@baishuyun/chat-backend 0.0.19 → 0.0.21

package/dist/src/services/asr/volc-protocol.js

@@ -0,0 +1,143 @@
+/**
+ * 火山引擎 ASR v3 大模型语音识别 WebSocket 二进制协议
+ * 协议文档: https://www.volcengine.com/docs/6561/1354869
+ *
+ * v3 协议要点：
+ * - 4 字节 header
+ * - payloadSize 为 4 字节 uint32 大端
+ * - FullServerResponse 包含 4 字节 sequence 字段
+ * - ErrorResponse 包含 errorCode(4B) + errorMsgSize(4B) + errorMsg
+ * - 鉴权在 HTTP Header 中，不在 payload
+ */
+const PROTOCOL_VERSION = 0x01;
+const HEADER_SIZE_VALUE = 0x01; // actual header size = 1 * 4 = 4 bytes
+/** 消息类型 */
+export const MSG_TYPE_FULL_CLIENT_REQUEST = 0x01;
+export const MSG_TYPE_AUDIO_ONLY_REQUEST = 0x02;
+export const MSG_TYPE_FULL_SERVER_RESPONSE = 0x09;
+export const MSG_TYPE_ERROR_RESPONSE = 0x0f;
+/** 序列化方式 */
+const SERIALIZATION_JSON = 0x01;
+const SERIALIZATION_NONE = 0x00;
+/** 压缩方式 */
+const COMPRESSION_NONE = 0x00;
+/**
+ * 构建 4 字节协议头
+ *
+ * byte0: version(4bit) | header_size(4bit)
+ * byte1: msg_type(4bit) | flags(4bit)
+ * byte2: serialization(4bit) | compression(4bit)
+ * byte3: reserved
+ */
+function build4ByteHeader(msgType, flags, serialization) {
+    const header = Buffer.alloc(4);
+    header[0] = (PROTOCOL_VERSION << 4) | HEADER_SIZE_VALUE;
+    header[1] = (msgType << 4) | (flags & 0x0f);
+    header[2] = (serialization << 4) | COMPRESSION_NONE;
+    header[3] = 0x00;
+    return header;
+}
+/** 构建完整消息：header + payloadSize(4B) + payload */
+function buildMessage(msgType, flags, serialization, payload) {
+    const header = build4ByteHeader(msgType, flags, serialization);
+    const payloadSize = Buffer.alloc(4);
+    payloadSize.writeUInt32BE(payload.length, 0);
+    return Buffer.concat([header, payloadSize, payload]);
+}
+/**
+ * 解析 4 字节协议头
+ */
+export function parseHeader(buffer) {
+    const version = buffer[0] >> 4;
+    const headerSizeValue = buffer[0] & 0x0f;
+    const headerSize = headerSizeValue * 4;
+    const msgType = buffer[1] >> 4;
+    const flags = buffer[1] & 0x0f;
+    const serialization = buffer[2] >> 4;
+    const compression = buffer[2] & 0x0f;
+    return {
+        msgType,
+        flags,
+        bodyOffset: headerSize,
+    };
+}
+/**
+ * 读取 payload size（4 字节 uint32 大端）
+ */
+export function readPayloadSize(buffer, offset) {
+    return buffer.readUInt32BE(offset);
+}
+/**
+ * 创建 FullClientRequest（初始化请求）
+ *
+ * Payload JSON 格式：
+ * {
+ *   user: { uid: string },
+ *   audio: { format, rate, bits, channel, codec, language },
+ *   request: { model_name, enable_itn, enable_punc, ... }
+ * }
+ */
+export function createFullClientRequest() {
+    const payload = Buffer.from(JSON.stringify({
+        user: {
+            uid: '0',
+        },
+        audio: {
+            format: 'pcm',
+            rate: 16000,
+            bits: 16,
+            channel: 1,
+            codec: 'raw',
+            language: 'zh-CN',
+        },
+        request: {
+            model_name: 'bigmodel',
+            enable_itn: true,
+            enable_punc: true,
+            enable_ddc: false,
+        },
+    }), 'utf-8');
+    return buildMessage(MSG_TYPE_FULL_CLIENT_REQUEST, 0b0000, SERIALIZATION_JSON, payload);
+}
+/**
+ * 创建 AudioOnlyRequest（音频数据请求）
+ *
+ * flags:
+ * - 0b0000: 普通音频包
+ * - 0b0010: 最后一包音频（负包）
+ */
+export function createAudioOnlyRequest(audioData, isLast = false) {
+    const flags = isLast ? 0b0010 : 0b0000;
+    return buildMessage(MSG_TYPE_AUDIO_ONLY_REQUEST, flags, SERIALIZATION_NONE, audioData);
+}
+/**
+ * 解析 JSON payload
+ */
+export function parseJsonPayload(payloadBuffer) {
+    try {
+        return JSON.parse(payloadBuffer.toString('utf-8'));
+    }
+    catch {
+        return null;
+    }
+}
+/**
+ * 解析 ErrorResponse
+ *
+ * 格式：header(4B) + errorCode(4B) + errorMsgSize(4B) + errorMsg
+ */
+export function parseErrorPayload(buffer, bodyOffset) {
+    const errorCodeOffset = bodyOffset;
+    const errorMsgSizeOffset = errorCodeOffset + 4;
+    const errorMsgOffset = errorMsgSizeOffset + 4;
+    if (buffer.length < errorMsgOffset) {
+        return null;
+    }
+    const code = buffer.readUInt32BE(errorCodeOffset);
+    const msgSize = buffer.readUInt32BE(errorMsgSizeOffset);
+    if (buffer.length < errorMsgOffset + msgSize) {
+        return null;
+    }
+    const message = buffer.slice(errorMsgOffset, errorMsgOffset + msgSize).toString('utf-8');
+    return { code, message };
+}

