc: true +weight: 143 +--- + +![](/imgs/versatile_assistant_1.png) + +众所周知 GPT 只是一个语言模型，功能上有很多局限，但只要综合利用高级编排各模块功能，就可以轻松突破原有 GPT 的局限，实现更多功能。 + +当然，所谓“全能助手”只是一个遥远的设想，高级编排的玩法有很大的可能性，本文只是扩展了诸如【天气查询】、【微博热搜查询】的功能，主要还是希望大家能通过案例来了解下高级编排的思路，然后可以分享更多有意思的玩法。 + +## 简要介绍一下“全能助手”的思路 + +思路说来也简单，以下分别用文字和图片两种方式介绍下 + +- **文字描述：** + +1. 对于用户输入的问题，通过【问题分类】模块进行区分，分出【询问天气】、【微博热搜】、【其他问题】等 +2. 对于【询问天气】的情况，调用第三方 API 查询天气（后文会介绍），将查询到的 json 结果丢给【AI 对话】模块，让它根据用户问题来给出回答 +3. 对于【微博热搜】的情况，同理，也是调的第三方 API +4. 对于【其他问题】的情况，直接走【AI 对话】模块就好了，跟普通的 GPT 聊天一样 + +- **流程图（方便理解）：** + +![](/imgs/versatile_assistant_2.png) + +## 详细步骤 + +以下对于相同的步骤不会赘述，对于第三方接口只介绍了【天气查询】，而【微博热榜】跟【天气查询】的步骤是一样的，只是接口和提示词不同，所以不再赘述。后文会发出完整的高级编排配置，可以导入自行查看~ + +### 第三方 API 获取 + +案例中第三方接口来源目前都是在 https://api.vvhan.com/ 里获得，里面有许多花里胡哨的接口可以用，当然你有其他的接口可以对接也可以，反正主要是返回的数据。 + +举个查询天气的例子： + +1. 找到查询天气的 API 接口 + +![](/imgs/versatile_assistant_3.jpg) + +2. 由于我想要的效果是用户可以随意问接下来一周内任意时间的天气（比如用户可以问“接下来一周的天气适合晾被子吗”），所以选择了上面接口的这个格式：https://api.vvhan.com/api/weather?city=徐州&type=week + +返回 json： + +```json +{"success":true,"city":"徐州市","data":[{"date":"2023-09-21","week":"星期四","type":"多云","low":"14°C","high":"24°C","fengxiang":"东北风","fengli":"3级","night":{"type":"多云","fengxiang":"南风","fengli":"3级"}},{"date":"2023-09-22","week":"星期五","type":"阴","low":"19°C","high":"25°C","fengxiang":"东风","fengli":"3级","night":{"type":"阴","fengxiang":"东风","fengli":"3级"}},{"date":"2023-09-23","week":"星期六","type":"小雨","low":"20°C","high":"23°C","fengxiang":"东北风","fengli":"3级","night":{"type":"小雨","fengxiang":"东北风","fengli":"3级"}},{"date":"2023-09-24","week":"星期日","type":"中雨","low":"20°C","high":"23°C","fengxiang":"东风","fengli":"3级","night":{"type":"中雨","fengxiang":"东北风","fengli":"3级"}},{"date":"2023-09-25","week":"星期一","type":"小雨","low":"20°C","high":"24°C","fengxiang":"北风","fengli":"3级","night":{"type":"阴","fengxiang":"北风","fengli":"3级"}},{"date":"2023-09-26","week":"星期二","type":"阴","low":"21°C","high":"27°C","fengxiang":"北风","fengli":"3级","night":{"type":"阴","fengxiang":"北风","fengli":"3级"}},{"date":"2023-09-27","week":"星期三","type":"阴","low":"21°C","high":"25°C","fengxiang":"东北风","fengli":"3级","night":{"type":"阴","fengxiang":"北风","fengli":"3级"}}]} +``` + +3. 由于 FastGPT 的 【http 模块】，对于返回的 json 是以对象形式接收，而我们期望得到的是上述 json 中的“data”字段，而“data”又是数组格式，无法直接丢给【AI 对话】模块（我丢过，非字符串格式报错了，不知道后面会不会更新），所以需要对其做一层中转，将“data”字段转成字符串格式。思路如此，中转方式多样，这里介绍我自己的做法：用 python 起一个服务，来负责对 API 的中转，代码如下（包含了天气接口和微博热搜接口）： + +```python +from flask import Flask, request, Response +import requests +import json + +app = Flask(__name__) + +@app.route('/weather', methods=['GET','POST']) +def weather(): + if request.method == 'POST': + city = request.form.get('city') + if not city: + data = request.get_json() + if data: + city = data.get('city') + else: + city = request.args.get('city') + + api_url = "https://api.vvhan.com/api/weather" + # 为了方便，这里强行写死一周了，只有城市是外部传进来的 + params = {"city": city, "type": "week"} + + response = requests.get(api_url, params=params) + + res = json.loads(response.text) + # 将data字段转成字符串格式 + res['data'] = json.dumps(res['data'], ensure_ascii=False) + return Response(json.dumps(res, ensure_ascii=False), mimetype="application/json") + +@app.route('/wbhot', methods=['GET','POST']) +def wbhot(): + api_url = "https://api.vvhan.com/api/wbhot" + + response = requests.get(api_url) + + res = json.loads(response.text) + # 只返回前10条热搜（免得数据太多耗token） + res['data'] = res['data'][:10] + # 将data字段转成字符串格式 + res['data'] = json.dumps(res['data'], ensure_ascii=False) + return Response(json.dumps(res, ensure_ascii=False), mimetype="application/json") + +if __name__ == '__main__': + #部署在3017端口，可自行修改 + app.run(host='0.0.0.0', port=3017) +``` + +4. 