@lobehub/chat 1.68.11 → 1.69.1

package/src/features/ChatInput/ActionBar/Model/ReasoningTokenSlider.tsx

@@ -6,7 +6,6 @@ import useMergeState from 'use-merge-value';
 const Kibi = 1024;
 
 const exponent = (num: number) => Math.log2(num);
-const getRealValue = (num: number) => Math.round(Math.pow(2, num));
 const powerKibi = (num: number) => Math.round(Math.pow(2, num) * Kibi);
 
 interface MaxTokenSliderProps {
@@ -15,7 +14,7 @@ interface MaxTokenSliderProps {
   value?: number;
 }
 
-const MaxTokenSlider = memo<MaxTokenSliderProps>(({ value, onChange, defaultValue }) => {
+const ReasoningTokenSlider = memo<MaxTokenSliderProps>(({ value, onChange, defaultValue }) => {
   const [token, setTokens] = useMergeState(0, {
     defaultValue,
     onChange,
@@ -30,7 +29,7 @@ const MaxTokenSlider = memo<MaxTokenSliderProps>(({ value, onChange, defaultValu
   const updateWithPowValue = (value: number) => {
     setPowValue(value);
 
-    setTokens(powerKibi(value));
+    setTokens(Math.min(powerKibi(value), 64_000));
   };
 
   const updateWithRealValue = (value: number) => {
@@ -52,7 +51,7 @@ const MaxTokenSlider = memo<MaxTokenSliderProps>(({ value, onChange, defaultValu
   }, []);
 
   return (
-    <Flexbox align={'center'} gap={12} horizontal>
+    <Flexbox align={'center'} gap={12} horizontal paddingInline={'4px 0'}>
       <Flexbox flex={1}>
         <Slider
           marks={marks}
@@ -60,21 +59,14 @@ const MaxTokenSlider = memo<MaxTokenSliderProps>(({ value, onChange, defaultValu
           min={exponent(1)}
           onChange={updateWithPowValue}
           step={null}
-          tooltip={{
-            formatter: (x) => {
-              if (typeof x === 'undefined') return;
-
-              let value = getRealValue(x);
-
-              if (value < Kibi) return ((value * Kibi) / 1000).toFixed(0) + 'k';
-            },
-          }}
+          tooltip={{ open: false }}
           value={powValue}
         />
       </Flexbox>
       <div>
         <InputNumber
           changeOnWheel
+          max={64_000}
           min={0}
           onChange={(e) => {
             if (!e && e !== 0) return;
@@ -89,4 +81,4 @@ const MaxTokenSlider = memo<MaxTokenSliderProps>(({ value, onChange, defaultValu
     </Flexbox>
   );
 });
-export default MaxTokenSlider;
+export default ReasoningTokenSlider;

package/src/features/ChatInput/ActionBar/Search/ModelBuiltinSearch.tsx

@@ -7,7 +7,7 @@ import { useTranslation } from 'react-i18next';
 import { Flexbox } from 'react-layout-kit';
 
 import { useAgentStore } from '@/store/agent';
-import { agentSelectors } from '@/store/agent/selectors';
+import { agentChatConfigSelectors, agentSelectors } from '@/store/agent/selectors';
 import { aiModelSelectors, useAiInfraStore } from '@/store/aiInfra';
 
 import ExaIcon from './ExaIcon';
@@ -37,7 +37,7 @@ const ModelBuiltinSearch = memo(() => {
   const [model, provider, checked, updateAgentChatConfig] = useAgentStore((s) => [
     agentSelectors.currentAgentModel(s),
     agentSelectors.currentAgentModelProvider(s),
-    agentSelectors.currentAgentChatConfig(s).useModelBuiltinSearch,
+    agentChatConfigSelectors.useModelBuiltinSearch(s),
     s.updateAgentChatConfig,
   ]);
 

package/src/features/ChatInput/ActionBar/Search/SwitchPanel.tsx

@@ -8,7 +8,7 @@ import { useTranslation } from 'react-i18next';
 import { Flexbox } from 'react-layout-kit';
 
 import { useAgentStore } from '@/store/agent';
-import { agentSelectors } from '@/store/agent/slices/chat';
+import { agentChatConfigSelectors, agentSelectors } from '@/store/agent/slices/chat';
 import { aiModelSelectors, useAiInfraStore } from '@/store/aiInfra';
 import { SearchMode } from '@/types/search';
 