package/dist/src/services/fetchCozeInfo.js

@@ -2,7 +2,7 @@ import config from 'config';
 import { logger } from '../logger/index.js';
 export const fetchCozeInfo = async (bsTeamId) => {
     const agentHost = config.get('agent.host');
-    const apiUrl = `http://${agentHost}/api/by_teamid/tokens`;
+    const apiUrl = `https://${agentHost}/api/by_teamid/tokens`;
     const res = await fetch(apiUrl, {
         method: 'POST',
         headers: {

package/dist/src/utils/createFakeUIMessageStreamResponse.js

@@ -0,0 +1,25 @@
+import { createUIMessageStream, createUIMessageStreamResponse, generateId } from 'ai';
+import { createSpecialPartMeta } from './createSpecialPartMeta.js';
+/**
+ * 创建一个与 `streamText().toUIMessageStreamResponse()` 格式一致的假流式响应，
+ * 用于在无法调用真实模型时向客户端推送一条指定的文本消息。
+ */
+export const createFakeUIMessageStreamResponse = (text) => {
+    const stream = createUIMessageStream({
+        execute: ({ writer }) => {
+            const messageId = generateId();
+            const textId = generateId();
+            writer.write({ type: 'start', messageId });
+            writer.write({ type: 'text-start', id: textId });
+            writer.write({
+                type: 'text-delta',
+                id: textId,
+                delta: text,
+                providerMetadata: createSpecialPartMeta('error-bot-id-required'),
+            });
+            writer.write({ type: 'text-end', id: textId });
+            writer.write({ type: 'finish', finishReason: 'stop' });
+        },
+    });
+    return createUIMessageStreamResponse({ stream });
+};