接口测试返回数据： + +```json +{"success": true, "city": "广州市", "data": "[{\"date\": \"2023-09-21\", \"week\": \"星期四\", \"type\": \"雷阵雨\", \"low\": \"25°C\", \"high\": \"34°C\", \"fengxiang\": \"微风\", \"fengli\": \"3级\", \"night\": {\"type\": \"雷阵雨\", \"fengxiang\": \"微风\", \"fengli\": \"3级\"}}, {\"date\": \"2023-09-22\", \"week\": \"星期五\", \"type\": \"雷阵雨\", \"low\": \"25°C\", \"high\": \"32°C\", \"fengxiang\": \"微风\", \"fengli\": \"3级\", \"night\": {\"type\": \"多云\", \"fengxiang\": \"微风\", \"fengli\": \"3级\"}}, {\"date\": \"2023-09-23\", \"week\": \"星期六\", \"type\": \"多云\", \"low\": \"25°C\", \"high\": \"32°C\", \"fengxiang\": \"微风\", \"fengli\": \"3级\", \"night\": {\"type\": \"多云\", \"fengxiang\": \"微风\", \"fengli\": \"3级\"}}, {\"date\": \"2023-09-24\", \"week\": \"星期日\", \"type\": \"多云\", \"low\": \"25°C\", \"high\": \"34°C\", \"fengxiang\": \"微风\", \"fengli\": \"3级\", \"night\": {\"type\": \"多云\", \"fengxiang\": \"微风\", \"fengli\": \"3级\"}}, {\"date\": \"2023-09-25\", \"week\": \"星期一\", \"type\": \"多云\", \"low\": \"25°C\", \"high\": \"34°C\", \"fengxiang\": \"微风\", \"fengli\": \"3级\", \"night\": {\"type\": \"多云\", \"fengxiang\": \"微风\", \"fengli\": \"3级\"}}, {\"date\": \"2023-09-26\", \"week\": \"星期二\", \"type\": \"多云\", \"low\": \"25°C\", \"high\": \"34°C\", \"fengxiang\": \"微风\", \"fengli\": \"3级\", \"night\": {\"type\": \"多云\", \"fengxiang\": \"微风\", \"fengli\": \"3级\"}}, {\"date\": \"2023-09-27\", \"week\": \"星期三\", \"type\": \"中雨\", \"low\": \"26°C\", \"high\": \"33°C\", \"fengxiang\": \"微风\", \"fengli\": \"3级\", \"night\": {\"type\": \"中雨\", \"fengxiang\": \"微风\", \"fengli\": \"3级\"}}]"} +``` + +### 用户问题分类 + +第一步就是对用户问题进行分类，如图红框部分： + +![](/imgs/versatile_assistant_4.jpg) + +### 接口参数获取及处理 + +**数据获取：** + +由于天气接口需要传入的是“城市”字段，所以需要我们从用户的问题中提取出“城市”字段，所以【文本内容提取】模块登场。 + +提取要求描述（自行调试另一个 prompt 也行）： + +``` +你是一个天气查询助手。根据用户问题，提取出城市。注意不是简单的文本提取，而是上下文理解后的提取。如果用户问题中不包含城市则不提取 +``` + +目标字段：城市 + +**数据处理：** + +1. 设计一个好用的功能往往需要把用户当成小白，所以用户的问题中很可能是没有我们需要的参数的，所以当“提取字段缺失”时，我们需要【指定回复】模块来提示用户输入城市 +2. 若提取成功，则将提取出来的“城市”发给 http 模块 + +如图： + +![](/imgs/versatile_assistant_5.jpg) + +### AI 总结回复 + +上述步骤已经拿到了天气的 json 结果，但我们需要的是语义化的结果，所以就要把“json 结果”、“当前时间”（方便用户问今天还是明天天气时可以判断）、“上下文聊天记录”（方便用户的问题涉及上下文关联时能区分）这三个参数传给【AI 对话】模块，让它来总结回复。 +限定词（我自己调试的，你有更好的也可以替换）： + +``` +已知条件：1. 当前时间是{{cTime}}；2. 这份json数据是要询问的地方的天气数据，比如用户问的是“北京”的天气，那这份json就是“北京”的天气数据。 + +现在请自行解析json后回复用户 +``` + +如图： + +![](/imgs/versatile_assistant_6.jpg) + +## 模块编排 + +复制下面配置，点击「高级编排」右上角的导入按键，导入该配置。 + +PS1：接口的第三方域名已打码，需要自行替换 + +PS2：配置中的问题分类还包含着“联网搜索”，这个是另一个案例中整合进来的，这里不做介绍，有兴趣看另一篇“联网 GPT”案例。没兴趣也可以在问题分类中删掉这个分支 + +{{% details title="编排配置" closed="true" %}} + +```json +[ + { + "moduleId": "userChatInput", + "name": "用户问题(对话入口)", + "flowType": "questionInput", + "position": { + "x": 464.32198615344566, + "y": 1602.2698463081606 + }, + "inputs": [ + { + "key": "userChatInput", + "type": "systemInput", + "label": "用户问题", + "connected": true + } + ], + "outputs": [ + { + "key": "userChatInput", + "label": "用户问题", + "type": "source", + "valueType": "string", + "targets": [ + { + "moduleId": "toho1d", + "key": "userChatInput" + }, + { + "moduleId": "rov9zf", + "key": "content" + }, + { + "moduleId": "6q1n0a", + "key": "userChatInput" + }, + { + "moduleId": "i0u1iy", + "key": "userChatInput" + }, + { + "moduleId": "uo68aj", + "key": "userChatInput" + }, + { + "moduleId": "3k4zw1", + "key": "content" + }, + { + "moduleId": "01fwnb", + "key": "userChatInput" + } + ] + } + ] + }, + { + "moduleId": "history", + "name": "聊天记录", + "flowType": "historyNode", + "position": { + "x": 452.5466249541586, + "y": 1276.3930310334215 + }, + "inputs": [ + { + "key": "maxContext", + "type": "numberInput", + "label": "最长记录数", + "value": 6, + "min": 0, + "max": 50, + "connected": true + }, + { + "key": "history", + "type": "hidden", + "label": "聊天记录", + "connected": true + } + ], + "outputs": [ + { + "key": "history", + "label": "聊天记录", + "valueType": "chat_history", + "type": "source", + "targets": [ + { + "moduleId": "toho1d", + "key": "history" + }, + { + "moduleId": "6q1n0a", + "key": "history" + }, + { + "moduleId": "rov9zf", + "key": "history" + }, + { + "moduleId": "uo68aj", + "key": "history" + }, + { + "moduleId": "3k4zw1", + "key": "history" + }, + { + "moduleId": "01fwnb", + "key": "history" + } + ] + } + ] + }, + { + "moduleId": "toho1d", + "name": "问题分类", + "flowType": "classifyQuestion", + "showStatus": true, + "position": { + "x": 942.1068912757241, + "y": 1044.6701989335747 + }, + "inputs": [ + { + "key": "systemPrompt", + "type": "textarea", + "valueType": "string", + "value": "", + "label": "系统提示词", + "description": "你可以添加一些特定内容的介绍，从而更好的识别用户的问题类型。这个内容通常是给模型介绍一个它不知道的内容。", + "placeholder": "例如: \n1. Laf 是一个云函数开发平台……\n2. Sealos 是一个集群操作系统", + "connected": true + }, + { + "key": "history", + "type": "target", + "label": "聊天记录", + "valueType": "chat_history", + "connected": true + }, + { + "key": "userChatInput", + "type": "target", + "label": "用户问题", + "required": true, + "valueType": "string", + "connected": true + }, + { + "key": "agents", + "type": "custom", + "label": "", + "value": [ + { + "value": "询问天气", + "key": "fasw" + }, + { + "value": "其它问题", + "key": "wl9i" + }, + { + "value": "微博热榜", + "key": "sf09" + }, + { + "value": "联网搜索", + "key": "6p8b" + } + ], + "connected": true + } + ], + "outputs": [ + { + "key": "fasw", + "label": "", + "type": "hidden", + "targets": [ + { + "moduleId": "rov9zf", + "key": "switch" + } + ] + }, + { + "key": "fqsw", + "label": "", + "type": "hidden", + "targets": [] + }, + { + "key": "fesw", + "label": "", + "type": "hidden", + "targets": [] + }, + { + "key": "wl9i", + "label": "", + "type": "hidden", + "targets": [ + { + "moduleId": "i0u1iy", + "key": "switch" + } + ] + }, + { + "key": "sf09", + "label": "", + "type": "hidden", + "targets": [ + { + "moduleId": "3m320f", + "key": "switch" + } + ] + }, + { + "key": "6p8b", + "label": "", + "type": "hidden", + "targets": [ + { + "moduleId": "3k4zw1", + "key": "switch" + } + ] + } + ] + }, + { + "moduleId": "rov9zf", + "name": "文本内容提取", + "flowType": "contentExtract", + "showStatus": true, + "position": { + "x": 1632.5948304111266, + "y": 331.84468967718163 + }, + "inputs": [ + { + "key": "switch", + "type": "target", + "label": "触发器", + "valueType": "any", + "connected": true + }, + { + "key": "description", + "type": "textarea", + "valueType": "string", + "value": "你是一个天气查询助手。根据用户问题，提取出城市。注意不是简单的文本提取，而是上下文理解后的提取。如果用户问题中不包含城市则不提取", + "label": "提取要求描述", + "description": "写一段提取要求，告诉 AI 需要提取哪些内容", + "required": true, + "placeholder": "例如: \n1. 你是一个实验室预约助手。根据用户问题，提取出姓名、实验室号和预约时间", + "connected": true + }, + { + "key": "history", + "type": "target", + "label": "聊天记录", + "valueType": "chat_history", + "connected": true + }, + { + "key": "content", + "type": "target", + "label": "需要提取的文本", + "required": true, + "valueType": "string", + "connected": true + }, + { + "key": "extractKeys", + "type": "custom", + "label": "目标字段", + "description": "由 '描述' 和 'key' 组成一个目标字段，可提取多个目标字段", + "value": [ + { + "desc": "城市", + "key": "city", + "required": true + } + ], + "connected": true + } + ], + "outputs": [ + { + "key": "success", + "label": "字段完全提取", + "valueType": "boolean", + "type": "source", + "targets": [ + { + "moduleId": "4gy7tw", + "key": "switch" + } + ] + }, + { + "key": "failed", + "label": "提取字段缺失", + "valueType": "boolean", + "type": "source", + "targets": [ + { + "moduleId": "eu1xhx", + "key": "switch" + } + ] + }, + { + "key": "fields", + "label": "完整提取结果", + "description": "一个 JSON 字符串，例如：{\"name:\":\"YY\",\"Time\":\"2023/7/2 