@lobehub/chat 1.68.2 → 1.68.4

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

@@ -222,6 +222,10 @@ describe('AnthropicStream', () => {
         'id: msg_017aTuY86wNxth5TE544yqJq',
         'event: stop',
         'data: "tool_use"\n',
+
+        'id: msg_017aTuY86wNxth5TE544yqJq',
+        'event: usage',
+        'data: {"inputTokens":457,"outputTokens":84,"totalTokens":541}\n',
       ].map((item) => `${item}\n`),
     );
 
@@ -375,6 +379,10 @@ describe('AnthropicStream', () => {
         'event: stop',
         'data: "tool_use"\n',
 
+        'id: msg_0175ryA67RbGrnRrGBXFQEYK',
+        'event: usage',
+        'data: {"inputTokens":485,"outputTokens":154,"totalTokens":639}\n',
+
         'id: msg_0175ryA67RbGrnRrGBXFQEYK',
         'event: stop',
         'data: "message_stop"\n',
@@ -506,6 +514,9 @@ describe('AnthropicStream', () => {
           'event: stop',
           'data: "end_turn"\n',
           'id: msg_01MNsLe7n1uVLtu6W8rCFujD',
+          'event: usage',
+          'data: {"inputTokens":46,"outputTokens":365,"totalTokens":411}\n',
+          'id: msg_01MNsLe7n1uVLtu6W8rCFujD',
           'event: stop',
           'data: "message_stop"\n',
         ].map((item) => `${item}\n`),
@@ -663,6 +674,9 @@ describe('AnthropicStream', () => {
           'event: stop',
           'data: "end_turn"\n',
           'id: msg_019q32esPvu3TftzZnL6JPys',
+          'event: usage',
+          'data: {"inputTokens":92,"outputTokens":263,"totalTokens":355}\n',
+          'id: msg_019q32esPvu3TftzZnL6JPys',
           'event: stop',
           'data: "message_stop"\n',
         ].map((item) => `${item}\n`),

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

@@ -1,6 +1,8 @@
 import Anthropic from '@anthropic-ai/sdk';
 import type { Stream } from '@anthropic-ai/sdk/streaming';
 
+import { ModelTokensUsage } from '@/types/message';
+
 import { ChatStreamCallbacks } from '../../types';
 import {
   StreamContext,
@@ -20,6 +22,11 @@ export const transformAnthropicStream = (
   switch (chunk.type) {
     case 'message_start': {
       context.id = chunk.message.id;
+      context.usage = {
+        inputTokens: chunk.message.usage?.input_tokens,
+        outputTokens: chunk.message.usage?.output_tokens,
+      };
+
       return { data: chunk.message, id: chunk.message.id, type: 'data' };
     }
     case 'content_block_start': {
@@ -133,6 +140,24 @@ export const transformAnthropicStream = (
     }
 
     case 'message_delta': {
+      const outputTokens = chunk.usage?.output_tokens + (context.usage?.outputTokens || 0);
+      const inputTokens = context.usage?.inputTokens || 0;
+      const totalTokens = inputTokens + outputTokens;
+
+      if (totalTokens > 0) {
+        return [
+          { data: chunk.delta.stop_reason, id: context.id, type: 'stop' },
+          {
+            data: {
+              inputTokens: inputTokens,
+              outputTokens: outputTokens,
+              totalTokens: inputTokens + outputTokens,
+            } as ModelTokensUsage,
+            id: context.id,
+            type: 'usage',
+          },
+        ];
+      }
       return { data: chunk.delta.stop_reason, id: context.id, type: 'stop' };
     }
 

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

@@ -348,6 +348,96 @@ describe('OpenAIStream', () => {
     ]);
   });
 