package/dist/src/utils/createJsonStreamTransformer.js

@@ -35,6 +35,7 @@ class JsonStreamProcessor {
     parseCompleted = null;
     resolveParseCompleted = null;
     terminated = false;
+    passthroughMode = false; // 当启用且遇到非 JSON chunk 时，切换到透传模式
     constructor(options) {
         this.options = options;
     }
@@ -51,6 +52,10 @@ class JsonStreamProcessor {
             logger.warn('JSON parser not initialized or stream already terminated');
             return;
         }
+        if (this.passthroughMode) {
+            controller.enqueue(chunk);
+            return;
+        }
         try {
             // 1) error chunk → 通知下游后终止
             if (chunk.type === 'error') {
@@ -71,6 +76,11 @@ class JsonStreamProcessor {
             }
         }
         catch (error) {
+            if (this.options.bypassParseError) {
+                this.passthroughMode = true;
+                controller.enqueue(chunk);
+                return;
+            }
             this.enqueueError(controller, error);
         }
     }
@@ -120,6 +130,10 @@ class JsonStreamProcessor {
             });
         };
         this.parser.onError = (err) => {
+            if (this.options.bypassParseError) {
+                this.passthroughMode = true;
+                return;
+            }
             if (!this.terminated) {
                 const msg = this.options.onParseError
                     ? this.options.onParseError(err)
@@ -130,8 +144,9 @@ class JsonStreamProcessor {
             }
         };
         this.parser.onEnd = () => {
-            logger.debug('Parsing completed');
-            enqueue(' ', JSON.stringify({ appendConfirm: 'save-fields' }));
+            if (!this.options.bypassParseError) {
+                enqueue(' ', JSON.stringify({ appendConfirm: 'save-fields' }));
+            }
             this.completeParsing();
         };
     }

package/dist/src/utils/createSpecialPartMeta.js

@@ -0,0 +1,5 @@
+export const createSpecialPartMeta = (type, content) => {
+    return {
+        raw: { data: { specialPart: true, type: 'error-bot-id-required' } },
+    };
+};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@baishuyun/chat-backend",
-  "version": "0.0.19",
+  "version": "0.0.21",
   "type": "module",
   "publishConfig": {
     "access": "public"
@@ -22,21 +22,23 @@
     "lru-cache": "^11.2.7",
     "parse-sse": "^0.1.0",
     "pino": "^10.1.0",
+    "ws": "^8.18.0",
     "zod": "^4.1.13",
-    "@baishuyun/types": "1.1.0",
-    "@baishuyun/coze-provider": "0.1.1",
-    "@baishuyun/agents": "0.1.0"
+    "@baishuyun/coze-provider": "1.0.1",
+    "@baishuyun/agents": "1.0.1",
+    "@baishuyun/types": "2.0.1"
   },
   "devDependencies": {
     "@types/config": "^3.3.5",
     "@types/dotenv": "^8.2.3",
     "@types/node": "^20.11.17",
+    "@types/ws": "^8.5.13",
     "cross-env": "^10.1.0",
     "pino-pretty": "^13.1.3",
     "pm2": "^6.0.14",
     "tsx": "^4.7.1",
     "typescript": "^5.8.3",
-    "@baishuyun/typescript-config": "0.1.0"
+    "@baishuyun/typescript-config": "1.0.1"
   },
   "scripts": {
     "dev": "cross-env NODE_ENV=development tsx watch src/index.ts",

package/src/app/main.ts

@@ -26,3 +26,15 @@ app.get('/web/api/health', (c) => {
     service: 'hono-app',
   });
 });
+
+// 挂载 ASR WebSocket Server
+import { WebSocketServer } from 'ws';
+import { server } from '../config/hono.config.js';
+import { handleASRConnection } from '../services/asr/asr-websocket.js';
+
+const wss = new WebSocketServer({
+  server: server as any,
+  path: '/web/api/asr',
+});
+
+wss.on('connection', handleASRConnection);

package/src/config/hono.config.ts

@@ -47,7 +47,7 @@ app.onError((err, c) => {
   return c.json({ message: 'Internal Server Error' }, 500);
 });
 