18:00\"}", + "valueType": "string", + "type": "source", + "targets": [] + }, + { + "key": "city", + "label": "提取结果-城市", + "description": "无法提取时不会返回", + "valueType": "string", + "type": "source", + "targets": [ + { + "moduleId": "4gy7tw", + "key": "city" + } + ] + } + ] + }, + { + "moduleId": "eu1xhx", + "name": "指定回复", + "flowType": "answerNode", + "position": { + "x": 2137.9125850753494, + "y": 326.06694967444105 + }, + "inputs": [ + { + "key": "switch", + "type": "target", + "label": "触发器", + "valueType": "any", + "connected": true + }, + { + "key": "text", + "type": "textarea", + "valueType": "string", + "value": "请告诉我你要查询的是哪个城市的天气", + "label": "回复的内容", + "description": "可以使用 \\n 来实现换行。也可以通过外部模块输入实现回复，外部模块输入时会覆盖当前填写的内容", + "connected": true + } + ], + "outputs": [ + { + "key": "finish", + "label": "回复结束", + "description": "回复完成后触发", + "valueType": "boolean", + "type": "source", + "targets": [] + } + ] + }, + { + "moduleId": "4gy7tw", + "name": "HTTP模块", + "flowType": "httpRequest", + "showStatus": true, + "position": { + "x": 2156.411722495609, + "y": 661.4677041198821 + }, + "inputs": [ + { + "key": "url", + "value": "http://api.xxx.cn/weather", + "type": "input", + "label": "请求地址", + "description": "请求目标地址", + "placeholder": "https://api.fastgpt.run/getInventory", + "required": true, + "connected": true + }, + { + "key": "switch", + "type": "target", + "label": "触发器", + "valueType": "any", + "connected": true + }, + { + "valueType": "string", + "type": "target", + "label": "城市", + "edit": true, + "required": false, + "connected": true, + "key": "city" + } + ], + "outputs": [ + { + "label": "结果", + "valueType": "string", + "type": "source", + "edit": true, + "targets": [ + { + "moduleId": "6q1n0a", + "key": "systemPrompt" + } + ], + "key": "data" + }, + { + "key": "finish", + "label": "请求结束", + "valueType": "boolean", + "type": "source", + "targets": [ + { + "moduleId": "6q1n0a", + "key": "switch" + } + ] + } + ] + }, + { + "moduleId": "6q1n0a", + "name": "AI 对话", + "flowType": "chatNode", + "showStatus": true, + "position": { + "x": 2771.9325168087653, + "y": 262.8526145591803 + }, + "inputs": [ + { + "key": "model", + "type": "custom", + "label": "对话模型", + "value": "gpt-3.5-turbo", + "list": [ + { + "label": "GPT35-4k", + "value": "gpt-3.5-turbo" + }, + { + "label": "GPT35-16k", + "value": "gpt-3.5-turbo-16k" + }, + { + "label": "GPT4-8k", + "value": "gpt-4" + } + ], + "connected": true + }, + { + "key": "temperature", + "type": "slider", + "label": "温度", + "value": 0, + "min": 0, + "max": 10, + "step": 1, + "markList": [ + { + "label": "严谨", + "value": 0 + }, + { + "label": "发散", + "value": 10 + } + ], + "connected": true + }, + { + "key": "maxToken", + "type": "custom", + "label": "回复上限", + "value": 2000, + "min": 100, + "max": 4000, + "step": 50, + "markList": [ + { + "label": "100", + "value": 100 + }, + { + "label": "4000", + "value": 4000 + } + ], + "connected": true + }, + { + "key": "systemPrompt", + "type": "textarea", + "label": "系统提示词", + "max": 300, + "valueType": "string", + "description": "模型固定的引导词，通过调整该内容，可以引导模型聊天方向。该内容会被固定在上下文的开头。可使用变量，例如 {{language}}", + "placeholder": "模型固定的引导词，通过调整该内容，可以引导模型聊天方向。该内容会被固定在上下文的开头。可使用变量，例如 {{language}}", + "value": "", + "connected": true + }, + { + "key": "quoteTemplate", + "type": "hidden", + "label": "引用内容模板", + "valueType": "string", + "value": "", + "connected": true + }, + { + "key": "quotePrompt", + "type": "hidden", + "label": "引用内容提示词", + "valueType": "string", + "value": "", + "connected": true + }, + { + "key": "switch", + "type": "target", + "label": "触发器", + "valueType": "any", + "connected": true + }, + { + "key": "quoteQA", + "type": "custom", + "label": "引用内容", + "description": "对象数组格式，结构：\n [{q:'问题',a:'回答'}]", + "valueType": "kb_quote", + "connected": false + }, + { + "key": "history", + "type": "target", + "label": "聊天记录", + "valueType": "chat_history", + "connected": true + }, + { + "key": "userChatInput", + "type": "target", + "label": "用户问题", + "required": true, + "valueType": "string", + "connected": true + }, + { + "key": "limitPrompt", + "type": "textarea", + "valueType": "string", + "label": "限定词", + "description": "限定模型对话范围，会被放置在本次提问前，拥有强引导和限定性。