feat: 添加网页搜索功能，优化意图识别模式，更新依赖项

.vscode/settings.json

@@ -1,5 +1,7 @@
 {
   "cSpell.words": [
+    "BOCHA",
+    "bochaai",
     "bunx",
     "CEINTL",
     "Chakroun",

constant/function.ts

@@ -29,6 +29,11 @@ const functionMap = {
     xAuthor: "wangyifei15 🕹️ Yingbo",
     xIcon: "🕹️",
   },
+  searchAgent: {
+    xName: "小煎蛋 Search Agent",
+    xAuthor: "Yingbo",
+    xIcon: "🔍",
+  },
 }
 
 export default functionMap

controller/intentAgent/index.ts

@@ -22,6 +22,9 @@ const BriefingSchema = BaseIntentSchema.extend({
   userDescription: z.string().min(1),
 })
 
+/**
+ * 简报存储链接模式
+ */
 const BriefingLinkSchema = z.object({
   intent: z.literal(14),
   link: z.string().url(),
@@ -35,6 +38,14 @@ const CommonResponseSchema = BaseIntentSchema.extend({
   message: z.string().min(1),
 })
 
+/**
+ * 联网检索模式
+ */
+const NetSearchSchema = BaseIntentSchema.extend({
+  intent: z.literal(15),
+  query: z.string().min(1),
+})
+
 /**
  * 意图模式
  */
@@ -43,12 +54,13 @@ const IntentSchema = z.union([
   CommonResponseSchema,
   BaseIntentSchema,
   BriefingLinkSchema,
+  NetSearchSchema,
 ])
 
-type BaseIntent = z.infer<typeof BaseIntentSchema>
 type Briefing = z.infer<typeof BriefingSchema>
 type BriefingLink = z.infer<typeof BriefingLinkSchema>
 type CommonResponse = z.infer<typeof CommonResponseSchema>
+type NetSearch = z.infer<typeof NetSearchSchema>
 export type Intent = z.infer<typeof IntentSchema>
 
 /**
@@ -58,8 +70,8 @@ const jsonSchema = zodToJsonSchema(IntentSchema)
 
 /**
  * 代理函数
- * @param {Context} ctx - 上下文对象
+ * @param ctx - 上下文对象
- * @returns {Promise<Intent>} - 返回意图对象
+ * @returns 返回意图对象
  */
 const agent = async (ctx: Context): Promise<Intent> => {
   const {
@@ -71,7 +83,7 @@ const agent = async (ctx: Context): Promise<Intent> => {
   let attempts = 0
   while (attempts < 3) {
     const res = await llm.invoke(
-      "intentRecognitionNext",
+      "intentRecognition",
       {
         userInput: msgText,
         time: new Date().toLocaleString("zh-CN", { timeZone: "Asia/Shanghai" }),
@@ -105,31 +117,51 @@ const agent = async (ctx: Context): Promise<Intent> => {
   }
 }
 
-const isBaseIntent = (intent: Intent): intent is BaseIntent => {
+/**
-  return intent.intent !== 5 && intent.intent !== 2
+ * 判断是否为简报意图
-}
+ * @param intent - 意图对象
-
+ * @returns 如果是简报意图则返回true，否则返回false
+ */
 const isBriefing = (intent: Intent): intent is Briefing => {
   return intent.intent === 5
 }
 
+/**
+ * 判断是否为简报链接意图
+ * @param intent - 意图对象
+ * @returns 如果是简报链接意图则返回true，否则返回false
+ */
 const isBriefingLink = (intent: Intent): intent is BriefingLink => {
   return intent.intent === 14
 }
 
+/**
+ * 判断是否为通用响应意图
+ * @param intent - 意图对象
+ * @returns 如果是通用响应意图则返回true，否则返回false
+ */
 const isCommonResponse = (intent: Intent): intent is CommonResponse => {
   return intent.intent === 2
 }
 
