@lobehub/chat 1.79.4 → 1.79.5

package/CHANGELOG.md

@@ -2,6 +2,31 @@
 
 # Changelog
 
+### [Version 1.79.5](https://github.com/lobehub/lobe-chat/compare/v1.79.4...v1.79.5)
+
+<sup>Released on **2025-04-10**</sup>
+
+#### 🐛 Bug Fixes
+
+- **misc**: Fix langfuse intergation.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **misc**: Fix langfuse intergation, closes [#7367](https://github.com/lobehub/lobe-chat/issues/7367) ([22b5236](https://github.com/lobehub/lobe-chat/commit/22b5236))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ### [Version 1.79.4](https://github.com/lobehub/lobe-chat/compare/v1.79.3...v1.79.4)
 
 <sup>Released on **2025-04-10**</sup>

package/changelog/v1.json

@@ -1,4 +1,13 @@
 [
+  {
+    "children": {
+      "fixes": [
+        "Fix langfuse intergation."
+      ]
+    },
+    "date": "2025-04-10",
+    "version": "1.79.5"
+  },
   {
     "children": {
       "improvements": [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.79.4",
+  "version": "1.79.5",
   "description": "Lobe Chat - an open-source, high-performance chatbot framework that supports speech synthesis, multimodal, and extensible Function Call plugin system. Supports one-click free deployment of your private ChatGPT/LLM web application.",
   "keywords": [
     "framework",
@@ -189,8 +189,8 @@
     "js-sha256": "^0.11.0",
     "jsonl-parse-stringify": "^1.0.3",
     "langchain": "^0.3.19",
-    "langfuse": "3.29.1",
-    "langfuse-core": "3.29.1",
+    "langfuse": "^3.37.1",
+    "langfuse-core": "^3.37.1",
     "lodash-es": "^4.17.21",
     "lucide-react": "^0.487.0",
     "mammoth": "^1.9.0",

package/src/app/(backend)/webapi/chat/[provider]/route.ts

@@ -33,10 +33,7 @@ export const POST = checkAuth(async (req: Request, { params, jwtPayload, createR
     let traceOptions = {};
     // If user enable trace
     if (tracePayload?.enabled) {
-      traceOptions = createTraceOptions(data, {
-        provider,
-        trace: tracePayload,
-      });
+      traceOptions = createTraceOptions(data, { provider, trace: tracePayload });
     }
 
     return await agentRuntime.chat(data, {

package/src/app/(backend)/webapi/plugin/gateway/route.ts

@@ -52,6 +52,7 @@ export const POST = async (req: Request) => {
     return createErrorResponse(result.error as ErrorType);
   }
 
+  // TODO: need to be replace by better telemetry system
   // add trace
   const tracePayload = getTracePayload(req);
   const traceClient = new TraceClient();

package/src/app/(backend)/webapi/trace/route.ts

@@ -1,3 +1,5 @@
+import { after } from 'next/server';
+
 import { TraceEventType } from '@/const/trace';
 import { TraceClient } from '@/libs/traces';
 import { TraceEventBasePayload, TraceEventPayloads } from '@/types/trace';
@@ -35,6 +37,9 @@ export const POST = async (req: Request) => {
     }
   }
 
-  await traceClient.shutdownAsync();
+  after(async () => {
+    await traceClient.shutdownAsync();
+  });
+
   return new Response(undefined, { status: 201 });
 };

package/src/const/trace.ts

@@ -7,15 +7,13 @@ export enum TraceNameMap {
   Conversation = 'Conversation',
   EmojiPicker = 'Emoji Picker',
   FetchPluginAPI = 'Fetch Plugin API',
-  InvokePlugin = 'Invoke Plugin',
   LanguageDetect = 'Language Detect',
+  SearchIntentRecognition = 'Search Intent Recognition',
   SummaryAgentDescription = 'Summary Agent Description',
   SummaryAgentTags = 'Summary Agent Tags',
   SummaryAgentTitle = 'Summary Agent Title',
   SummaryTopicTitle = 'Summary Topic Title',
   Translator = 'Translator',
-  // mean user have relative events
-  UserEvents = 'User Events',
 }
 
 export enum TraceEventType {
@@ -33,7 +31,7 @@ export enum TraceTagMap {
   Chat = 'Chat Competition',
   SystemChain = 'System Chain',
   ToolCalling = 'Tool Calling',
-  ToolsCall = 'Tools Call',
+  ToolsCalling = 'Tools Calling',
 }
 
 export interface TracePayload {

package/src/database/models/__tests__/_util.ts

@@ -1,11 +1,13 @@
 import { clientDB, initializeDB } from '@/database/client/db';
-import { getTestDBInstance } from '@/database/core/dbForTest';
 import { LobeChatDatabase } from '@/database/type';
 
 const isServerDBMode = process.env.TEST_SERVER_DB === '1';
 
 export const getTestDB = async () => {
-  if (isServerDBMode) return await getTestDBInstance();
+  if (isServerDBMode) {
+    const { getTestDBInstance } = await import('@/database/core/dbForTest');
+    return await getTestDBInstance();
+  }
 
   await initializeDB();
   return clientDB as LobeChatDatabase;

package/src/libs/agent-runtime/AgentRuntime.test.ts

@@ -66,10 +66,7 @@ const testRuntime = (providerId: string, payload?: any) => {
 let mockModelRuntime: AgentRuntime;
 beforeEach(async () => {
   const jwtPayload: JWTPayload = { apiKey: 'user-openai-key', baseURL: 'user-endpoint' };
-  mockModelRuntime = await AgentRuntime.initializeWithProvider(
-    ModelProvider.OpenAI,
-    jwtPayload,
-  );
+  mockModelRuntime = await AgentRuntime.initializeWithProvider(ModelProvider.OpenAI, jwtPayload);
 });
 
 describe('AgentRuntime', () => {
@@ -112,7 +109,7 @@ describe('AgentRuntime', () => {
         provider: 'openai',
         trace: {
           traceId: 'test-trace-id',
-          traceName: TraceNameMap.Conversation,
+          traceName: TraceNameMap.SummaryTopicTitle,
           sessionId: 'test-session-id',
           topicId: 'test-topic-id',
           tags: [],
@@ -136,7 +133,7 @@ describe('AgentRuntime', () => {
         provider: 'openai',
         trace: {
           traceId: 'test-trace-id',
-          traceName: TraceNameMap.Conversation,
+          traceName: TraceNameMap.SummaryTopicTitle,
           sessionId: 'test-session-id',
           topicId: 'test-topic-id',
           tags: [],
@@ -147,7 +144,7 @@ describe('AgentRuntime', () => {
 
       const updateMock = vi.fn();
 
