@lobehub/chat 1.88.17 → 1.88.19

package/CHANGELOG.md

@@ -2,6 +2,57 @@
 
 # Changelog
 
+### [Version 1.88.19](https://github.com/lobehub/lobe-chat/compare/v1.88.18...v1.88.19)
+
+<sup>Released on **2025-05-30**</sup>
+
+#### 💄 Styles
+
+- **misc**: Support Web Search Tools and Beta Header from Anthropic.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### Styles
+
+- **misc**: Support Web Search Tools and Beta Header from Anthropic, closes [#7964](https://github.com/lobehub/lobe-chat/issues/7964) ([a47ddc5](https://github.com/lobehub/lobe-chat/commit/a47ddc5))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
+### [Version 1.88.18](https://github.com/lobehub/lobe-chat/compare/v1.88.17...v1.88.18)
+
+<sup>Released on **2025-05-29**</sup>
+
+#### 🐛 Bug Fixes
+
+- **misc**: Close historySummary correctly, Enable thinking output only for supported Gemini thinking models.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **misc**: Close historySummary correctly, closes [#7010](https://github.com/lobehub/lobe-chat/issues/7010) ([90a6f68](https://github.com/lobehub/lobe-chat/commit/90a6f68))
+- **misc**: Enable thinking output only for supported Gemini thinking models, closes [#7987](https://github.com/lobehub/lobe-chat/issues/7987) ([f503c53](https://github.com/lobehub/lobe-chat/commit/f503c53))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ### [Version 1.88.17](https://github.com/lobehub/lobe-chat/compare/v1.88.16...v1.88.17)
 
 <sup>Released on **2025-05-29**</sup>

package/changelog/v1.json

@@ -1,4 +1,22 @@
 [
+  {
+    "children": {
+      "improvements": [
+        "Support Web Search Tools and Beta Header from Anthropic."
+      ]
+    },
+    "date": "2025-05-30",
+    "version": "1.88.19"
+  },
+  {
+    "children": {
+      "fixes": [
+        "Close historySummary correctly, Enable thinking output only for supported Gemini thinking models."
+      ]
+    },
+    "date": "2025-05-29",
+    "version": "1.88.18"
+  },
   {
     "children": {
       "improvements": [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.88.17",
+  "version": "1.88.19",
   "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",

package/src/config/aiModels/anthropic.ts

@@ -5,6 +5,7 @@ const anthropicChatModels: AIChatModelCard[] = [
     abilities: {
       functionCall: true,
       reasoning: true,
+      search: true,
       vision: true,
     },
     contextWindowTokens: 200_000,
@@ -23,6 +24,7 @@ const anthropicChatModels: AIChatModelCard[] = [
     releasedAt: '2025-05-23',
     settings: {
       extendParams: ['disableContextCaching', 'enableReasoning', 'reasoningBudgetToken'],
+      searchImpl: 'params',
     },
     type: 'chat',
   },
@@ -30,6 +32,7 @@ const anthropicChatModels: AIChatModelCard[] = [
     abilities: {
       functionCall: true,
       reasoning: true,
+      search: true,
       vision: true,
     },
     contextWindowTokens: 200_000,
@@ -48,6 +51,7 @@ const anthropicChatModels: AIChatModelCard[] = [
     releasedAt: '2025-05-23',
     settings: {
       extendParams: ['disableContextCaching', 'enableReasoning', 'reasoningBudgetToken'],
+      searchImpl: 'params',
     },
     type: 'chat',
   },
@@ -55,6 +59,7 @@ const anthropicChatModels: AIChatModelCard[] = [
     abilities: {
       functionCall: true,
       reasoning: true,
+      search: true,
       vision: true,
     },
     contextWindowTokens: 200_000,
@@ -73,12 +78,14 @@ const anthropicChatModels: AIChatModelCard[] = [
     releasedAt: '2025-02-24',
     settings: {
       extendParams: ['disableContextCaching', 'enableReasoning', 'reasoningBudgetToken'],
+      searchImpl: 'params',
     },
     type: 'chat',
   },
   {
     abilities: {
       functionCall: true,
+      search: true,
       vision: true,
     },
     contextWindowTokens: 200_000,
@@ -96,6 +103,7 @@ const anthropicChatModels: AIChatModelCard[] = [
     releasedAt: '2024-10-22',
     settings: {
       extendParams: ['disableContextCaching'],
+      searchImpl: 'params',
     },
     type: 'chat',
   },
@@ -119,6 +127,7 @@ const anthropicChatModels: AIChatModelCard[] = [
     releasedAt: '2024-06-20',
     settings: {
       extendParams: ['disableContextCaching'],
+      searchImpl: 'params',
     },
     type: 'chat',
   },