@@ -84,7 +84,7 @@ const Item = memo<NetworkOption>(({ value, description, icon, label, disable })
   const { t } = useTranslation('chat');
   const { styles } = useStyles();
   const [mode, updateAgentChatConfig] = useAgentStore((s) => [
-    agentSelectors.agentSearchMode(s),
+    agentChatConfigSelectors.agentSearchMode(s),
     s.updateAgentChatConfig,
   ]);
 

package/src/features/ChatInput/ActionBar/Token/TokenTag.tsx

@@ -10,7 +10,7 @@ import { useModelContextWindowTokens } from '@/hooks/useModelContextWindowTokens
 import { useModelSupportToolUse } from '@/hooks/useModelSupportToolUse';
 import { useTokenCount } from '@/hooks/useTokenCount';
 import { useAgentStore } from '@/store/agent';
-import { agentSelectors } from '@/store/agent/selectors';
+import { agentChatConfigSelectors, agentSelectors } from '@/store/agent/selectors';
 import { useChatStore } from '@/store/chat';
 import { topicSelectors } from '@/store/chat/selectors';
 import { useToolStore } from '@/store/tool';
@@ -31,15 +31,13 @@ const Token = memo<TokenTagProps>(({ total: messageString }) => {
   ]);
 
   const [systemRole, model, provider] = useAgentStore((s) => {
-    const config = agentSelectors.currentAgentChatConfig(s);
-
     return [
       agentSelectors.currentAgentSystemRole(s),
       agentSelectors.currentAgentModel(s) as string,
       agentSelectors.currentAgentModelProvider(s) as string,
       // add these two params to enable the component to re-render
-      config.historyCount,
-      config.enableHistoryCount,
+      agentChatConfigSelectors.historyCount(s),
+      agentChatConfigSelectors.enableHistoryCount(s),
     ];
   });
 

package/src/features/Conversation/Messages/Assistant/Tool/Render/CustomRender.tsx

@@ -57,4 +57,6 @@ const CustomRender = memo<CustomRenderProps>(
   },
 );
 
+CustomRender.displayName = 'CustomRender';
+
 export default CustomRender;

package/src/features/Conversation/Messages/Assistant/Tool/Render/index.tsx

@@ -1,3 +1,4 @@
+import isEqual from 'fast-deep-equal';
 import { Suspense, memo } from 'react';
 
 import { LOADING_FLAT } from '@/const/message';
@@ -16,10 +17,11 @@ interface RenderProps {
   toolCallId: string;
   toolIndex: number;
 }
+
 const Render = memo<RenderProps>(
   ({ toolCallId, toolIndex, messageId, requestArgs, showPluginRender, setShowPluginRender }) => {
     const loading = useChatStore(chatSelectors.isToolCallStreaming(messageId, toolIndex));
-    const toolMessage = useChatStore(chatSelectors.getMessageByToolCallId(toolCallId));
+    const toolMessage = useChatStore(chatSelectors.getMessageByToolCallId(toolCallId), isEqual);
 
     // 如果处于 loading 或者找不到 toolMessage 则展示 Arguments
     if (loading || !toolMessage) return <Arguments arguments={requestArgs} />;
@@ -48,4 +50,6 @@ const Render = memo<RenderProps>(
   },
 );
 
+Render.displayName = 'ToolRender';
+
 export default Render;

package/src/features/Conversation/Messages/Assistant/Tool/index.tsx

@@ -52,4 +52,6 @@ const Tool = memo<InspectorProps>(
   },
 );
 
+Tool.displayName = 'AssistantTool';
+
 export default Tool;

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

@@ -8,7 +8,7 @@ import { useTranslation } from 'react-i18next';
 import { Flexbox } from 'react-layout-kit';
 
 import { useAgentStore } from '@/store/agent';
-import { agentSelectors } from '@/store/agent/selectors';
+import { agentChatConfigSelectors } from '@/store/agent/selectors';
 import { useChatStore } from '@/store/chat';
 import { chatSelectors } from '@/store/chat/selectors';
 import { useUserStore } from '@/store/user';
@@ -65,15 +65,12 @@ const Item = memo<ChatListItemProps>(
     disableEditing,
     inPortalThread = false,
   }) => {
-    const fontSize = useUserStore(userGeneralSettingsSelectors.fontSize);
     const { t } = useTranslation('common');
     const { styles, cx } = useStyles();
-    const [type = 'chat'] = useAgentStore((s) => {
-      const config = agentSelectors.currentAgentChatConfig(s);
-      return [config.displayMode];
-    });
 