-      it('should call experimental_onToolCall correctly', async () => {
+      it('should call onToolsCalling correctly', async () => {
         vi.spyOn(langfuseCfg, 'getLangfuseConfig').mockReturnValue({
           ENABLE_LANGFUSE: true,
           LANGFUSE_PUBLIC_KEY: 'abc',
@@ -157,9 +154,9 @@ describe('AgentRuntime', () => {
         // 使用 spyOn 模拟 chat 方法
         vi.spyOn(LobeOpenAI.prototype, 'chat').mockImplementation(
           async (payload, { callback }: any) => {
-            // 模拟 experimental_onToolCall 回调的触发
-            if (callback?.experimental_onToolCall) {
-              await callback.experimental_onToolCall();
+            // 模拟 onToolCall 回调的触发
+            if (callback?.onToolsCalling) {
+              await callback.onToolsCalling();
             }
             return new Response('abc');
           },
@@ -168,7 +165,7 @@ describe('AgentRuntime', () => {
 
         await mockModelRuntime.chat(payload, createTraceOptions(payload, options));
 
-        expect(updateMock).toHaveBeenCalledWith({ tags: ['Tools Call'] });
+        expect(updateMock).toHaveBeenCalledWith({ tags: ['Tools Calling'] });
       });
       it('should call onStart correctly', async () => {
         vi.spyOn(langfuseCfg, 'getLangfuseConfig').mockReturnValue({
@@ -204,7 +201,7 @@ describe('AgentRuntime', () => {
         vi.spyOn(LobeOpenAI.prototype, 'chat').mockImplementation(
           async (payload, { callback }: any) => {
             if (callback?.onCompletion) {
-              await callback.onCompletion('Test completion');
+              await callback.onCompletion({ text: 'Test completion' });
             }
             return new Response('Success');
           },
@@ -215,14 +212,11 @@ describe('AgentRuntime', () => {
         // Verify onCompletion was called with expected output
         expect(updateMock).toHaveBeenCalledWith({
           endTime: expect.any(Date),
-          metadata: {
-            provider: 'openai',
-            tools: undefined,
-          },
+          metadata: {},
           output: 'Test completion',
         });
       });
-      it('should call onFinal correctly', async () => {
+      it.skip('should call onFinal correctly', async () => {
         vi.spyOn(langfuseCfg, 'getLangfuseConfig').mockReturnValue({
           ENABLE_LANGFUSE: true,
           LANGFUSE_PUBLIC_KEY: 'abc',

package/src/libs/agent-runtime/helpers/index.ts

@@ -0,0 +1 @@
+export * from './parseToolCalls';

package/src/{utils/fetch/__tests__ → libs/agent-runtime/helpers}/parseToolCalls.test.ts

@@ -1,7 +1,6 @@
 import { describe, expect, it } from 'vitest';
-import { ZodError } from 'zod';
 
-import { parseToolCalls } from '../parseToolCalls';
+import { parseToolCalls } from './parseToolCalls';
 
 describe('parseToolCalls', () => {
   it('should create add new item', () => {

package/src/libs/agent-runtime/index.ts

@@ -8,6 +8,7 @@ export { LobeDeepSeekAI } from './deepseek';
 export * from './error';
 export { LobeGoogleAI } from './google';
 export { LobeGroq } from './groq';
+export * from './helpers';
 export { LobeMinimaxAI } from './minimax';
 export { LobeMistralAI } from './mistral';
 export { LobeMoonshotAI } from './moonshot';

package/src/libs/agent-runtime/types/chat.ts

@@ -1,4 +1,25 @@
-import { MessageToolCall } from '@/types/message';
+import { DeepPartial } from 'utility-types';
+
+import { ModelTokensUsage, ToolFunction } from '@/types/message';
+
+export interface MessageToolCall {
+  /**
+   * The function that the model called.
+   */
+  function: ToolFunction;
+
+  /**
+   * The ID of the tool call.
+   */
+  id: string;
+
+  /**
+   * The type of the tool. Currently, only `function` is supported.
+   */
+  type: 'function' | string;
+}
+
+export type MessageToolCallChunk = DeepPartial<MessageToolCall> & { index: number };
 
 export type LLMRoleType = 'user' | 'system' | 'assistant' | 'function' | 'tool';
 
@@ -165,18 +186,29 @@ export interface ChatCompletionTool {
   type: 'function';
 }
 