package/src/features/Conversation/components/History/index.tsx

@@ -7,6 +7,8 @@ import { memo } from 'react';
 import { useTranslation } from 'react-i18next';
 import { Center, Flexbox } from 'react-layout-kit';
 
+import { agentChatConfigSelectors } from '@/store/agent/selectors';
+import { useAgentStore } from '@/store/agent/store';
 import { useChatStore } from '@/store/chat';
 import { topicSelectors } from '@/store/chat/selectors';
 
@@ -35,10 +37,14 @@ const History = memo(() => {
     return [history?.content, history?.model];
   });
 
+  const enableCompressHistory = useAgentStore(
+    (s) => agentChatConfigSelectors.currentChatConfig(s).enableCompressHistory,
+  );
+
   return (
     <Flexbox paddingInline={16} style={{ paddingBottom: 8 }}>
       <HistoryDivider enable />
-      {!!content && (
+      {enableCompressHistory && !!content && (
         <Flexbox className={styles.container} gap={8}>
           <Flexbox align={'flex-start'} gap={8} horizontal>
             <Center height={20} width={20}>

package/src/libs/model-runtime/anthropic/index.test.ts

@@ -302,6 +302,64 @@ describe('LobeAnthropicAI', () => {
           { enabledContextCaching: true },
         );
       });
+
+      it('should build payload with tools and web search enabled', async () => {
+        const tools: ChatCompletionTool[] = [
+          { function: { name: 'tool1', description: 'desc1' }, type: 'function' }
+        ];
+
+        const mockAnthropicTools = [{ name: 'tool1', description: 'desc1' }];
+
+        vi.spyOn(anthropicHelpers, 'buildAnthropicTools').mockReturnValue(mockAnthropicTools as any);
+
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Search and get info', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.5,
+          tools,
+          enabledSearch: true,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(anthropicHelpers.buildAnthropicTools).toHaveBeenCalledWith(tools, {
+          enabledContextCaching: true,
+        });
+
+        // Should include both the converted tools and web search tool
+        expect(result.tools).toEqual([
+          ...mockAnthropicTools,
+          {
+            name: 'web_search',
+            type: 'web_search_20250305',
+          },
+        ]);
+      });
+
+      it('should build payload with web search enabled but no other tools', async () => {
+        vi.spyOn(anthropicHelpers, 'buildAnthropicTools').mockReturnValue(undefined);
+
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Search for information', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.5,
+          enabledSearch: true,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(anthropicHelpers.buildAnthropicTools).toHaveBeenCalledWith(undefined, {
+          enabledContextCaching: true,
+        });
+
+        // Should only include web search tool
+        expect(result.tools).toEqual([
+          {
+            name: 'web_search',
+            type: 'web_search_20250305',
+          },
+        ]);
+      });
     });
 
     describe('Error', () => {

package/src/libs/model-runtime/anthropic/index.ts

@@ -23,6 +23,8 @@ export interface AnthropicModelCard {
   id: string;
 }
 
+type anthropicTools = Anthropic.Tool | Anthropic.WebSearchTool20250305;
+
 const modelsWithSmallContextWindow = new Set(['claude-3-opus-20240229', 'claude-3-haiku-20240307']);
 
 const DEFAULT_BASE_URL = 'https://api.anthropic.com';
@@ -45,7 +47,14 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
   constructor({ apiKey, baseURL = DEFAULT_BASE_URL, id, ...res }: AnthropicAIParams = {}) {
     if (!apiKey) throw AgentRuntimeError.createError(AgentRuntimeErrorType.InvalidProviderAPIKey);
 