可使用变量，例如 {{language}}。引导例子:\n1. 知识库是关于 Laf 的介绍，参考知识库回答问题，与 \"Laf\" 无关内容，直接回复: \"我不知道\"。\n2. 你仅回答关于 \"xxx\" 的问题，其他问题回复: \"xxxx\"", + "placeholder": "限定模型对话范围，会被放置在本次提问前，拥有强引导和限定性。可使用变量，例如 {{language}}。引导例子:\n1. 知识库是关于 Laf 的介绍，参考知识库回答问题，与 \"Laf\" 无关内容，直接回复: \"我不知道\"。\n2. 你仅回答关于 \"xxx\" 的问题，其他问题回复: \"xxxx\"", + "value": "已知条件：1. 当前时间是{{cTime}}；2. 这份json数据是要询问的地方的天气数据，比如用户问的是“北京”的天气，那这份json就是“北京”的天气数据。\n\n现在请自行解析json后回复用户", + "connected": true + } + ], + "outputs": [ + { + "key": "answerText", + "label": "模型回复", + "description": "将在 stream 回复完毕后触发", + "valueType": "string", + "type": "source", + "targets": [] + }, + { + "key": "finish", + "label": "回复结束", + "description": "AI 回复完成后触发", + "valueType": "boolean", + "type": "source", + "targets": [] + } + ] + }, + { + "moduleId": "i0u1iy", + "name": "AI 对话", + "flowType": "chatNode", + "showStatus": true, + "position": { + "x": 1636.416225126142, + "y": 1243.2398251366028 + }, + "inputs": [ + { + "key": "model", + "type": "custom", + "label": "对话模型", + "value": "gpt-3.5-turbo", + "list": [ + { + "label": "GPT35-4k", + "value": "gpt-3.5-turbo" + }, + { + "label": "GPT35-16k", + "value": "gpt-3.5-turbo-16k" + }, + { + "label": "GPT4-8k", + "value": "gpt-4" + } + ], + "connected": true + }, + { + "key": "temperature", + "type": "slider", + "label": "温度", + "value": 0, + "min": 0, + "max": 10, + "step": 1, + "markList": [ + { + "label": "严谨", + "value": 0 + }, + { + "label": "发散", + "value": 10 + } + ], + "connected": true + }, + { + "key": "maxToken", + "type": "custom", + "label": "回复上限", + "value": 4000, + "min": 100, + "max": 4000, + "step": 50, + "markList": [ + { + "label": "100", + "value": 100 + }, + { + "label": "4000", + "value": 4000 + } + ], + "connected": true + }, + { + "key": "systemPrompt", + "type": "textarea", + "label": "系统提示词", + "max": 300, + "valueType": "string", + "description": "模型固定的引导词，通过调整该内容，可以引导模型聊天方向。该内容会被固定在上下文的开头。可使用变量，例如 {{language}}", + "placeholder": "模型固定的引导词，通过调整该内容，可以引导模型聊天方向。该内容会被固定在上下文的开头。可使用变量，例如 {{language}}", + "value": "", + "connected": true + }, + { + "key": "quoteTemplate", + "type": "hidden", + "label": "引用内容模板", + "valueType": "string", + "value": "", + "connected": true + }, + { + "key": "quotePrompt", + "type": "hidden", + "label": "引用内容提示词", + "valueType": "string", + "value": "", + "connected": true + }, + { + "key": "switch", + "type": "target", + "label": "触发器", + "valueType": "any", + "connected": true + }, + { + "key": "quoteQA", + "type": "custom", + "label": "引用内容", + "description": "对象数组格式，结构：\n [{q:'问题',a:'回答'}]", + "valueType": "kb_quote", + "connected": false + }, + { + "key": "history", + "type": "target", + "label": "聊天记录", + "valueType": "chat_history", + "connected": true + }, + { + "key": "userChatInput", + "type": "target", + "label": "用户问题", + "required": true, + "valueType": "string", + "connected": true + }, + { + "key": "limitPrompt", + "type": "textarea", + "valueType": "string", + "label": "限定词", + "description": "限定模型对话范围，会被放置在本次提问前，拥有强引导和限定性。可使用变量，例如 {{language}}。引导例子:\n1. 知识库是关于 Laf 的介绍，参考知识库回答问题，与 \"Laf\" 无关内容，直接回复: \"我不知道\"。\n2. 你仅回答关于 \"xxx\" 的问题，其他问题回复: \"xxxx\"", + "placeholder": "限定模型对话范围，会被放置在本次提问前，拥有强引导和限定性。可使用变量，例如 {{language}}。引导例子:\n1. 知识库是关于 Laf 的介绍，参考知识库回答问题，与 \"Laf\" 无关内容，直接回复: \"我不知道\"。\n2. 