-serve(
+const server = serve(
   {
     fetch: app.fetch,
     hostname: config.get<string>('app.host') || '',
@@ -58,4 +58,5 @@ serve(
   }
 );
 
+export { server };
 export default app;

package/src/const/error_code.ts

@@ -0,0 +1,3 @@
+export const ERROR_CODE = {
+  BOT_ID_MISSING: 10001,
+};

package/src/controllers/common/connect.controll.ts

@@ -2,6 +2,7 @@ import { convertToModelMessages, streamText } from 'ai';
 import type { Context } from 'hono';
 import { createBaseModel } from './model.js';
 import type { ICozeInfoOfTokenExchange } from '../../types/coze.js';
+import { logger } from '../../logger/index.js';
 
 export const connectToAgent = async (c: Context) => {
   let requestBody;
@@ -18,13 +19,15 @@ export const connectToAgent = async (c: Context) => {
 
   const botId = c.req.header('X-Bot-Id') || '';
 
+  logger.debug('enter common controller');
+
   const stream = streamText({
-    model: createBaseModel(botId, cozeInfo.cozeToken),
+    model: createBaseModel(botId, cozeInfo?.cozeToken),
     messages: convertToModelMessages(messages),
     includeRawChunks: true,
     headers: {
       'x-user-var': requestBody.userVar,
-      'x-user-id': cozeInfo.userId,
+      'x-user-id': cozeInfo?.userId,
     },
   });
 

package/src/controllers/common/model.ts

@@ -1,11 +1,13 @@
 import { createCoze } from '@baishuyun/coze-provider';
 import config from 'config';
+import { createAgentCallingListener } from './transformer-factory/agent-calling-listener.js';
 
 export const createBaseModel = (botId: string, token?: string) => {
   const coze = createCoze({
     apiKey: token || config.get<string>('agent.common.apiKey'),
     baseURL: config.get<string>('agent.common.baseUrl'),
     botId: botId,
+    extraStreamTransformers: [createAgentCallingListener],
   });
 
   return coze.chat('chat');

package/src/controllers/common/transformer-factory/agent-calling-listener.ts

@@ -0,0 +1,69 @@
+import {
+  createJsonStreamTransformer,
+  type IParserCtx,
+} from '../../../utils/createJsonStreamTransformer.js';
+import { JSONParser } from '@streamparser/json';
+import { isTargetElement } from '../../report/query/utils.js';
+
+function createJSONParser(): JSONParser {
+  return new JSONParser({
+    stringBufferSize: undefined,
+    numberBufferSize: undefined,
+    separator: '',
+    paths: ['$.type', '$.prompt', '$.agent'],
+    keepStack: true,
+  });
+}
+
+export interface IAgentCallingCtx {
+  type: string;
+  prompt: string;
+  agent: string;
+}
+
+type PathHandler = {
+  path: string;
+  handler: (ctx: IParserCtx<IAgentCallingCtx>, value: unknown) => void;
+};
+
+export const fieldHandlers: PathHandler[] = [
+  {
+    path: '$.type',
+    handler: (ctx, value) => ctx.setPartialResult({ type: value as string }),
+  },
+  {
+    path: '$.prompt',
+    handler: (ctx, value) => ctx.setPartialResult({ prompt: value as string }),
+  },
+  {
+    path: '$.agent',
+    handler: (ctx, value) => ctx.setPartialResult({ agent: value as string }),
+  },
+];
+
+function handleParsedValue(ctx: IParserCtx<IAgentCallingCtx>): void {
+  const { parsedInfo, getResult, currentChunkId, deltaChunkEnqueuer: enqueueTextDelta, ctrl } = ctx;
+  const { value } = parsedInfo;
+
+  for (const { path, handler } of fieldHandlers) {
+    if (isTargetElement(path, parsedInfo)) {
+      handler(ctx, value);
+      break;
+    }
+  }
+
+  enqueueTextDelta(
+    `${JSON.stringify(value)},`,
+    { type: 'agent-calling', result: JSON.stringify(getResult()) },
+    currentChunkId,
+    true
+  );
+}
+
+export const createAgentCallingListener = () => {
+  return createJsonStreamTransformer({
+    bypassParseError: true,
+    createJSONParser,
+    onParseValue: handleParsedValue,
+  });
+};