+interface OnFinishData {
+  grounding?: any;
+  text: string;
+  thinking?: string;
+  toolsCalling?: MessageToolCall[];
+  usage?: ModelTokensUsage;
+}
+
 export interface ChatStreamCallbacks {
+  onCompletion?: (data: OnFinishData) => Promise<void> | void;
   /**
-   * `onCompletion`: Called for each tokenized message.
+   * `onFinal`: Called once when the stream is closed with the final completion message.
    **/
-  onCompletion?: (completion: string) => Promise<void> | void;
-  /** `onFinal`: Called once when the stream is closed with the final completion message. */
-  onFinal?: (completion: string) => Promise<void> | void;
+  onFinal?: (data: OnFinishData) => Promise<void> | void;
+  onGrounding?: (grounding: any) => Promise<void> | void;
   /** `onStart`: Called once when the stream is initialized. */
   onStart?: () => Promise<void> | void;
   /** `onText`: Called for each text chunk. */
-  onText?: (text: string) => Promise<void> | void;
-  /** `onToken`: Called for each tokenized message. */
-  onToken?: (token: string) => Promise<void> | void;
-  onToolCall?: () => Promise<void> | void;
+  onText?: (content: string) => Promise<void> | void;
+  onThinking?: (content: string) => Promise<void> | void;
+  onToolsCalling?: (data: {
+    chunk: MessageToolCallChunk[];
+    toolsCalling: MessageToolCall[];
+  }) => Promise<void> | void;
+  onUsage?: (usage: ModelTokensUsage) => Promise<void> | void;
 }

package/src/libs/agent-runtime/utils/openaiCompatibleFactory/index.test.ts

@@ -776,7 +776,7 @@ describe('LobeOpenAICompatibleFactory', () => {
         // 准备 callback 和 headers
         const mockCallback: ChatStreamCallbacks = {
           onStart: vi.fn(),
-          onToken: vi.fn(),
+          onCompletion: vi.fn(),
         };
         const mockHeaders = { 'Custom-Header': 'TestValue' };
 
@@ -793,7 +793,9 @@ describe('LobeOpenAICompatibleFactory', () => {
         // 验证 callback 被调用
         await result.text(); // 确保流被消费
         expect(mockCallback.onStart).toHaveBeenCalled();
-        expect(mockCallback.onToken).toHaveBeenCalledWith('hello');
+        expect(mockCallback.onCompletion).toHaveBeenCalledWith({
+          text: 'hello',
+        });
 
         // 验证 headers 被正确传递
         expect(result.headers.get('Custom-Header')).toEqual('TestValue');

package/src/libs/agent-runtime/utils/streams/anthropic.test.ts

@@ -58,13 +58,11 @@ describe('AnthropicStream', () => {
 
     const onStartMock = vi.fn();
     const onTextMock = vi.fn();
-    const onTokenMock = vi.fn();
     const onCompletionMock = vi.fn();
 
     const protocolStream = AnthropicStream(mockAnthropicStream, {
       onStart: onStartMock,
       onText: onTextMock,
-      onToken: onTokenMock,
       onCompletion: onCompletionMock,
     });
 
@@ -92,9 +90,8 @@ describe('AnthropicStream', () => {
     ]);
 
     expect(onStartMock).toHaveBeenCalledTimes(1);
-    expect(onTextMock).toHaveBeenNthCalledWith(1, '"Hello"');
-    expect(onTextMock).toHaveBeenNthCalledWith(2, '" world!"');
-    expect(onTokenMock).toHaveBeenCalledTimes(2);
+    expect(onTextMock).toHaveBeenNthCalledWith(1, 'Hello');
+    expect(onTextMock).toHaveBeenNthCalledWith(2, ' world!');
     expect(onCompletionMock).toHaveBeenCalledTimes(1);
   });
 
@@ -168,7 +165,7 @@ describe('AnthropicStream', () => {
     const onToolCallMock = vi.fn();
 
     const protocolStream = AnthropicStream(mockReadableStream, {
-      onToolCall: onToolCallMock,
+      onToolsCalling: onToolCallMock,
     });
 
     const decoder = new TextDecoder();
@@ -320,7 +317,7 @@ describe('AnthropicStream', () => {
     const onToolCallMock = vi.fn();
 
     const protocolStream = AnthropicStream(mockReadableStream, {
-      onToolCall: onToolCallMock,
+      onToolsCalling: onToolCallMock,
     });
 
     const decoder = new TextDecoder();

package/src/libs/agent-runtime/utils/streams/bedrock/llama.test.ts

@@ -20,13 +20,11 @@ describe('AWSBedrockLlamaStream', () => {
 
     const onStartMock = vi.fn();
     const onTextMock = vi.fn();
-    const onTokenMock = vi.fn();
     const onCompletionMock = vi.fn();
 
     const protocolStream = AWSBedrockLlamaStream(mockBedrockStream, {
       onStart: onStartMock,
       onText: onTextMock,
-      onToken: onTokenMock,
       onCompletion: onCompletionMock,
     });
 
@@ -51,9 +49,8 @@ describe('AWSBedrockLlamaStream', () => {
     ]);
 
     expect(onStartMock).toHaveBeenCalledTimes(1);
-    expect(onTextMock).toHaveBeenNthCalledWith(1, '"Hello"');
-    expect(onTextMock).toHaveBeenNthCalledWith(2, '" world!"');
-    expect(onTokenMock).toHaveBeenCalledTimes(2);
+    expect(onTextMock).toHaveBeenNthCalledWith(1, 'Hello');
+    expect(onTextMock).toHaveBeenNthCalledWith(2, ' world!');
     expect(onCompletionMock).toHaveBeenCalledTimes(1);
   });
 
@@ -73,13 +70,11 @@ describe('AWSBedrockLlamaStream', () => {
 
     const onStartMock = vi.fn();
     const onTextMock = vi.fn();
-    const onTokenMock = vi.fn();
     const onCompletionMock = vi.fn();
 
     const protocolStream = AWSBedrockLlamaStream(mockBedrockStream, {
       onStart: onStartMock,
       onText: onTextMock,
-      onToken: onTokenMock,
       onCompletion: onCompletionMock,
     });
 
@@ -104,9 +99,8 @@ describe('AWSBedrockLlamaStream', () => {
     ]);
 
     expect(onStartMock).toHaveBeenCalledTimes(1);
-    expect(onTextMock).toHaveBeenNthCalledWith(1, '"Hello"');
-    expect(onTextMock).toHaveBeenNthCalledWith(2, '" world!"');
-    expect(onTokenMock).toHaveBeenCalledTimes(2);
+    expect(onTextMock).toHaveBeenNthCalledWith(1, 'Hello');
+    expect(onTextMock).toHaveBeenNthCalledWith(2, ' world!');
     expect(onCompletionMock).toHaveBeenCalledTimes(1);
   });
 