-    this.client = new Anthropic({ apiKey, baseURL, ...res });
+    const betaHeaders = process.env.ANTHROPIC_BETA_HEADERS;
+
+    this.client = new Anthropic({
+      apiKey,
+      baseURL,
+      ...(betaHeaders ? { defaultHeaders: { "anthropic-beta": betaHeaders } } : {}),
+      ...res
+    });
     this.baseURL = this.client.baseURL;
     this.apiKey = apiKey;
     this.id = id || ModelProvider.Anthropic;
@@ -99,6 +108,7 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
       tools,
       thinking,
       enabledContextCaching = true,
+      enabledSearch,
     } = payload;
 
     const { default: anthropicModels } = await import('@/config/aiModels/anthropic');
@@ -127,7 +137,27 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
 
     const postMessages = await buildAnthropicMessages(user_messages, { enabledContextCaching });
 
-    const postTools = buildAnthropicTools(tools, { enabledContextCaching });
+    let postTools: anthropicTools[] | undefined = buildAnthropicTools(tools, { enabledContextCaching });
+
+    if (enabledSearch) {
+      // Limit the number of searches per request
+      const maxUses = process.env.ANTHROPIC_MAX_USES;
+
+      const webSearchTool: Anthropic.WebSearchTool20250305 = {
+        name: 'web_search',
+        type: 'web_search_20250305',
+        ...(maxUses && Number.isInteger(Number(maxUses)) && Number(maxUses) > 0 && { 
+          max_uses: Number(maxUses) 
+        }),
+      };
+
+      // 如果已有工具，则添加到现有工具列表中；否则创建新的工具列表
+      if (postTools && postTools.length > 0) {
+        postTools = [...postTools, webSearchTool];
+      } else {
+        postTools = [webSearchTool];
+      }
+    }
 
     if (!!thinking && thinking.type === 'enabled') {
       const maxTokens = getMaxTokens() || 32_000; // Claude Opus 4 has minimum maxOutput

package/src/libs/model-runtime/google/index.ts

@@ -117,8 +117,17 @@ export class LobeGoogleAI implements LobeRuntimeAI {
       const { model, thinking } = payload;
 
       const thinkingConfig: GoogleAIThinkingConfig = {
-        includeThoughts: true,
-        thinkingBudget: thinking?.type === 'enabled' ? Math.min(thinking.budget_tokens, 24_576) : 0,
+        includeThoughts:
+          (thinking?.type === 'enabled') || 
+          (!thinking && model && (model.includes('-2.5-') || model.includes('thinking'))) 
+            ? true
+            : undefined,
+        thinkingBudget:
+          thinking?.type === 'enabled'
+            ? Math.min(thinking.budget_tokens, 24_576)
+            : thinking?.type === 'disabled'
+              ? 0
+              : undefined,
       };
 
       const contents = await this.buildGoogleMessages(payload.messages);
@@ -132,8 +141,8 @@ export class LobeGoogleAI implements LobeRuntimeAI {
               // @ts-expect-error - Google SDK 0.24.0 doesn't have this property for now with
               response_modalities: modelsWithModalities.has(model) ? ['Text', 'Image'] : undefined,
               temperature: payload.temperature,
-              thinkingConfig,
               topP: payload.top_p,
+              ...(modelsDisableInstuction.has(model) || model.toLowerCase().includes('learnlm') ? {} : { thinkingConfig }),
             },
             model,
             // avoid wide sensitive words

package/src/libs/model-runtime/utils/streams/anthropic.ts

@@ -1,7 +1,7 @@
 import Anthropic from '@anthropic-ai/sdk';
 import type { Stream } from '@anthropic-ai/sdk/streaming';
 
-import { ModelTokensUsage } from '@/types/message';
+import { ModelTokensUsage, CitationItem } from '@/types/message';
 
 import { ChatStreamCallbacks } from '../../types';
 import {
@@ -23,6 +23,7 @@ export const transformAnthropicStream = (
   switch (chunk.type) {
     case 'message_start': {
       context.id = chunk.message.id;
+      context.returnedCitationArray = [];
       let totalInputTokens = chunk.message.usage?.input_tokens;
 
       if (
@@ -59,6 +60,7 @@ export const transformAnthropicStream = (
           return { data: chunk.content_block.text, id: context.id, type: 'data' };
         }
 