你仅回答关于 \"xxx\" 的问题，其他问题回复: \"xxxx\"", + "value": "", + "connected": true + } + ], + "outputs": [ + { + "key": "answerText", + "label": "模型回复", + "description": "将在 stream 回复完毕后触发", + "valueType": "string", + "type": "source", + "targets": [] + }, + { + "key": "finish", + "label": "回复结束", + "description": "AI 回复完成后触发", + "valueType": "boolean", + "type": "source", + "targets": [] + } + ] + }, + { + "moduleId": "3m320f", + "name": "HTTP模块", + "flowType": "httpRequest", + "showStatus": true, + "position": { + "x": 1640.5198770218628, + "y": 2420.3111570417573 + }, + "inputs": [ + { + "key": "url", + "value": "http://api.xxx.cn/wbhot", + "type": "input", + "label": "请求地址", + "description": "请求目标地址", + "placeholder": "https://api.fastgpt.run/getInventory", + "required": true, + "connected": true + }, + { + "key": "switch", + "type": "target", + "label": "触发器", + "valueType": "any", + "connected": true + } + ], + "outputs": [ + { + "label": "data", + "valueType": "string", + "type": "source", + "edit": true, + "targets": [ + { + "moduleId": "uo68aj", + "key": "systemPrompt" + } + ], + "key": "data" + }, + { + "key": "finish", + "label": "请求结束", + "valueType": "boolean", + "type": "source", + "targets": [ + { + "moduleId": "uo68aj", + "key": "switch" + } + ] + } + ] + }, + { + "moduleId": "uo68aj", + "name": "AI 对话", + "flowType": "chatNode", + "showStatus": true, + "position": { + "x": 2248.9999960823247, + "y": 2411.459363346701 + }, + "inputs": [ + { + "key": "model", + "type": "custom", + "label": "对话模型", + "value": "gpt-3.5-turbo-16k", + "list": [ + { + "label": "GPT35-4k", + "value": "gpt-3.5-turbo" + }, + { + "label": "GPT35-16k", + "value": "gpt-3.5-turbo-16k" + }, + { + "label": "GPT4-8k", + "value": "gpt-4" + } + ], + "connected": true + }, + { + "key": "temperature", + "type": "slider", + "label": "温度", + "value": 0, + "min": 0, + "max": 10, + "step": 1, + "markList": [ + { + "label": "严谨", + "value": 0 + }, + { + "label": "发散", + "value": 10 + } + ], + "connected": true + }, + { + "key": "maxToken", + "type": "custom", + "label": "回复上限", + "value": 16000, + "min": 100, + "max": 4000, + "step": 50, + "markList": [ + { + "label": "100", + "value": 100 + }, + { + "label": "4000", + "value": 4000 + } + ], + "connected": true + }, + { + "key": "systemPrompt", + "type": "textarea", + "label": "系统提示词", + "max": 300, + "valueType": "string", + "description": "模型固定的引导词，通过调整该内容，可以引导模型聊天方向。该内容会被固定在上下文的开头。可使用变量，例如 {{language}}", + "placeholder": "模型固定的引导词，通过调整该内容，可以引导模型聊天方向。该内容会被固定在上下文的开头。可使用变量，例如 {{language}}", + "value": "", + "connected": true + }, + { + "key": "quoteTemplate", + "type": "hidden", + "label": "引用内容模板", + "valueType": "string", + "value": "", + "connected": true + }, + { + "key": "quotePrompt", + "type": "hidden", + "label": "引用内容提示词", + "valueType": "string", + "value": "", + "connected": true + }, + { + "key": "switch", + "type": "target", + "label": "触发器", + "valueType": "any", + "connected": true + }, + { + "key": "quoteQA", + "type": "custom", + "label": "引用内容", + "description": "对象数组格式，结构：\n [{q:'问题',a:'回答'}]", + "valueType": "kb_quote", + "connected": false + }, + { + "key": "history", + "type": "target", + "label": "聊天记录", + "valueType": "chat_history", + "connected": true + }, + { + "key": "userChatInput", + "type": "target", + "label": "用户问题", + "required": true, + "valueType": "string", + "connected": true + }, + { + "key": "limitPrompt", + "type": "textarea", + "valueType": "string", + "label": "限定词", + "description": "限定模型对话范围，会被放置在本次提问前，拥有强引导和限定性。可使用变量，例如 {{language}}。引导例子:\n1. 知识库是关于 Laf 的介绍，参考知识库回答问题，与 \"Laf\" 无关内容，直接回复: \"我不知道\"。\n2. 你仅回答关于 \"xxx\" 的问题，其他问题回复: \"xxxx\"", + "placeholder": "限定模型对话范围，会被放置在本次提问前，拥有强引导和限定性。可使用变量，例如 {{language}}。引导例子:\n1. 