@@ -143,7 +137,6 @@ describe('AWSBedrockLlamaStream', () => {
     const protocolStream = AWSBedrockLlamaStream(mockBedrockStream, {
       onStart: onStartMock,
       onText: onTextMock,
-      onToken: onTokenMock,
       onCompletion: onCompletionMock,
     });
 
@@ -168,9 +161,8 @@ describe('AWSBedrockLlamaStream', () => {
     ]);
 
     expect(onStartMock).toHaveBeenCalledTimes(1);
-    expect(onTextMock).toHaveBeenNthCalledWith(1, '"Hello"');
-    expect(onTextMock).toHaveBeenNthCalledWith(2, '" world!"');
-    expect(onTokenMock).toHaveBeenCalledTimes(2);
+    expect(onTextMock).toHaveBeenNthCalledWith(1, 'Hello');
+    expect(onTextMock).toHaveBeenNthCalledWith(2, ' world!');
     expect(onCompletionMock).toHaveBeenCalledTimes(1);
   });
 

package/src/libs/agent-runtime/utils/streams/google-ai.test.ts

@@ -30,15 +30,13 @@ describe('GoogleGenerativeAIStream', () => {
 
     const onStartMock = vi.fn();
     const onTextMock = vi.fn();
-    const onTokenMock = vi.fn();
     const onToolCallMock = vi.fn();
     const onCompletionMock = vi.fn();
 
     const protocolStream = GoogleGenerativeAIStream(mockGoogleStream, {
       onStart: onStartMock,
       onText: onTextMock,
-      onToken: onTokenMock,
-      onToolCall: onToolCallMock,
+      onToolsCalling: onToolCallMock,
       onCompletion: onCompletionMock,
     });
 
@@ -68,9 +66,8 @@ describe('GoogleGenerativeAIStream', () => {
     ]);
 
     expect(onStartMock).toHaveBeenCalledTimes(1);
-    expect(onTextMock).toHaveBeenNthCalledWith(1, '"Hello"');
-    expect(onTextMock).toHaveBeenNthCalledWith(2, '" world!"');
-    expect(onTokenMock).toHaveBeenCalledTimes(2);
+    expect(onTextMock).toHaveBeenNthCalledWith(1, 'Hello');
+    expect(onTextMock).toHaveBeenNthCalledWith(2, ' world!');
     expect(onToolCallMock).toHaveBeenCalledTimes(1);
     expect(onCompletionMock).toHaveBeenCalledTimes(1);
   });

package/src/libs/agent-runtime/utils/streams/ollama.test.ts

@@ -22,13 +22,11 @@ describe('OllamaStream', () => {
 
       const onStartMock = vi.fn();
       const onTextMock = vi.fn();
-      const onTokenMock = vi.fn();
       const onCompletionMock = vi.fn();
 
       const protocolStream = OllamaStream(mockOllamaStream, {
         onStart: onStartMock,
         onText: onTextMock,
-        onToken: onTokenMock,
         onCompletion: onCompletionMock,
       });
 
@@ -53,9 +51,8 @@ describe('OllamaStream', () => {
       ]);
 
       expect(onStartMock).toHaveBeenCalledTimes(1);
-      expect(onTextMock).toHaveBeenNthCalledWith(1, '"Hello"');
-      expect(onTextMock).toHaveBeenNthCalledWith(2, '" world!"');
-      expect(onTokenMock).toHaveBeenCalledTimes(2);
+      expect(onTextMock).toHaveBeenNthCalledWith(1, 'Hello');
+      expect(onTextMock).toHaveBeenNthCalledWith(2, ' world!');
       expect(onCompletionMock).toHaveBeenCalledTimes(1);
     });
 
@@ -100,16 +97,14 @@ describe('OllamaStream', () => {
       });
       const onStartMock = vi.fn();
       const onTextMock = vi.fn();
-      const onTokenMock = vi.fn();
       const onToolCall = vi.fn();
       const onCompletionMock = vi.fn();
 
       const protocolStream = OllamaStream(mockOllamaStream, {
         onStart: onStartMock,
         onText: onTextMock,
-        onToken: onTokenMock,
         onCompletion: onCompletionMock,
-        onToolCall,
+        onToolsCalling: onToolCall,
       });
 
       const decoder = new TextDecoder();
@@ -134,7 +129,6 @@ describe('OllamaStream', () => {
       expect(onTextMock).toHaveBeenCalledTimes(0);
       expect(onStartMock).toHaveBeenCalledTimes(1);
       expect(onToolCall).toHaveBeenCalledTimes(1);
-      expect(onTokenMock).toHaveBeenCalledTimes(0);
       expect(onCompletionMock).toHaveBeenCalledTimes(1);
     });
   });

package/src/libs/agent-runtime/utils/streams/openai.test.ts

@@ -44,14 +44,12 @@ describe('OpenAIStream', () => {
 
     const onStartMock = vi.fn();
     const onTextMock = vi.fn();
-    const onTokenMock = vi.fn();
     const onCompletionMock = vi.fn();
 
     const protocolStream = OpenAIStream(mockOpenAIStream, {
       callbacks: {
         onStart: onStartMock,
         onText: onTextMock,
-        onToken: onTokenMock,
         onCompletion: onCompletionMock,
       },
     });
@@ -77,9 +75,8 @@ describe('OpenAIStream', () => {
     ]);
 
     expect(onStartMock).toHaveBeenCalledTimes(1);
-    expect(onTextMock).toHaveBeenNthCalledWith(1, '"Hello"');
-    expect(onTextMock).toHaveBeenNthCalledWith(2, '" world!"');
-    expect(onTokenMock).toHaveBeenCalledTimes(2);
+    expect(onTextMock).toHaveBeenNthCalledWith(1, 'Hello');
+    expect(onTextMock).toHaveBeenNthCalledWith(2, ' world!');
     expect(onCompletionMock).toHaveBeenCalledTimes(1);
   });
 