+/**
+ * 判断是否为联网检索意图
+ * @param intent - 意图对象
+ * @returns 如果是联网检索意图则返回true，否则返回false
+ */
+const isNetSearch = (intent: Intent): intent is NetSearch => {
+  return intent.intent === 15
+}
+
 /**
  * 意图代理对象
  */
 const intentAgent = {
   agent,
-  isBaseIntent,
   isBriefing,
   isCommonResponse,
   isBriefingLink,
+  isNetSearch,
 }
 
 export default intentAgent

controller/reportAgent/index.ts

@@ -154,7 +154,7 @@ const agent = async (ctx: Context, link: string, userDescription: string) => {
     validateLink(link)
     // 发送一个loading卡片
     await message.updateOrReply(
-      cardGender.genSuccessCard("正在为您收集简报，请稍等片刻~")
+      cardGender.genPendingCard("正在为您收集简报，请稍等片刻~")
     )
     // // 抓取网页
     // const crawRes = await crawlWebPage(ctx, link)

controller/searchAgent/index.ts

@@ -0,0 +1,70 @@
+import { Context } from "../../types"
+import llm from "../../utils/llm"
+
+/**
+ * 生成网页简报
+ * @param {Context} ctx - 上下文对象
+ * @param {string} userDescription - 用户描述
+ * @param {any} content - 网页内容
+ * @returns {Promise<string>} - 返回简报内容
+ * @throws {Error} - 当生成简报失败时抛出错误
+ */
+export const generateAnswer = async (
+  ctx: Context,
+  webSearchResults:
+    | {
+        siteName: string
+        summary: string
+      }[]
+    | null
+): Promise<string> => {
+  const {
+    requestId,
+    larkBody: { msgText },
+  } = ctx
+  const llmRes = (await llm.invoke(
+    "searchAgent",
+    {
+      webSearchResults,
+      time: new Date().toLocaleString("zh-CN", { timeZone: "Asia/Shanghai" }),
+      userInput: msgText,
+    },
+    requestId,
+    1
+  )) as string
+  if (!llmRes) throw new Error("模型侧错误")
+  return llmRes
+}
+
+const agent = async (ctx: Context, query: string) => {
+  const {
+    larkService: { message },
+    attachService,
+    larkCard,
+    logger,
+  } = ctx
+  const cardGender = larkCard.child("searchAgent")
+  try {
+    // 发送一个loading卡片
+    await message.updateOrReply(
+      cardGender.genPendingCard("正在检索网络信息，请稍等片刻~")
+    )
+    const searchRes = await attachService.webSearch(query)
+    await message.updateOrReply(
+      cardGender.genPendingCard("LLM输出中，请稍等...")
+    )
+    const answer = await generateAnswer(ctx, searchRes)
+    await message.updateOrReply(cardGender.genSuccessCard(answer))
+  } catch (error: any) {
+    logger.error(`searchAgent error: ${error}`)
+    await message.updateOrReply(
+      cardGender.genErrorCard(`检索失败: ${error.message}`)
+    )
+  }
+}
+
+const searchAgent = {
+  agent,
+}
+
+export default searchAgent

package.json

@@ -17,12 +17,12 @@
     ]
   },
   "devDependencies": {
-    "@commitlint/cli": "^19.6.1",
+    "@commitlint/cli": "^19.7.1",
-    "@commitlint/config-conventional": "^19.6.0",
+    "@commitlint/config-conventional": "^19.7.1",
     "@eslint/js": "^9.19.0",
     "@types/node-schedule": "^2.1.7",
     "@types/uuid": "^10.0.0",
-    "bun-types": "^1.2.1",
+    "bun-types": "^1.2.2",
     "eslint": "^9.19.0",
     "eslint-plugin-simple-import-sort": "^12.1.1",
     "eslint-plugin-unused-imports": "^4.1.4",
@@ -30,7 +30,7 @@
     "lint-staged": "^15.4.3",
     "oxlint": "^0.13.2",
     "prettier": "^3.4.2",
-    "typescript-eslint": "^8.22.0"
+    "typescript-eslint": "^8.23.0"
   },
   "peerDependencies": {
     "typescript": "^5.5.4"
@@ -39,13 +39,13 @@
     "@egg/hooks": "^1.2.0",
     "@egg/lark-msg-tool": "^1.21.0",
     "@egg/logger": "^1.6.0",
-    "@egg/net-tool": "^1.31.1",
+    "@egg/net-tool": "^1.31.2",
     "@egg/path-tool": "^1.4.1",
-    "@langchain/core": "^0.3.36",
+    "@langchain/core": "^0.3.38",
-    "@langchain/langgraph": "^0.2.41",
+    "@langchain/langgraph": "^0.2.44",
     "@langchain/openai": "^0.3.17",
     "joi": "^17.13.3",
-    "langfuse-langchain": "^3.32.3",
+    "langfuse-langchain": "^3.35.1",
     "node-schedule": "^2.1.1",
     "p-limit": "^6.2.0",
     "pocketbase": "^0.23.0",