+    const type = useAgentStore(agentChatConfigSelectors.displayMode);
     const item = useChatStore(chatSelectors.getMessageById(id), isEqual);
+    const fontSize = useUserStore(userGeneralSettingsSelectors.fontSize);
 
     const [
       isMessageLoading,

package/src/features/Portal/Thread/Chat/ChatItem.tsx

@@ -3,7 +3,7 @@ import React, { memo, useMemo } from 'react';
 import { ChatItem } from '@/features/Conversation';
 import ActionsBar from '@/features/Conversation/components/ChatItem/ActionsBar';
 import { useAgentStore } from '@/store/agent';
-import { agentSelectors } from '@/store/agent/selectors';
+import { agentChatConfigSelectors } from '@/store/agent/selectors';
 import { useChatStore } from '@/store/chat';
 import { threadSelectors } from '@/store/chat/selectors';
 
@@ -35,14 +35,9 @@ const ThreadChatItem = memo<ThreadChatItemProps>(({ id, index }) => {
     [id, isParentMessage],
   );
 
-  const enableHistoryDivider = useAgentStore((s) => {
-    const config = agentSelectors.currentAgentChatConfig(s);
-    return (
-      config.enableHistoryCount &&
-      historyLength > (config.historyCount ?? 0) &&
-      config.historyCount === historyLength - index
-    );
-  });
+  const enableHistoryDivider = useAgentStore(
+    agentChatConfigSelectors.enableHistoryDivider(historyLength, index),
+  );
 
   return (
     <ChatItem

package/src/hooks/useAgentEnableSearch.ts

@@ -1,12 +1,12 @@
 import { useAgentStore } from '@/store/agent';
-import { agentSelectors } from '@/store/agent/slices/chat';
+import { agentChatConfigSelectors, agentSelectors } from '@/store/agent/selectors';
 import { aiModelSelectors, useAiInfraStore } from '@/store/aiInfra';
 
 export const useAgentEnableSearch = () => {
   const [model, provider, agentSearchMode] = useAgentStore((s) => [
     agentSelectors.currentAgentModel(s),
     agentSelectors.currentAgentModelProvider(s),
-    agentSelectors.agentSearchMode(s),
+    agentChatConfigSelectors.agentSearchMode(s),
   ]);
 
   const isModelSupportToolUse = useAiInfraStore(

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

@@ -81,7 +81,12 @@ describe('LobeAnthropicAI', () => {
       expect(instance['client'].messages.create).toHaveBeenCalledWith(
         {
           max_tokens: 4096,
-          messages: [{ content: 'Hello', role: 'user' }],
+          messages: [
+            {
+              content: [{ cache_control: { type: 'ephemeral' }, text: 'Hello', type: 'text' }],
+              role: 'user',
+            },
+          ],
           model: 'claude-3-haiku-20240307',
           stream: true,
           temperature: 0,
@@ -117,10 +122,21 @@ describe('LobeAnthropicAI', () => {
       expect(instance['client'].messages.create).toHaveBeenCalledWith(
         {
           max_tokens: 4096,
-          messages: [{ content: 'Hello', role: 'user' }],
+          messages: [
+            {
+              content: [{ cache_control: { type: 'ephemeral' }, text: 'Hello', type: 'text' }],
+              role: 'user',
+            },
+          ],
           model: 'claude-3-haiku-20240307',
           stream: true,
-          system: 'You are an awesome greeter',
+          system: [
+            {
+              cache_control: { type: 'ephemeral' },
+              type: 'text',
+              text: 'You are an awesome greeter',
+            },
+          ],
           temperature: 0,
         },
         {},
@@ -152,7 +168,12 @@ describe('LobeAnthropicAI', () => {
       expect(instance['client'].messages.create).toHaveBeenCalledWith(
         {
           max_tokens: 2048,
-          messages: [{ content: 'Hello', role: 'user' }],
+          messages: [
+            {
+              content: [{ cache_control: { type: 'ephemeral' }, text: 'Hello', type: 'text' }],
+              role: 'user',
+            },
+          ],
           model: 'claude-3-haiku-20240307',
           stream: true,
           temperature: 0.25,
@@ -189,7 +210,12 @@ describe('LobeAnthropicAI', () => {
       expect(instance['client'].messages.create).toHaveBeenCalledWith(
         {
           max_tokens: 2048,
-          messages: [{ content: 'Hello', role: 'user' }],
+          messages: [
+            {
+              content: [{ cache_control: { type: 'ephemeral' }, text: 'Hello', type: 'text' }],
+              role: 'user',
+            },
+          ],
           model: 'claude-3-haiku-20240307',
           stream: true,
           temperature: 0.25,
@@ -240,7 +266,7 @@ describe('LobeAnthropicAI', () => {
     });
 