@@ -195,7 +192,7 @@ describe('OpenAIStream', () => {
 
     const protocolStream = OpenAIStream(mockOpenAIStream, {
       callbacks: {
-        onToolCall: onToolCallMock,
+        onToolsCalling: onToolCallMock,
       },
     });
 
@@ -578,7 +575,7 @@ describe('OpenAIStream', () => {
 
       const protocolStream = OpenAIStream(mockOpenAIStream, {
         callbacks: {
-          onToolCall: onToolCallMock,
+          onToolsCalling: onToolCallMock,
         },
       });
 
@@ -711,7 +708,7 @@ describe('OpenAIStream', () => {
 
       const protocolStream = OpenAIStream(mockOpenAIStream, {
         callbacks: {
-          onToolCall: onToolCallMock,
+          onToolsCalling: onToolCallMock,
         },
       });
 

package/src/libs/agent-runtime/utils/streams/protocol.ts

@@ -1,7 +1,8 @@
-import { ChatStreamCallbacks } from '@/libs/agent-runtime';
 import { ModelTokensUsage } from '@/types/message';
 
 import { AgentRuntimeErrorType } from '../../error';
+import { parseToolCalls } from '../../helpers';
+import { ChatStreamCallbacks } from '../../types';
 
 /**
  * context in the stream to save temporarily data
@@ -140,18 +141,31 @@ export const createSSEProtocolTransformer = (
 
 export function createCallbacksTransformer(cb: ChatStreamCallbacks | undefined) {
   const textEncoder = new TextEncoder();
-  let aggregatedResponse = '';
-  let currentType = '';
+  let aggregatedText = '';
+  let aggregatedThinking: string | undefined = undefined;
+  let usage: ModelTokensUsage | undefined;
+  let grounding: any;
+  let toolsCalling: any;
+
+  let currentType = '' as unknown as StreamProtocolChunk['type'];
   const callbacks = cb || {};
 
   return new TransformStream({
     async flush(): Promise<void> {
+      const data = {
+        grounding,
+        text: aggregatedText,
+        thinking: aggregatedThinking,
+        toolsCalling,
+        usage,
+      };
+
       if (callbacks.onCompletion) {
-        await callbacks.onCompletion(aggregatedResponse);
+        await callbacks.onCompletion(data);
       }
 
       if (callbacks.onFinal) {
-        await callbacks.onFinal(aggregatedResponse);
+        await callbacks.onFinal(data);
       }
     },
 
@@ -164,22 +178,53 @@ export function createCallbacksTransformer(cb: ChatStreamCallbacks | undefined)
 
       // track the type of the chunk
       if (chunk.startsWith('event:')) {
-        currentType = chunk.split('event:')[1].trim();
+        currentType = chunk.split('event:')[1].trim() as unknown as StreamProtocolChunk['type'];
       }
       // if the message is a data chunk, handle the callback
       else if (chunk.startsWith('data:')) {
         const content = chunk.split('data:')[1].trim();
 
+        let data: any = undefined;
+        try {
+          data = JSON.parse(content);
+        } catch {}
+
+        if (!data) return;
+
         switch (currentType) {
           case 'text': {
-            await callbacks.onText?.(content);
-            await callbacks.onToken?.(JSON.parse(content));
+            aggregatedText += data;
+            await callbacks.onText?.(data);
+            break;
+          }
+
+          case 'reasoning': {
+            if (!aggregatedThinking) {
+              aggregatedThinking = '';
+            }
+
+            aggregatedThinking += data;
+            await callbacks.onThinking?.(data);
+            break;
+          }
+
+          case 'usage': {
+            usage = data;
+            await callbacks.onUsage?.(data);
+            break;
+          }
+
+          case 'grounding': {
+            grounding = data;
+            await callbacks.onGrounding?.(data);
             break;
           }
 
           case 'tool_calls': {
-            // TODO: make on ToolCall callback
-            await callbacks.onToolCall?.();
+            if (!toolsCalling) toolsCalling = [];
+            toolsCalling = parseToolCalls(toolsCalling, data);
+
+            await callbacks.onToolsCalling?.({ chunk: data, toolsCalling });
           }
         }
       }

package/src/libs/agent-runtime/utils/streams/qwen.test.ts

@@ -43,13 +43,11 @@ describe('QwenAIStream', () => {
 
     const onStartMock = vi.fn();
     const onTextMock = vi.fn();
-    const onTokenMock = vi.fn();
     const onCompletionMock = vi.fn();
 
     const protocolStream = QwenAIStream(mockOpenAIStream, {
       onStart: onStartMock,
       onText: onTextMock,
-      onToken: onTokenMock,
       onCompletion: onCompletionMock,
     });
 
@@ -74,9 +72,8 @@ describe('QwenAIStream', () => {
     ]);
 
     expect(onStartMock).toHaveBeenCalledTimes(1);
-    expect(onTextMock).toHaveBeenNthCalledWith(1, '"Hello"');
-    expect(onTextMock).toHaveBeenNthCalledWith(2, '" world!"');
-    expect(onTokenMock).toHaveBeenCalledTimes(2);
+    expect(onTextMock).toHaveBeenNthCalledWith(1, 'Hello');
+    expect(onTextMock).toHaveBeenNthCalledWith(2, ' world!');
     expect(onCompletionMock).toHaveBeenCalledTimes(1);
   });
 