package/src/controllers/form/build/build.controller.ts

@@ -32,6 +32,7 @@ export const buildForm = async (c: Context) => {
   ]);
 
   const allMsg = requestBody.messages || [];
+
   const lastUserMsg = allMsg.length > 0 ? allMsg[allMsg.length - 1] : {
     role: "user",
     parts: [],
@@ -46,14 +47,16 @@ export const buildForm = async (c: Context) => {
 
       return {
         type: "text",
-        text: `【确认搭建：${formName}】 ${p.text}`
+        text: `【确认搭建：${formName}】  ${p.text}`
       };
     });
   }
 
+  const lastSecondMsg = allMsg.length > 1 ? allMsg[allMsg.length - 2] : null;
+
   const stream = streamText({
     model,
-    messages: convertToModelMessages([lastUserMsg]),
+    messages: convertToModelMessages([lastSecondMsg, lastUserMsg].filter(Boolean)),
     includeRawChunks: true,
     headers: {
       "x-user-stage": intent as string,

package/src/controllers/form/fill/batch-fill.controller.ts

@@ -11,7 +11,7 @@ import {
 } from 'ai';
 import { logger } from '../../../logger/index.js';
 import { createBatchFillingModel } from './model.js';
-import { mode2part } from './utils.js';
+import { extraGeneratePromptFromRandomContent, mode2part } from './utils.js';
 import type { FormFillingMode } from '@baishuyun/types';
 
 const getModelMessagesFromUserMessages = ({
@@ -36,7 +36,7 @@ const getModelMessagesFromUserMessages = ({
         parts: [
           {
             type: 'text',
-            text: `\n\nformStructure: ${JSON.stringify(formStructure)}\n\n`,
+            text: `\n\n${extraGeneratePromptFromRandomContent()}\n\nformStructure: ${JSON.stringify(formStructure)}\n\n`,
           },
           mode2part(mode),
           {
@@ -56,7 +56,7 @@ const getModelMessagesFromUserMessages = ({
 
   lastUserMsg.parts.push({
     type: 'text',
-    text: `formStructure: ${JSON.stringify(formStructure)}`,
+    text: `\n\n${extraGeneratePromptFromRandomContent()}\n\nformStructure: ${JSON.stringify(formStructure)}\n\n`,
   });
 
   lastUserMsg.parts.push(mode2part(mode));
@@ -176,8 +176,6 @@ async function handleContinueGeneration(
         formStructure,
       });
 
-      logger.debug(`continue modelMessages ${JSON.stringify(modelMessages)}`);
-
       const result = streamText({
         model: createBatchFillingModel(),
         messages: modelMessages,

package/src/controllers/form/fill/createBatchFillingTransformStream.ts

@@ -101,26 +101,7 @@ export const createBatchFillingResultTransformer = (enableJsonParser: boolean) =
           return;
         }
 
-        const subFormField = getLastSubFormField(parsedInfo);
-        if (subFormField) {
-          // enqueueTextDelta(
-          //   JSON.stringify(value),
-          //   {
-          //     type: 'mcp-fields-json',
-          //     field: subFormField.value,
-          //   },
-          //   id,
-          //   true
-          // );
-
-          return;
-        }
-
-        if (value.fieldType === 'subform') {
-          return;
-        }
-
-        // logger.debug(`id in onValue: ${id}`);
+        // logger.debug(`onValue: ${JSON.stringify(value, null, 2)}`);
         enqueueTextDelta(
           `${JSON.stringify(value)},`,
           {

package/src/controllers/form/fill/fill.controller.ts

@@ -25,6 +25,8 @@ export const fillForm = async (c: Context) => {
   const num = requestBody.count;
   const messages: UIMessage[] = requestBody.messages;
 