     describe('chat with tools', () => {
-      it('should call client.beta.tools.messages.create when tools are provided', async () => {
+      it('should call tools when tools are provided', async () => {
         // Arrange
         const tools: ChatCompletionTool[] = [
           { function: { name: 'tool1', description: 'desc1' }, type: 'function' },
@@ -257,7 +283,10 @@ describe('LobeAnthropicAI', () => {
 
         // Assert
         expect(instance['client'].messages.create).toHaveBeenCalled();
-        expect(spyOn).toHaveBeenCalledWith(tools);
+        expect(spyOn).toHaveBeenCalledWith(
+          [{ function: { name: 'tool1', description: 'desc1' }, type: 'function' }],
+          { enabledContextCaching: true },
+        );
       });
     });
 

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

@@ -97,10 +97,33 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
   }
 
   private async buildAnthropicPayload(payload: ChatStreamPayload) {
-    const { messages, model, max_tokens, temperature, top_p, tools, thinking } = payload;
+    const {
+      messages,
+      model,
+      max_tokens,
+      temperature,
+      top_p,
+      tools,
+      thinking,
+      enabledContextCaching = true,
+    } = payload;
     const system_message = messages.find((m) => m.role === 'system');
     const user_messages = messages.filter((m) => m.role !== 'system');
 
+    const systemPrompts = !!system_message?.content
+      ? ([
+          {
+            cache_control: enabledContextCaching ? { type: 'ephemeral' } : undefined,
+            text: system_message?.content as string,
+            type: 'text',
+          },
+        ] as Anthropic.TextBlockParam[])
+      : undefined;
+
+    const postMessages = await buildAnthropicMessages(user_messages, { enabledContextCaching });
+
+    const postTools = buildAnthropicTools(tools, { enabledContextCaching });
+
     if (!!thinking) {
       const maxTokens =
         max_tokens ?? (thinking?.budget_tokens ? thinking?.budget_tokens + 4096 : 4096);
@@ -109,22 +132,21 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
       // `top_p` must be unset when thinking is enabled.
       return {
         max_tokens: maxTokens,
-        messages: await buildAnthropicMessages(user_messages),
+        messages: postMessages,
         model,
-        system: system_message?.content as string,
-
+        system: systemPrompts,
         thinking,
-        tools: buildAnthropicTools(tools),
+        tools: postTools,
       } satisfies Anthropic.MessageCreateParams;
     }
 
     return {
       max_tokens: max_tokens ?? 4096,
-      messages: await buildAnthropicMessages(user_messages),
+      messages: postMessages,
       model,
-      system: system_message?.content as string,
+      system: systemPrompts,
       temperature: payload.temperature !== undefined ? temperature / 2 : undefined,
-      tools: buildAnthropicTools(tools),
+      tools: postTools,
       top_p,
     } satisfies Anthropic.MessageCreateParams;
   }

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

@@ -1,6 +1,8 @@
 import OpenAI, { AzureOpenAI } from 'openai';
 import type { Stream } from 'openai/streaming';
 
+import { systemToUserModels } from '@/const/models';
+
 import { LobeRuntimeAI } from '../BaseAI';
 import { AgentRuntimeErrorType } from '../error';
 import { ChatCompetitionOptions, ChatStreamPayload, ModelProvider } from '../types';
@@ -13,7 +15,7 @@ import { OpenAIStream } from '../utils/streams';
 export class LobeAzureOpenAI implements LobeRuntimeAI {
   client: AzureOpenAI;
 