@@ -114,7 +111,7 @@ describe('QwenAIStream', () => {
     const onToolCallMock = vi.fn();
 
     const protocolStream = QwenAIStream(mockOpenAIStream, {
-      onToolCall: onToolCallMock,
+      onToolsCalling: onToolCallMock,
     });
 
     const decoder = new TextDecoder();

package/src/libs/agent-runtime/utils/streams/spark.test.ts

@@ -1,41 +1,42 @@
+import type OpenAI from 'openai';
 import { beforeAll, describe, expect, it, vi } from 'vitest';
+
 import { SparkAIStream, transformSparkResponseToStream } from './spark';
-import type OpenAI from 'openai';
 
 describe('SparkAIStream', () => {
   beforeAll(() => {});
 
   it('should transform non-streaming response to stream', async () => {
     const mockResponse = {
-      id: "cha000ceba6@dx193d200b580b8f3532",
-      object: "chat.completion",
+      id: 'cha000ceba6@dx193d200b580b8f3532',
+      object: 'chat.completion',
       created: 1734395014,
-      model: "max-32k",
+      model: 'max-32k',
       choices: [
         {
           message: {
-            role: "assistant",
-            content: "",
+            role: 'assistant',
+            content: '',
             refusal: null,
             tool_calls: {
-              type: "function",
+              type: 'function',
               function: {
                 arguments: '{"city":"Shanghai"}',
-                name: "realtime-weather____fetchCurrentWeather"
+                name: 'realtime-weather____fetchCurrentWeather',
               },
-              id: "call_1"
-            }
+              id: 'call_1',
+            },
           },
           index: 0,
           logprobs: null,
-          finish_reason: "tool_calls"
-        }
+          finish_reason: 'tool_calls',
+        },
       ],
       usage: {
         prompt_tokens: 8,
         completion_tokens: 0,
-        total_tokens: 8
-      }
+        total_tokens: 8,
+      },
     } as unknown as OpenAI.ChatCompletion;
 
     const stream = transformSparkResponseToStream(mockResponse);
@@ -48,15 +49,17 @@ describe('SparkAIStream', () => {
     }
 
     expect(chunks).toHaveLength(2);
-    expect(chunks[0].choices[0].delta.tool_calls).toEqual([{
-      function: {
-        arguments: '{"city":"Shanghai"}',
-        name: "realtime-weather____fetchCurrentWeather"
+    expect(chunks[0].choices[0].delta.tool_calls).toEqual([
+      {
+        function: {
+          arguments: '{"city":"Shanghai"}',
+          name: 'realtime-weather____fetchCurrentWeather',
+        },
+        id: 'call_1',
+        index: 0,
+        type: 'function',
       },
-      id: "call_1",
-      index: 0,
-      type: "function"
-    }]);
+    ]);
     expect(chunks[1].choices[0].finish_reason).toBeDefined();
   });
 
@@ -64,36 +67,36 @@ describe('SparkAIStream', () => {
     const mockStream = new ReadableStream({
       start(controller) {
         controller.enqueue({
-          id: "cha000b0bf9@dx193d1ffa61cb894532",
-          object: "chat.completion.chunk",
+          id: 'cha000b0bf9@dx193d1ffa61cb894532',
+          object: 'chat.completion.chunk',
           created: 1734395014,
-          model: "max-32k",
+          model: 'max-32k',
           choices: [
             {
               delta: {
-                role: "assistant",
-                content: "",
+                role: 'assistant',
+                content: '',
                 tool_calls: {
-                  type: "function",
+                  type: 'function',
                   function: {
                     arguments: '{"city":"Shanghai"}',
-                    name: "realtime-weather____fetchCurrentWeather"
+                    name: 'realtime-weather____fetchCurrentWeather',
                   },
-                  id: "call_1"
-                }
+                  id: 'call_1',
+                },
               },
-              index: 0
-            }
-          ]
+              index: 0,
+            },
+          ],
         } as unknown as OpenAI.ChatCompletionChunk);
         controller.close();
-      }
+      },
     });
 
     const onToolCallMock = vi.fn();
 
     const protocolStream = SparkAIStream(mockStream, {
-      onToolCall: onToolCallMock
+      onToolsCalling: onToolCallMock,
     });
 
     const decoder = new TextDecoder();
@@ -107,7 +110,7 @@ describe('SparkAIStream', () => {
     expect(chunks).toEqual([
       'id: cha000b0bf9@dx193d1ffa61cb894532\n',
       'event: tool_calls\n',
-      `data: [{\"function\":{\"arguments\":\"{\\\"city\\\":\\\"Shanghai\\\"}\",\"name\":\"realtime-weather____fetchCurrentWeather\"},\"id\":\"call_1\",\"index\":0,\"type\":\"function\"}]\n\n`
+      `data: [{\"function\":{\"arguments\":\"{\\\"city\\\":\\\"Shanghai\\\"}\",\"name\":\"realtime-weather____fetchCurrentWeather\"},\"id\":\"call_1\",\"index\":0,\"type\":\"function\"}]\n\n`,
     ]);
 
     expect(onToolCallMock).toHaveBeenCalledTimes(1);
@@ -117,43 +120,43 @@ describe('SparkAIStream', () => {
     const mockStream = new ReadableStream({
       start(controller) {
         controller.enqueue({
-          id: "test-id",
-          object: "chat.completion.chunk",
+          id: 'test-id',
+          object: 'chat.completion.chunk',
           created: 1734395014,
-          model: "max-32k",
+          model: 'max-32k',
           choices: [
             {
               delta: {
-                content: "Hello",
-                role: "assistant"
+                content: 'Hello',
+                role: 'assistant',
               },
-              index: 0
-            }
-          ]
+              index: 0,
+            },
+          ],
         } as OpenAI.ChatCompletionChunk);
         controller.enqueue({
-          id: "test-id",
-          object: "chat.completion.chunk",
+          id: 'test-id',
+          object: 'chat.completion.chunk',
           created: 1734395014,
-          model: "max-32k",
+          model: 'max-32k',
           choices: [
             {
               delta: {
-                content: " World",
-                role: "assistant"
+                content: ' World',
+                role: 'assistant',
               },
-              index: 0
-            }
-          ]
+              index: 0,
+            },
+          ],
         } as OpenAI.ChatCompletionChunk);
         controller.close();
-      }
+      },
     });
 
     const onTextMock = vi.fn();
-    
+
     const protocolStream = SparkAIStream(mockStream, {
-      onText: onTextMock
+      onText: onTextMock,
     });
 
     const decoder = new TextDecoder();
@@ -170,18 +173,18 @@ describe('SparkAIStream', () => {
       'data: "Hello"\n\n',
       'id: test-id\n',
       'event: text\n',
-      'data: " World"\n\n'
+      'data: " World"\n\n',
     ]);
 
