@lobehub/chat 1.68.10 → 1.69.0

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' },
+        },
+      ]);
+    });
   });
 });

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

@@ -130,6 +130,7 @@ export const buildAnthropicMessage = async (
 
 export const buildAnthropicMessages = async (
   oaiMessages: OpenAIChatMessage[],
+  options: { enabledContextCaching?: boolean } = {},
 ): Promise<Anthropic.Messages.MessageParam[]> => {
   const messages: Anthropic.Messages.MessageParam[] = [];
   let pendingToolResults: Anthropic.ToolResultBlockParam[] = [];
@@ -180,13 +181,46 @@ export const buildAnthropicMessages = async (
     }
   }
 
+  const lastMessage = messages.at(-1);
+  if (options.enabledContextCaching && !!lastMessage) {
+    if (typeof lastMessage.content === 'string') {
+      lastMessage.content = [
+        {
+          cache_control: { type: 'ephemeral' },
+          text: lastMessage.content as string,
+          type: 'text',
+        },
+      ];
+    } else {
+      const lastContent = lastMessage.content.at(-1);
+
+      if (
+        lastContent &&
+        lastContent.type !== 'thinking' &&
+        lastContent.type !== 'redacted_thinking'
+      ) {
+        lastContent.cache_control = { type: 'ephemeral' };
+      }
+    }
+  }
   return messages;
 };
-export const buildAnthropicTools = (tools?: OpenAI.ChatCompletionTool[]) =>
-  tools?.map(
-    (tool): Anthropic.Tool => ({
+
+export const buildAnthropicTools = (
+  tools?: OpenAI.ChatCompletionTool[],
+  options: { enabledContextCaching?: boolean } = {},
+) => {
+  if (!tools) return;
+
+  return tools.map(
+    (tool, index): Anthropic.Tool => ({
+      cache_control:
+        options.enabledContextCaching && index === tools.length - 1
+          ? { type: 'ephemeral' }
+          : undefined,
       description: tool.function.description,
       input_schema: tool.function.parameters as Anthropic.Tool.InputSchema,
       name: tool.function.name,
     }),
   );
+};

package/src/libs/langchain/loaders/code/__tests__/long.json

@@ -164,11 +164,11 @@
     "metadata": { "loc": { "lines": { "from": 547, "to": 570 } } }
   },
   {
-    "pageContent": "const abortController = internal_toggleChatLoading(\n      true,\n      assistantId,\n      n('generateMessage(start)', { assistantId, messages }) as string,\n    );\n\n    const agentConfig = getAgentConfig();\n    const chatConfig = agentConfig.chatConfig;\n\n    const compiler = template(chatConfig.inputTemplate, { interpolate: /{{([\\S\\s]+?)}}/g });\n\n    // ================================== //\n    //   messages uniformly preprocess    //\n    // ================================== //\n\n    // 1. slice messages with config\n    let preprocessMsgs = chatHelpers.getSlicedMessagesWithConfig(messages, chatConfig);",
+    "pageContent": "const abortController = internal_toggleChatLoading(\n      true,\n      assistantId,\n      n('generateMessage(start)', { assistantId, messages }) as string,\n    );\n\n    const agentConfig = getAgentConfig();\n    const chatConfig = agentConfig.chatConfig;\n\n    const compiler = template(chatConfig.inputTemplate, { interpolate: /{{([\\S\\s]+?)}}/g });\n\n    // ================================== //\n    //   messages uniformly preprocess    //\n    // ================================== //\n\n    // 1. slice messages with config\n    let preprocessMsgs = chatHelpers.getSlicedMessages(messages, chatConfig);",
     "metadata": { "loc": { "lines": { "from": 566, "to": 582 } } }
   },
   {
-    "pageContent": "const compiler = template(chatConfig.inputTemplate, { interpolate: /{{([\\S\\s]+?)}}/g });\n\n    // ================================== //\n    //   messages uniformly preprocess    //\n    // ================================== //\n\n    // 1. slice messages with config\n    let preprocessMsgs = chatHelpers.getSlicedMessagesWithConfig(messages, chatConfig);\n\n    // 2. replace inputMessage template\n    preprocessMsgs = !chatConfig.inputTemplate\n      ? preprocessMsgs\n      : preprocessMsgs.map((m) => {\n          if (m.role === 'user') {\n            try {\n              return { ...m, content: compiler({ text: m.content }) };\n            } catch (error) {\n              console.error(error);\n\n              return m;\n            }\n          }\n\n          return m;\n        });",
+    "pageContent": "const compiler = template(chatConfig.inputTemplate, { interpolate: /{{([\\S\\s]+?)}}/g });\n\n    // ================================== //\n    //   messages uniformly preprocess    //\n    // ================================== //\n\n    // 1. slice messages with config\n    let preprocessMsgs = chatHelpers.getSlicedMessages(messages, chatConfig);\n\n    // 2. replace inputMessage template\n    preprocessMsgs = !chatConfig.inputTemplate\n      ? preprocessMsgs\n      : preprocessMsgs.map((m) => {\n          if (m.role === 'user') {\n            try {\n              return { ...m, content: compiler({ text: m.content }) };\n            } catch (error) {\n              console.error(error);\n\n              return m;\n            }\n          }\n\n          return m;\n        });",
     "metadata": { "loc": { "lines": { "from": 575, "to": 599 } } }
   },
   {