-  constructor(params: { apiKey?: string; apiVersion?: string, baseURL?: string; } = {}) {
+  constructor(params: { apiKey?: string; apiVersion?: string; baseURL?: string } = {}) {
     if (!params.apiKey || !params.baseURL)
       throw AgentRuntimeError.createError(AgentRuntimeErrorType.InvalidProviderAPIKey);
 
@@ -34,17 +36,10 @@ export class LobeAzureOpenAI implements LobeRuntimeAI {
     // o1 series models on Azure OpenAI does not support streaming currently
     const enableStreaming = model.includes('o1') ? false : (params.stream ?? true);
 
-    // Convert 'system' role to 'user' or 'developer' based on the model
-    const systemToUserModels = new Set([
-      'o1-preview',
-      'o1-preview-2024-09-12',
-      'o1-mini',
-      'o1-mini-2024-09-12',
-    ]);
-
     const updatedMessages = messages.map((message) => ({
       ...message,
       role:
+        // Convert 'system' role to 'user' or 'developer' based on the model
         (model.includes('o1') || model.includes('o3')) && message.role === 'system'
           ? [...systemToUserModels].some((sub) => model.includes(sub))
             ? 'user'

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

@@ -2,6 +2,8 @@ import createClient, { ModelClient } from '@azure-rest/ai-inference';
 import { AzureKeyCredential } from '@azure/core-auth';
 import OpenAI from 'openai';
 
+import { systemToUserModels } from '@/const/models';
+
 import { LobeRuntimeAI } from '../BaseAI';
 import { AgentRuntimeErrorType } from '../error';
 import { ChatCompetitionOptions, ChatStreamPayload, ModelProvider } from '../types';
@@ -30,24 +32,17 @@ export class LobeAzureAI implements LobeRuntimeAI {
     // o1 series models on Azure OpenAI does not support streaming currently
     const enableStreaming = model.includes('o1') ? false : (params.stream ?? true);
 
-    // Convert 'system' role to 'user' or 'developer' based on the model
-    const systemToUserModels = new Set([
-      'o1-preview',
-      'o1-preview-2024-09-12',
-      'o1-mini',
-      'o1-mini-2024-09-12',
-    ]);
-
     const updatedMessages = messages.map((message) => ({
       ...message,
       role:
+        // Convert 'system' role to 'user' or 'developer' based on the model
         (model.includes('o1') || model.includes('o3')) && message.role === 'system'
           ? [...systemToUserModels].some((sub) => model.includes(sub))
             ? 'user'
             : 'developer'
           : message.role,
     }));
-    
+
     try {
       const response = this.client.path('/chat/completions').post({
         body: {

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

@@ -1,25 +1,14 @@
+import { disableStreamModels, systemToUserModels } from '@/const/models';
+import type { ChatModelCard } from '@/types/llm';
+
 import { ChatStreamPayload, ModelProvider, OpenAIChatMessage } from '../types';
 import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory';
 
-import type { ChatModelCard } from '@/types/llm';
-
 export interface OpenAIModelCard {
   id: string;
 }
 
 export const pruneReasoningPayload = (payload: ChatStreamPayload) => {
-  // TODO: 临时写法，后续要重构成 model card 展示配置
-  const disableStreamModels = new Set([
-    'o1',
-    'o1-2024-12-17'
-  ]);
-  const systemToUserModels = new Set([
-    'o1-preview',
-    'o1-preview-2024-09-12',
-    'o1-mini',
-    'o1-mini-2024-09-12',
-  ]);
-
   return {
     ...payload,
     frequency_penalty: 0,
@@ -58,46 +47,40 @@ export const LobeOpenAI = LobeOpenAICompatibleFactory({
   models: async ({ client }) => {
     const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
 
-    const functionCallKeywords = [
-      'gpt-4',
-      'gpt-3.5',
-      'o3-mini',
-    ];
+    const functionCallKeywords = ['gpt-4', 'gpt-3.5', 'o3-mini'];
 
-    const visionKeywords = [
-      'gpt-4o',
-      'vision',
-    ];
+    const visionKeywords = ['gpt-4o', 'vision'];
 
-    const reasoningKeywords = [
-      'o1',
-      'o3',
-    ];
+    const reasoningKeywords = ['o1', 'o3'];
 