知识库是关于 Laf 的介绍，参考知识库回答问题，与 \"Laf\" 无关内容，直接回复: \"我不知道\"。\n2. 你仅回答关于 \"xxx\" 的问题，其他问题回复: \"xxxx\"", + "value": "以上json数据是当前的微博热榜数据，回答的时候用markdown格式，只需回复热搜标题的前10即可", + "connected": true + } + ], + "outputs": [ + { + "key": "answerText", + "label": "模型回复", + "description": "将在 stream 回复完毕后触发", + "valueType": "string", + "type": "source", + "targets": [] + }, + { + "key": "finish", + "label": "回复结束", + "description": "AI 回复完成后触发", + "valueType": "boolean", + "type": "source", + "targets": [] + } + ] + }, + { + "moduleId": "qoccls", + "name": "聊天记录", + "flowType": "historyNode", + "position": { + "x": 448.94080110453046, + "y": 990.48670949044 + }, + "inputs": [ + { + "key": "maxContext", + "type": "numberInput", + "label": "最长记录数", + "value": 50, + "min": 0, + "max": 50, + "connected": true + }, + { + "key": "history", + "type": "hidden", + "label": "聊天记录", + "connected": true + } + ], + "outputs": [ + { + "key": "history", + "label": "聊天记录", + "valueType": "chat_history", + "type": "source", + "targets": [ + { + "moduleId": "i0u1iy", + "key": "history" + } + ] + } + ] + }, + { + "moduleId": "3k4zw1", + "name": "文本内容提取", + "flowType": "contentExtract", + "showStatus": true, + "position": { + "x": 1608.4732867173993, + "y": 3651.5738821560017 + }, + "inputs": [ + { + "key": "switch", + "type": "target", + "label": "触发器", + "valueType": "any", + "connected": true + }, + { + "key": "description", + "type": "textarea", + "valueType": "string", + "value": "你是谷歌搜索机器人，可以生成搜索词。你需要自行判断是否需要生成搜索词，如果不需要则返回空字符串。", + "label": "提取要求描述", + "description": "写一段提取要求，告诉 AI 需要提取哪些内容", + "required": true, + "placeholder": "例如: \n1. 你是一个实验室预约助手。根据用户问题，提取出姓名、实验室号和预约时间", + "connected": true + }, + { + "key": "history", + "type": "target", + "label": "聊天记录", + "valueType": "chat_history", + "connected": true + }, + { + "key": "content", + "type": "target", + "label": "需要提取的文本", + "required": true, + "valueType": "string", + "connected": true + }, + { + "key": "extractKeys", + "type": "custom", + "label": "目标字段", + "description": "由 '描述' 和 'key' 组成一个目标字段，可提取多个目标字段", + "value": [ + { + "desc": "搜索词", + "key": "searchKey", + "required": true + } + ], + "connected": true + } + ], + "outputs": [ + { + "key": "success", + "label": "字段完全提取", + "valueType": "boolean", + "type": "source", + "targets": [] + }, + { + "key": "failed", + "label": "提取字段缺失", + "valueType": "boolean", + "type": "source", + "targets": [] + }, + { + "key": "fields", + "label": "完整提取结果", + "description": "一个 JSON 字符串，例如：{\"name:\":\"YY\",\"Time\":\"2023/7/2 18:00\"}", + "valueType": "string", + "type": "source", + "targets": [] + }, + { + "key": "searchKey", + "label": "提取结果-搜索词", + "description": "无法提取时不会返回", + "valueType": "string", + "type": "source", + "targets": [ + { + "moduleId": "3ojl65", + "key": "searchKey" + } + ] + } + ] + }, + { + "moduleId": "3ojl65", + "name": "HTTP模块", + "flowType": "httpRequest", + "showStatus": true, + "position": { + "x": 2250.5435150325084, + "y": 3647.785854643283 + }, + "inputs": [ + { + "key": "url", + "value": "http://api.xxx.cn/google", + "type": "input", + "label": "请求地址", + "description": "请求目标地址", + "placeholder": "https://api.fastgpt.run/getInventory", + "required": true, + "connected": true + }, + { + "key": "switch", + "type": "target", + "label": "触发器", + "valueType": "any", + "connected": false + }, + { + "valueType": "string", + "type": "target", + "label": "搜索词", + "edit": true, + "key": "searchKey", + "required": true, + "connected": true + } + ], + "outputs": [ + { + "label": "搜索词", + "valueType": "string", + "type": "source", + "edit": true, + "targets": [], + "key": "searchKey" + }, + { + "label": "搜索结果", + "valueType": "string", + "type": "source", + "edit": true, + "targets": [ + { + "moduleId": "01fwnb", + "key": "systemPrompt" + } + ], + "key": "prompt" + }, + { + "key": "finish", + "label": "请求结束", + "valueType": "boolean", + "type": "source", + "targets": [ + { + "moduleId": "01fwnb", + "key": "switch" + } + ] + } + ] + }, + { + "moduleId": "01fwnb", + "name": "AI 对话", + "flowType": "chatNode", + "showStatus": true, + "position": { + "x": 2913.2501313416833, + "y": 3642.3449136226823 + }, + "inputs": [ + { + "key": "model", + "type": "custom", + "label": "对话模型", + "value": "gpt-3.5-turbo-16k", + "list": [ + { + "label": "GPT35-4k", + "value": "gpt-3.5-turbo" + }, + { + "label": "GPT35-16k", + "value": "gpt-3.5-turbo-16k" + }, + { + "label": "GPT4-8k", + "value": "gpt-4" + } + ], + "connected": true + }, + { + "key": "temperature", + "type": "slider", + "label": "温度", + "value": 0, + "min": 0, + "max": 10, + "step": 1, + "markList": [ + { + "label": "严谨", + "value": 0 + }, + { + "label": "发散", + "value": 10 + } + ], + "connected": true + }, + { + "key": "maxToken", + "type": "custom", + "label": "回复上限", + "value": 16000, + "min": 100, + "max": 4000, + "step": 50, + "markList": [ + { + "label": "100", + "value": 100 + }, + { + "label": "4000", + "value": 4000 + } + ], + "connected": true + }, + { + "key": "systemPrompt", + "type": "textarea", + "label": "系统提示词", + "max": 300, + "valueType": "string", + "description": "模型固定的引导词，通过调整该内容，可以引导模型聊天方向。该内容会被固定在上下文的开头。可使用变量，例如 {{language}}", + "placeholder": "模型固定的引导词，通过调整该内容，可以引导模型聊天方向。该内容会被固定在上下文的开头。可使用变量，例如 {{language}}", + "value": "", + "connected": true + }, + { + "key": "quoteTemplate", + "type": "hidden", + "label": "引用内容模板", + "valueType": "string", + "value": "", + "connected": true + }, + { + "key": "quotePrompt", + "type": "hidden", + "label": "引用内容提示词", + "valueType": "string", + "value": "", + "connected": true + }, + { + "key": "switch", + "type": "target", + "label": "触发器", + "valueType": "any", + "connected": true + }, + { + "key": "quoteQA", + "type": "custom", + "label": "引用内容", + "description": "对象数组格式，结构：\n [{q:'问题',a:'回答'}]", + "valueType": "kb_quote", + "connected": false + }, + { + "key": "history", + "type": "target", + "label": "聊天记录", + "valueType": "chat_history", + "connected": true + }, + { + "key": "userChatInput", + "type": "target", + "label": "用户问题", + "required": true, + "valueType": "string", + "connected": true + }, + { + "key": "limitPrompt", + "type": "textarea", + "valueType": "string", + "label": "限定词", + "description": "限定模型对话范围，会被放置在本次提问前，拥有强引导和限定性。可使用变量，例如 {{language}}。引导例子:\n1. 知识库是关于 Laf 的介绍，参考知识库回答问题，与 \"Laf\" 无关内容，直接回复: \"我不知道\"。\n2. 你仅回答关于 \"xxx\" 的问题，其他问题回复: \"xxxx\"", + "placeholder": "限定模型对话范围，会被放置在本次提问前，拥有强引导和限定性。可使用变量，例如 {{language}}。引导例子:\n1. 知识库是关于 Laf 的介绍，参考知识库回答问题，与 \"Laf\" 无关内容，直接回复: \"我不知道\"。\n2. 你仅回答关于 \"xxx\" 的问题，其他问题回复: \"xxxx\"", + "value": "", + "connected": true + } + ], + "outputs": [ + { + "key": "answerText", + "label": "模型回复", + "description": "将在 stream 回复完毕后触发", + "valueType": "string", + "type": "source", + "targets": [] + }, + { + "key": "finish", + "label": "回复结束", + "description": "AI 回复完成后触发", + "valueType": "boolean", + "type": "source", + "targets": [] + } + ] + }, + { + "moduleId": "kq35bj", + "name": "用户引导", + "flowType": "userGuide", + "position": { + "x": 359.84546622310677, + "y": 686.3487640909323 + }, + "inputs": [ + { + "key": "welcomeText", + "type": "input", + "label": "开场白", + "value": "你好，我是你的全能助手，目前我拥有【查询天气】、【查看微博热搜】、【智能聊天】功能。来跟我对话吧~", + "connected": true + } + ], + "outputs": [] + } +] +``` + +{{% /details %}} + +## 效果图 + +![](/imgs/versatile_assistant_7.jpg) + +## 后记 + +1. 案例中的提示词不一定完美，如果有出现抽风的情况，可以自行调整提示词。 +2. 查询天气的 ai 对话，为了省 token 我用的是 GPT3.5，按理说 GPT4 理解力会高点，可以自行试试。 +3. 本案例中采用了“限定词”的方式引导 【AI 对话】模块，但最新版好像不支持限定词了（当然导入配置是没问题的），大家可以自行研究下新版的用法~