routes/bot/eventMsg.ts

@@ -1,6 +1,7 @@
 import groupAgent from "../../controller/groupAgent"
 import intentAgent from "../../controller/intentAgent"
 import reportAgent from "../../controller/reportAgent"
+import searchAgent from "../../controller/searchAgent"
 import soupAgent from "../../controller/soupAgent"
 import { Context } from "../../types"
 import { isNotP2POrAtBot } from "../../utils/message"
@@ -68,69 +69,69 @@ const manageIntent = async (ctx: Context) => {
   try {
     const intentRes = await intentAgent.agent(ctx)
     logger.info(`intentRes: ${JSON.stringify(intentRes)}`)
+    // 链接总结
     if (intentAgent.isBriefing(intentRes)) {
       reportAgent
         .agent(ctx, intentRes.link, intentRes.userDescription)
         .catch(() => null)
       return
     }
+    // 设置链接总结的结果存储表格
     if (intentAgent.isBriefingLink(intentRes)) {
       reportAgent.setSheet(ctx, intentRes.link).catch(() => null)
       return
     }
+    // 网络检索
+    if (intentAgent.isNetSearch(intentRes)) {
+      searchAgent.agent(ctx, intentRes.query).catch(() => null)
+      return
+    }
+    // 通用回复
     if (intentAgent.isCommonResponse(intentRes)) {
       await message.updateOrReply(larkCard.genSuccessCard(intentRes.message))
       return
     }
-    if (intentAgent.isBaseIntent(intentRes)) {
+    switch (intentRes.intent) {
-      switch (intentRes.intent) {
+      case 3:
-        case 3:
+        await message.updateOrReply(
-          await message.updateOrReply(
+          larkCard.genTempCard("chatId", { chat_id: chatId }) as string
-            larkCard.genTempCard("chatId", { chat_id: chatId }) as string
+        )
-          )
+        break
-          break
+      case 4:
-        case 4:
+        await attachService.ciMonitor(chatId)
-          await attachService.ciMonitor(chatId)
+        break
-          break
+      case 6:
-        case 6:
+        await message.updateOrReply(
-          await message.updateOrReply(
+          larkCard.genTempCard("eggGuide", { chat_id: chatId }) as string
-            larkCard.genTempCard("eggGuide", { chat_id: chatId }) as string
+        )
-          )
+        break
-          break
+      case 7:
-        case 7:
+        groupAgent.report.setSubscription(ctx, "daily", true).catch(() => null)
-          groupAgent.report
+        break
-            .setSubscription(ctx, "daily", true)
+      case 8:
-            .catch(() => null)
+        groupAgent.report.setSubscription(ctx, "weekly", true).catch(() => null)
-          break
+        break
-        case 8:
+      case 9:
-          groupAgent.report
+        groupAgent.report.setSubscription(ctx, "daily", false).catch(() => null)
-            .setSubscription(ctx, "weekly", true)
+        break
-            .catch(() => null)
+      case 10:
-          break
+        groupAgent.report
-        case 9:
+          .setSubscription(ctx, "weekly", false)
-          groupAgent.report
+          .catch(() => null)
-            .setSubscription(ctx, "daily", false)
+        break
-            .catch(() => null)
+      case 11:
-          break
+        groupAgent.report.gen4Test(ctx, "daily").catch(() => null)
-        case 10:
+        break
-          groupAgent.report
+      case 12:
-            .setSubscription(ctx, "weekly", false)
+        groupAgent.report.gen4Test(ctx, "weekly").catch(() => null)
-            .catch(() => null)
+        break
-          break
+      case 13:
-        case 11:
+        soupAgent.startOrStopGame(ctx, true, "manual").catch(() => null)
-          groupAgent.report.gen4Test(ctx, "daily").catch(() => null)
+        break
-          break
+      case 1:
-        case 12:
+      default:
-          groupAgent.report.gen4Test(ctx, "weekly").catch(() => null)
+        groupAgent.agent(ctx).catch(() => null)
-          break
+        break
-        case 13:
-          soupAgent.startOrStopGame(ctx, true, "manual").catch(() => null)
-          break
-        case 1:
-        default:
-          groupAgent.agent(ctx).catch(() => null)
-          break
-      }
     }
   } catch (error) {
     logger.error(`manageIntent error: ${error}`)