-    expect(onTextMock).toHaveBeenNthCalledWith(1, '"Hello"');
-    expect(onTextMock).toHaveBeenNthCalledWith(2, '" World"');
+    expect(onTextMock).toHaveBeenNthCalledWith(1, 'Hello');
+    expect(onTextMock).toHaveBeenNthCalledWith(2, ' World');
   });
 
   it('should handle empty stream', async () => {
     const mockStream = new ReadableStream({
       start(controller) {
         controller.close();
-      }
+      },
     });
 
     const protocolStream = SparkAIStream(mockStream);

package/src/libs/agent-runtime/utils/streams/vertex-ai.test.ts

@@ -99,15 +99,13 @@ describe('VertexAIStream', () => {
 
     const onStartMock = vi.fn();
     const onTextMock = vi.fn();
-    const onTokenMock = vi.fn();
     const onToolCallMock = vi.fn();
     const onCompletionMock = vi.fn();
 
     const protocolStream = VertexAIStream(mockGoogleStream, {
       onStart: onStartMock,
       onText: onTextMock,
-      onToken: onTokenMock,
-      onToolCall: onToolCallMock,
+      onToolsCalling: onToolCallMock,
       onCompletion: onCompletionMock,
     });
 
@@ -132,7 +130,7 @@ describe('VertexAIStream', () => {
     ]);
 
     expect(onStartMock).toHaveBeenCalledTimes(1);
-    expect(onTokenMock).toHaveBeenCalledTimes(2);
+    expect(onTextMock).toHaveBeenCalledTimes(2);
     expect(onCompletionMock).toHaveBeenCalledTimes(1);
   });
 
@@ -202,15 +200,13 @@ describe('VertexAIStream', () => {
 
     const onStartMock = vi.fn();
     const onTextMock = vi.fn();
-    const onTokenMock = vi.fn();
     const onToolCallMock = vi.fn();
     const onCompletionMock = vi.fn();
 
     const protocolStream = VertexAIStream(mockGoogleStream, {
       onStart: onStartMock,
       onText: onTextMock,
-      onToken: onTokenMock,
-      onToolCall: onToolCallMock,
+      onToolsCalling: onToolCallMock,
       onCompletion: onCompletionMock,
     });
 

package/src/libs/agent-runtime/zhipu/index.test.ts

@@ -63,7 +63,7 @@ describe('LobeZhipuAI', () => {
       // 准备 callback 和 headers
       const mockCallback: ChatStreamCallbacks = {
         onStart: vi.fn(),
-        onToken: vi.fn(),
+        onText: vi.fn(),
       };
       const mockHeaders = { 'Custom-Header': 'TestValue' };
 
@@ -80,7 +80,7 @@ describe('LobeZhipuAI', () => {
       // 验证 callback 被调用
       await result.text(); // 确保流被消费
       expect(mockCallback.onStart).toHaveBeenCalled();
-      expect(mockCallback.onToken).toHaveBeenCalledWith('hello');
+      expect(mockCallback.onText).toHaveBeenCalledWith('hello');
 
       // 验证 headers 被正确传递
       expect(result.headers.get('Custom-Header')).toEqual('TestValue');

package/src/server/modules/AgentRuntime/index.ts

@@ -1,22 +1,10 @@
 import { getLLMConfig } from '@/config/llm';
 import { JWTPayload } from '@/const/auth';
-import { INBOX_SESSION_ID } from '@/const/session';
-import {
-  LOBE_CHAT_OBSERVATION_ID,
-  LOBE_CHAT_TRACE_ID,
-  TracePayload,
-  TraceTagMap,
-} from '@/const/trace';
-import { AgentRuntime, ChatStreamPayload, ModelProvider } from '@/libs/agent-runtime';
-import { TraceClient } from '@/libs/traces';
+import { AgentRuntime, ModelProvider } from '@/libs/agent-runtime';
 
 import apiKeyManager from './apiKeyManager';
 
-export interface AgentChatOptions {
-  enableTrace?: boolean;
-  provider: string;
-  trace?: TracePayload;
-}
+export * from './trace';
 
 /**
  * Retrieves the options object from environment and apikeymanager
@@ -26,7 +14,7 @@ export interface AgentChatOptions {
  * @param payload - The JWT payload.
  * @returns The options object.
  */
-const getLlmOptionsFromPayload = (provider: string, payload: JWTPayload) => {
+const getParamsFromPayload = (provider: string, payload: JWTPayload) => {
   const llmConfig = getLLMConfig() as Record<string, any>;
 