-    const modelsPage = await client.models.list() as any;
+    const modelsPage = (await client.models.list()) as any;
     const modelList: OpenAIModelCard[] = modelsPage.data;
 
     return modelList
       .map((model) => {
-        const knownModel = LOBE_DEFAULT_MODEL_LIST.find((m) => model.id.toLowerCase() === m.id.toLowerCase());
+        const knownModel = LOBE_DEFAULT_MODEL_LIST.find(
+          (m) => model.id.toLowerCase() === m.id.toLowerCase(),
+        );
 
         return {
           contextWindowTokens: knownModel?.contextWindowTokens ?? undefined,
           displayName: knownModel?.displayName ?? undefined,
           enabled: knownModel?.enabled || false,
           functionCall:
-            functionCallKeywords.some(keyword => model.id.toLowerCase().includes(keyword)) && !model.id.toLowerCase().includes('audio')
-            || knownModel?.abilities?.functionCall
-            || false,
+            (functionCallKeywords.some((keyword) => model.id.toLowerCase().includes(keyword)) &&
+              !model.id.toLowerCase().includes('audio')) ||
+            knownModel?.abilities?.functionCall ||
+            false,
           id: model.id,
           reasoning:
-            reasoningKeywords.some(keyword => model.id.toLowerCase().includes(keyword))
-            || knownModel?.abilities?.reasoning
-            || false,
+            reasoningKeywords.some((keyword) => model.id.toLowerCase().includes(keyword)) ||
+            knownModel?.abilities?.reasoning ||
+            false,
           vision:
-            visionKeywords.some(keyword => model.id.toLowerCase().includes(keyword)) && !model.id.toLowerCase().includes('audio')
-            || knownModel?.abilities?.vision
-            || false,
+            (visionKeywords.some((keyword) => model.id.toLowerCase().includes(keyword)) &&
+              !model.id.toLowerCase().includes('audio')) ||
+            knownModel?.abilities?.vision ||
+            false,
         };
       })
       .filter(Boolean) as ChatModelCard[];

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

@@ -46,6 +46,10 @@ export interface OpenAIChatMessage {
  * @title Chat Stream Payload
  */
 export interface ChatStreamPayload {
+  /**
+   * 开启上下文缓存
+   */
+  enabledContextCaching?: boolean;
   /**
    * 是否开启搜索
    */

package/src/libs/agent-runtime/utils/anthropicHelpers.test.ts

@@ -619,6 +619,26 @@ describe('anthropicHelpers', () => {
         { content: '继续', role: 'user' },
       ]);
     });
+
+    it('should enable cache control', async () => {
+      const messages: OpenAIChatMessage[] = [
+        { content: 'Hello', role: 'user' },
+        { content: 'Hello', role: 'user' },
+        { content: 'Hi', role: 'assistant' },
+      ];
+
+      const contents = await buildAnthropicMessages(messages, { enabledContextCaching: true });
+
+      expect(contents).toHaveLength(3);
+      expect(contents).toEqual([
+        { content: 'Hello', role: 'user' },
+        { content: 'Hello', role: 'user' },
+        {
+          content: [{ cache_control: { type: 'ephemeral' }, text: 'Hi', type: 'text' }],
+          role: 'assistant',
+        },
+      ]);
+    });
   });
 
   describe('buildAnthropicTools', () => {
@@ -656,5 +676,40 @@ describe('anthropicHelpers', () => {
         },
       ]);
     });
+    it('should enable cache control', () => {
+      const tools: OpenAI.ChatCompletionTool[] = [
+        {
+          type: 'function',
+          function: {
+            name: 'search',
+            description: 'Searches the web',
+            parameters: {
+              type: 'object',
+              properties: {
+                query: { type: 'string' },
+              },
+              required: ['query'],
+            },
+          },
+        },
+      ];
+
+      const result = buildAnthropicTools(tools, { enabledContextCaching: true });
+
+      expect(result).toEqual([
+        {
+          name: 'search',
+          description: 'Searches the web',
+          input_schema: {
+            type: 'object',
+            properties: {
+              query: { type: 'string' },
+            },
+            required: ['query'],
+          },
+          cache_control: { type: 'ephemeral' },
+        },
+      ]);
+    });
   });
 });