services/attach/index.ts

@@ -142,6 +142,37 @@ class AttachService extends NetToolBase {
         throw new Error("MIFY爬虫请求失败")
       })
   }
+
+  /**
+   * 使用bochaai搜索网页
+   * @param {string} query - 搜索关键词
+   * @returns  返回搜索结果
+   */
+  async webSearch(query: string) {
+    const URL = "https://api.bochaai.com/v1/web-search"
+    return this.post(
+      URL,
+      {
+        query,
+        summary: true,
+      },
+      {},
+      {
+        Authorization: `Bearer ${APP_CONFIG.BOCHA_SK}`,
+      }
+    )
+      .then((res) => {
+        const { value } = res.data.webPages
+        return value.map(({ siteName, summary }: any) => ({
+          siteName,
+          summary,
+        })) as {
+          siteName: string
+          summary: string
+        }[]
+      })
+      .catch(() => null)
+  }
 }
 
 export default AttachService

test/llm/intentRecognition.ts

@@ -8,34 +8,63 @@ import { cleanLLMRes } from "../../utils/llm/base"
 
 await initAppConfig()
 
+/**
+ * 基础意图模式
+ */
 const BaseIntentSchema = z.object({
   intent: z.number().min(1).max(13),
 })
 
+/**
+ * 简报模式
+ */
 const BriefingSchema = BaseIntentSchema.extend({
-  intent: z.literal(3),
+  intent: z.literal(5),
   link: z.string().url(),
   userDescription: z.string().min(1),
 })
 
+/**
+ * 简报存储链接模式
+ */
+const BriefingLinkSchema = z.object({
+  intent: z.literal(14),
+  link: z.string().url(),
+})
+
+/**
+ * 通用响应模式
+ */
 const CommonResponseSchema = BaseIntentSchema.extend({
-  intent: z.literal(12),
+  intent: z.literal(2),
   message: z.string().min(1),
 })
 
+/**
+ * 联网检索模式
+ */
+const NetSearchSchema = BaseIntentSchema.extend({
+  intent: z.literal(15),
+  query: z.string().min(1),
+})
+
+/**
+ * 意图模式
+ */
 const IntentSchema = z.union([
   BriefingSchema,
   CommonResponseSchema,
   BaseIntentSchema,
+  BriefingLinkSchema,
+  NetSearchSchema,
 ])
 
 const jsonSchema = zodToJsonSchema(IntentSchema)
 
 const res = await llm.invoke(
-  "intentRecognitionNext",
+  "intentRecognition",
   {
-    userInput:
+    userInput: "今天是哪天",
-      "https://mp.weixin.qq.com/s/-0J8XbXJU6Bu-UihRtgGAQ Airbnb死磕React Native惨败，微软却玩出花！Office、Outlook全线接入，Copilot成最大赢家 推荐大家看一下，rn助力微软copilot 跨平台实现",
     time: new Date().toLocaleString("zh-CN", { timeZone: "Asia/Shanghai" }),
     jsonSchema,
   },

test/server/getBatchUserInfo.ts

@@ -0,0 +1,11 @@
+import initAppConfig from "../../constant/config"
+import genLarkService from "../../utils/genLarkService"
+
+await initAppConfig()
+
+const larkService = genLarkService("egg", "test")
+
+larkService.user
+  .batchGet(["ou_5d9c2da2870802fc47fc2066f28b1b49"], "open_id")
+  .then(console.log)
+  .catch(console.error)

test/server/webSearch.ts

@@ -0,0 +1,8 @@
+import initAppConfig from "../../constant/config"
+import { AttachService } from "../../services"
+
+await initAppConfig()
+
+const service = new AttachService()
+
+service.webSearch("北京今天天气").then(console.log).catch(console.error)