+        case 'server_tool_use':
         case 'tool_use': {
           const toolChunk = chunk.content_block;
 
@@ -85,6 +87,29 @@ export const transformAnthropicStream = (
 
           return { data: [toolCall], id: context.id, type: 'tool_calls' };
         }
+
+        /*
+        case 'web_search_tool_result': {
+          const citations = chunk.content_block.content;
+
+          return [
+            {
+              data: {
+                citations: (citations as any[]).map(
+                  (item) =>
+                    ({
+                      title: item.title,
+                      url: item.url,
+                    }) as CitationItem,
+                ),
+              },
+              id: context.id,
+              type: 'grounding',
+            },
+          ];
+        }
+        */
+
         case 'thinking': {
           const thinkingChunk = chunk.content_block;
 
@@ -148,6 +173,19 @@ export const transformAnthropicStream = (
           };
         }
 
+        case 'citations_delta': {
+          const citations = (chunk as any).delta.citation;
+
+          if (context.returnedCitationArray) {
+            context.returnedCitationArray.push({
+              title: citations.title,
+              url: citations.url,
+            } as CitationItem)
+          }
+
+          return { data: null, id: context.id, type: 'text' };
+        }
+
         default: {
           break;
         }
@@ -180,7 +218,17 @@ export const transformAnthropicStream = (
     }
 
     case 'message_stop': {
-      return { data: 'message_stop', id: context.id, type: 'stop' };
+      return [
+        ...(context.returnedCitationArray?.length 
+          ? [{ 
+              data: { citations: context.returnedCitationArray }, 
+              id: context.id, 
+              type: 'grounding' 
+            }] 
+          : []
+        ),
+        { data: 'message_stop', id: context.id, type: 'stop' }
+      ] as any;
     }
 
     default: {

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

@@ -1,4 +1,4 @@
-import { ModelSpeed, ModelTokensUsage } from '@/types/message';
+import { CitationItem, ModelSpeed, ModelTokensUsage } from '@/types/message';
 import { safeParseJSON } from '@/utils/safeParseJSON';
 
 import { AgentRuntimeErrorType } from '../../error';
@@ -16,6 +16,13 @@ export interface StreamContext {
    * Same as Hunyuan and Wenxin
    */
   returnedCitation?: boolean;
+  /**
+   * Claude's citations are inline and interleaved with text output.
+   * Each text segment may carry references to sources (e.g., web search results) 
+   * relevant to that specific portion of the generated content.
+   * This array accumulates all citation items received during the streaming response.
+   */
+  returnedCitationArray?: CitationItem[];
   thinking?: {
     id: string;
     name: string;

package/src/services/chat.ts

@@ -225,12 +225,20 @@ class ChatService {
       )(getAiInfraStoreState());
       // if model has extended params, then we need to check if the model can use reasoning
 
-      if (modelExtendParams!.includes('enableReasoning') && chatConfig.enableReasoning) {
-        extendParams.thinking = {
-          budget_tokens: chatConfig.reasoningBudgetToken || 1024,
-          type: 'enabled',
-        };
+      if (modelExtendParams!.includes('enableReasoning')) {
+        if (chatConfig.enableReasoning) {
+          extendParams.thinking = {
+            budget_tokens: chatConfig.reasoningBudgetToken || 1024,
+            type: 'enabled',
+          };
+        } else {
+          extendParams.thinking = {
+            budget_tokens: 0,
+            type: 'disabled',
+          };
+        }
       }
+      
       if (
         modelExtendParams!.includes('disableContextCaching') &&
         chatConfig.disableContextCaching

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

@@ -552,7 +552,9 @@ export const generateAIChat: StateCreator<
     // to upload image
     const uploadTasks: Map<string, Promise<{ id?: string; url?: string }>> = new Map();
 
-    const historySummary = topicSelectors.currentActiveTopicSummary(get());
+    const historySummary = chatConfig.enableCompressHistory
+      ? topicSelectors.currentActiveTopicSummary(get())
+      : undefined;
     await chatService.createAssistantMessageStream({
       abortController,
       params: {