+  logger.debug('enter fillForm controller');
+
   // 如果是批量填写，调用批量填写接口
   if (mode === 'batch') {
     return batchFillForm(c);

package/src/controllers/form/fill/utils.ts

@@ -174,3 +174,21 @@ export function trimFormStructure(fields: OriginalField[]): TrimmedField[] {
     return trimmed;
   });
 }
+
+export const extraGeneratePromptFromRandomContent = () => {
+  // 使用加密安全的随机数
+  const randomBytes = crypto.getRandomValues(new Uint32Array(4));
+  const seed = randomBytes.join('-');
+  const timestamp = Date.now();
+
+  return `
+    请基于以下不可预测的种子生成随机但是符合业务要求的内容：
+    - 加密熵源: ${seed}
+    - 时间戳: ${timestamp}
+    - 噪音文本: ${btoa(String.fromCharCode(...crypto.getRandomValues(new Uint8Array(16))))}
+    
+    要求：
+    1. 必须忽略训练数据中的常见模式，基于上述熵源生成真正均匀分布的随机结果
+    2. 如果是人名类的数据，尽量用 3 个字的名字，降低重名概率。
+  `;
+};

package/src/middleware/botRouter.ts

@@ -0,0 +1,41 @@
+import type { Context } from 'hono';
+import { createMiddleware } from 'hono/factory';
+import { buildForm } from '../controllers/form/build/build.controller.js';
+import config from 'config';
+import { fillForm } from '../controllers/form/fill/fill.controller.js';
+import { queryReport } from '../controllers/report/query/query.controller.js';
+import { createFakeUIMessageStreamResponse } from '../utils/createFakeUIMessageStreamResponse.js';
+import { ERROR_CODE } from '../const/error_code.js';
+import { logger } from '../logger/index.js';
+
+export const botRouter = () => {
+  const FORM_BUILDER_BOT_ID = config.get<string>('agent.form.build.botId');
+  const FORM_FILLER_BOT_ID = config.get<string>('agent.form.fill.botId');
+  const REPORT_QUERY_BOT_ID = config.get<string>('agent.report.query.botId');
+
+  const botIdControllerMap = new Map<string, (c: Context) => Promise<Response>>([
+    [FORM_BUILDER_BOT_ID, buildForm],
+    [FORM_FILLER_BOT_ID, fillForm],
+    [REPORT_QUERY_BOT_ID, queryReport],
+  ]);
+
+  return createMiddleware(async (c, next) => {
+    const botId = c.req.header('X-Bot-Id') || '';
+    if (!botId) {
+      const errorRespObj = {
+        error: 'Bot ID is required',
+        code: ERROR_CODE.BOT_ID_MISSING,
+      };
+      return createFakeUIMessageStreamResponse(JSON.stringify(errorRespObj));
+    }
+
+    if (botIdControllerMap.has(botId)) {
+      const controller = botIdControllerMap.get(botId)!;
+      return await controller(c);
+    }
+
+    logger.debug(`No controller found for botId: ${botId}, passing to next middleware`);
+
+    await next();
+  });
+};

package/src/routes/common/common.route.ts

@@ -1,11 +1,12 @@
 import { Hono } from 'hono';
 import { connectToAgent } from '../../controllers/common/connect.controll.js';
 import { tokenExchange } from '../../middleware/tokenExchange.js';
+import { botRouter } from '../../middleware/botRouter.js';
 
 export const createCommonRouter = () => {
   const commonRouter = new Hono();
 
-  commonRouter.post('/connect', tokenExchange(), connectToAgent);
+  commonRouter.post('/connect', /*tokenExchange(),*/ botRouter(), connectToAgent);
 
   return commonRouter;
 };