   switch (provider) {
@@ -131,66 +119,7 @@ export const initAgentRuntimeWithUserPayload = (
   params: any = {},
 ) => {
   return AgentRuntime.initializeWithProvider(provider, {
-    ...getLlmOptionsFromPayload(provider, payload),
+    ...getParamsFromPayload(provider, payload),
     ...params,
   });
 };
-
-export const createTraceOptions = (
-  payload: ChatStreamPayload,
-  { trace: tracePayload, provider }: AgentChatOptions,
-) => {
-  const { messages, model, tools, ...parameters } = payload;
-  // create a trace to monitor the completion
-  const traceClient = new TraceClient();
-  const trace = traceClient.createTrace({
-    id: tracePayload?.traceId,
-    input: messages,
-    metadata: { provider },
-    name: tracePayload?.traceName,
-    sessionId: `${tracePayload?.sessionId || INBOX_SESSION_ID}@${tracePayload?.topicId || 'start'}`,
-    tags: tracePayload?.tags,
-    userId: tracePayload?.userId,
-  });
-
-  const generation = trace?.generation({
-    input: messages,
-    metadata: { provider },
-    model,
-    modelParameters: parameters as any,
-    name: `Chat Completion (${provider})`,
-    startTime: new Date(),
-  });
-
-  return {
-    callback: {
-      experimental_onToolCall: async () => {
-        trace?.update({
-          tags: [...(tracePayload?.tags || []), TraceTagMap.ToolsCall],
-        });
-      },
-
-      onCompletion: async (completion: string) => {
-        generation?.update({
-          endTime: new Date(),
-          metadata: { provider, tools },
-          output: completion,
-        });
-
-        trace?.update({ output: completion });
-      },
-
-      onFinal: async () => {
-        await traceClient.shutdownAsync();
-      },
-
-      onStart: () => {
-        generation?.update({ completionStartTime: new Date() });
-      },
-    },
-    headers: {
-      [LOBE_CHAT_OBSERVATION_ID]: generation?.id,
-      [LOBE_CHAT_TRACE_ID]: trace?.id,
-    },
-  };
-};

package/src/server/modules/AgentRuntime/trace.ts

@@ -0,0 +1,107 @@
+import { after } from 'next/server';
+
+import { INBOX_SESSION_ID } from '@/const/session';
+import {
+  LOBE_CHAT_OBSERVATION_ID,
+  LOBE_CHAT_TRACE_ID,
+  TracePayload,
+  TraceTagMap,
+} from '@/const/trace';
+import { ChatStreamCallbacks, ChatStreamPayload } from '@/libs/agent-runtime';
+import { TraceClient } from '@/libs/traces';
+
+export interface AgentChatOptions {
+  enableTrace?: boolean;
+  provider: string;
+  trace?: TracePayload;
+}
+
+export const createTraceOptions = (
+  payload: ChatStreamPayload,
+  { trace: tracePayload, provider }: AgentChatOptions,
+) => {
+  const { messages, model, tools, ...parameters } = payload;
+  // create a trace to monitor the completion
+  const traceClient = new TraceClient();
+  const messageLength = messages.length;
+  const systemRole = messages.find((message) => message.role === 'system')?.content;
+
+  const trace = traceClient.createTrace({
+    id: tracePayload?.traceId,
+    input: messages,
+    metadata: { messageLength, model, provider, systemRole, tools },
+    name: tracePayload?.traceName,
+    sessionId: tracePayload?.topicId
+      ? tracePayload.topicId
+      : `${tracePayload?.sessionId || INBOX_SESSION_ID}@default`,
+    tags: tracePayload?.tags,
+    userId: tracePayload?.userId,
+  });
+
+  const generation = trace?.generation({
+    input: messages,
+    metadata: { messageLength, model, provider },
+    model,
+    modelParameters: parameters as any,
+    name: `Chat Completion (${provider})`,
+    startTime: new Date(),
+  });
+
+  return {
+    callback: {
+      onCompletion: async ({ text, thinking, usage, grounding, toolsCalling }) => {
+        const output =
+          // if the toolsCalling is not empty, we need to return the toolsCalling
+          !!toolsCalling && toolsCalling.length > 0
+            ? !!text
+              ? // tools calling with thinking and text
+                { text, thinking, toolsCalling }
+              : toolsCalling
+            : !!thinking
+              ? { text, thinking }
+              : text;
+
+        generation?.update({
+          endTime: new Date(),
+          metadata: { grounding, thinking },
+          output,
+          usage: usage
+            ? {
+                completionTokens: usage.outputTextTokens,
+                input: usage.totalInputTokens,
+                output: usage.totalOutputTokens,
+                promptTokens: usage.inputTextTokens,
+                totalTokens: usage.totalTokens,
+              }
+            : undefined,
+        });
+
+        trace?.update({ output });
+      },
+
+      onFinal: () => {
+        after(async () => {
+          try {
+            await traceClient.shutdownAsync();
+          } catch (e) {
+            console.error('TraceClient shutdown error:', e);
+          }
+        });
+      },
+
+      onStart: () => {
+        generation?.update({ completionStartTime: new Date() });
+      },
+
+      onToolsCalling: async () => {
+        trace?.update({
+          tags: [...(tracePayload?.tags || []), TraceTagMap.ToolsCalling],
+        });
+      },
+    } as ChatStreamCallbacks,
+    headers: {
+      [LOBE_CHAT_OBSERVATION_ID]: generation?.id,
+      [LOBE_CHAT_TRACE_ID]: trace?.id,
+    },
+  };
+};

package/src/store/chat/slices/aiChat/actions/generateAIChat.ts

@@ -388,6 +388,12 @@ export const generateAIChat: StateCreator<
             });
           }
         },
+        trace: {
+          traceId: params?.traceId,
+          sessionId: get().activeId,
+          topicId: get().activeTopicId,
+          traceName: TraceNameMap.SearchIntentRecognition,
+        },
         abortController,
         onMessageHandle: async (chunk) => {
           if (chunk.type === 'tool_calls') {

package/src/utils/fetch/fetchSSE.ts

@@ -2,6 +2,7 @@ import { isObject } from 'lodash-es';
 
 import { MESSAGE_CANCEL_FLAT } from '@/const/message';
 import { LOBE_CHAT_OBSERVATION_ID, LOBE_CHAT_TRACE_ID } from '@/const/trace';
+import { parseToolCalls } from '@/libs/agent-runtime';
 import { ChatErrorType } from '@/types/fetch';
 import { SmoothingParams } from '@/types/llm';
 import {
@@ -18,7 +19,6 @@ import { nanoid } from '@/utils/uuid';
 
 import { fetchEventSource } from './fetchEventSource';
 import { getMessageError } from './parseError';
-import { parseToolCalls } from './parseToolCalls';
 
 type SSEFinishType = 'done' | 'error' | 'abort';