package/src/libs/langchain/loaders/code/__tests__/long.txt

@@ -579,7 +579,7 @@ export const chatMessage: StateCreator<
     // ================================== //
 
     // 1. slice messages with config
-    let preprocessMsgs = chatHelpers.getSlicedMessagesWithConfig(messages, chatConfig);
+    let preprocessMsgs = chatHelpers.getSlicedMessages(messages, chatConfig);
 
     // 2. replace inputMessage template
     preprocessMsgs = !chatConfig.inputTemplate

package/src/locales/default/chat.ts

@@ -33,7 +33,12 @@ export default {
   duplicateTitle: '{{title}} 副本',
   emptyAgent: '暂无助手',
   extendParams: {
+    disableContextCaching: {
+      desc: '单条对话生成成本最高可降低 90%，响应速度提升 4 倍（<1>了解更多</1>）。开启后将自动禁用历史消息数限制',
+      title: '开启上下文缓存',
+    },
     enableReasoning: {
+      desc: '基于 Claude Thinking 机制限制（<1>了解更多</1>），开启后将自动禁用历史消息数限制',
       title: '开启深度思考',
     },
     reasoningBudgetToken: {
@@ -41,6 +46,9 @@ export default {
     },
     title: '模型扩展功能',
   },
+  history: {
+    title: '助手将只记住最后{{count}}条消息',
+  },
   historyRange: '历史范围',
   historySummary: '历史消息总结',
   inbox: {

package/src/store/agent/initialState.ts

@@ -1,7 +1,7 @@
 import { AgentState, initialAgentChatState } from './slices/chat/initialState';
 
-export type SessionStoreState = AgentState;
+export type AgentStoreState = AgentState;
 
-export const initialState: SessionStoreState = {
+export const initialState: AgentStoreState = {
   ...initialAgentChatState,
 };

package/src/store/agent/selectors.ts

@@ -1 +1 @@
-export { agentSelectors } from './slices/chat/selectors';
+export { agentChatConfigSelectors,agentSelectors } from './slices/chat/selectors';

package/src/store/agent/slices/chat/{selectors.test.ts → selectors/agent.test.ts}

@@ -6,8 +6,8 @@ import { AgentStore } from '@/store/agent';
 import { AgentState } from '@/store/agent/slices/chat/initialState';
 import { merge } from '@/utils/merge';
 
-import { initialState } from '../../initialState';
-import { agentSelectors } from './selectors';
+import { initialState } from '../../../initialState';
+import { agentSelectors } from './agent';
 
 vi.mock('i18next', () => ({
   t: vi.fn((key) => key), // Simplified mock return value