+  it('should streaming token usage', async () => {
+    const data = [
+      {
+        id: 'chatcmpl-B7CcnaeK3jqWBMOhxg7SSKFwlk7dC',
+        object: 'chat.completion.chunk',
+        created: 1741056525,
+        model: 'gpt-4o-mini-2024-07-18',
+        choices: [{ index: 0, delta: { role: 'assistant', content: '' } }],
+        service_tier: 'default',
+        system_fingerprint: 'fp_06737a9306',
+      },
+      {
+        id: 'chatcmpl-B7CcnaeK3jqWBMOhxg7SSKFwlk7dC',
+        object: 'chat.completion.chunk',
+        created: 1741056525,
+        model: 'gpt-4o-mini-2024-07-18',
+        choices: [{ index: 0, delta: { content: '你好！' } }],
+        service_tier: 'default',
+        system_fingerprint: 'fp_06737a9306',
+      },
+      {
+        id: 'chatcmpl-B7CcnaeK3jqWBMOhxg7SSKFwlk7dC',
+        object: 'chat.completion.chunk',
+        created: 1741056525,
+        model: 'gpt-4o-mini-2024-07-18',
+        choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
+        service_tier: 'default',
+        system_fingerprint: 'fp_06737a9306',
+      },
+      {
+        id: 'chatcmpl-B7CcnaeK3jqWBMOhxg7SSKFwlk7dC',
+        object: 'chat.completion.chunk',
+        created: 1741056525,
+        model: 'gpt-4o-mini-2024-07-18',
+        choices: [],
+        service_tier: 'default',
+        system_fingerprint: 'fp_06737a9306',
+        usage: {
+          prompt_tokens: 1646,
+          completion_tokens: 11,
+          total_tokens: 1657,
+          prompt_tokens_details: { audio_tokens: 0, cached_tokens: 0 },
+          completion_tokens_details: {
+            accepted_prediction_tokens: 0,
+            audio_tokens: 0,
+            reasoning_tokens: 0,
+            rejected_prediction_tokens: 0,
+          },
+        },
+      },
+    ];
+
+    const mockOpenAIStream = new ReadableStream({
+      start(controller) {
+        data.forEach((chunk) => {
+          controller.enqueue(chunk);
+        });
+
+        controller.close();
+      },
+    });
+
+    const protocolStream = OpenAIStream(mockOpenAIStream);
+
+    const decoder = new TextDecoder();
+    const chunks = [];
+
+    // @ts-ignore
+    for await (const chunk of protocolStream) {
+      chunks.push(decoder.decode(chunk, { stream: true }));
+    }
+
+    expect(chunks).toEqual(
+      [
+        'id: chatcmpl-B7CcnaeK3jqWBMOhxg7SSKFwlk7dC',
+        'event: text',
+        `data: ""\n`,
+        'id: chatcmpl-B7CcnaeK3jqWBMOhxg7SSKFwlk7dC',
+        'event: text',
+        `data: "你好！"\n`,
+        'id: chatcmpl-B7CcnaeK3jqWBMOhxg7SSKFwlk7dC',
+        'event: stop',
+        `data: "stop"\n`,
+        'id: chatcmpl-B7CcnaeK3jqWBMOhxg7SSKFwlk7dC',
+        'event: usage',
+        `data: {"acceptedPredictionTokens":0,"cachedTokens":0,"inputAudioTokens":0,"inputTokens":1646,"outputAudioTokens":0,"outputTokens":11,"reasoningTokens":0,"rejectedPredictionTokens":0,"totalTokens":1657}\n`,
+      ].map((i) => `${i}\n`),
+    );
+  });
+
   describe('Tools Calling', () => {
     it('should handle OpenAI official tool calls', async () => {
       const mockOpenAIStream = new ReadableStream({
@@ -749,8 +839,8 @@ describe('OpenAIStream', () => {
           'event: text',
           `data: "帮助。"\n`,
           'id: 1',
-          'event: stop',
-          `data: "stop"\n`,
+          'event: usage',
+          `data: {"cachedTokens":0,"inputCacheMissTokens":6,"inputTokens":6,"outputTokens":104,"reasoningTokens":70,"totalTokens":110}\n`,
         ].map((i) => `${i}\n`),
       );
     });
@@ -968,8 +1058,8 @@ describe('OpenAIStream', () => {
           'event: text',
           `data: "帮助。"\n`,
           'id: 1',
-          'event: stop',
-          `data: "stop"\n`,
+          'event: usage',
+          `data: {"cachedTokens":0,"inputCacheMissTokens":6,"inputTokens":6,"outputTokens":104,"reasoningTokens":70,"totalTokens":110}\n`,
         ].map((i) => `${i}\n`),
       );
     });
@@ -1169,8 +1259,8 @@ describe('OpenAIStream', () => {
           'event: text',
           `data: "帮助。"\n`,
           'id: 1',
-          'event: stop',
-          `data: "stop"\n`,
+          'event: usage',
+          `data: {"cachedTokens":0,"inputCacheMissTokens":6,"inputTokens":6,"outputTokens":104,"reasoningTokens":70,"totalTokens":110}\n`,
         ].map((i) => `${i}\n`),
       );
     });
@@ -1370,8 +1460,8 @@ describe('OpenAIStream', () => {
           'event: text',
           `data: "帮助。"\n`,
           'id: 1',
-          'event: stop',
-          `data: "stop"\n`,
+          'event: usage',
+          `data: {"cachedTokens":0,"inputCacheMissTokens":6,"inputTokens":6,"outputTokens":104,"reasoningTokens":70,"totalTokens":110}\n`,
         ].map((i) => `${i}\n`),
       );
     });
@@ -1571,8 +1661,8 @@ describe('OpenAIStream', () => {
           'event: text',
           `data: "帮助。"\n`,
           'id: 1',
-          'event: stop',
-          `data: "stop"\n`,
+          'event: usage',
+          `data: {"cachedTokens":0,"inputCacheMissTokens":6,"inputTokens":6,"outputTokens":104,"reasoningTokens":70,"totalTokens":110}\n`,
         ].map((i) => `${i}\n`),
       );
     });

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

@@ -1,7 +1,7 @@
 import OpenAI from 'openai';
 import type { Stream } from 'openai/streaming';
 
-import { ChatMessageError, CitationItem } from '@/types/message';
+import { ChatMessageError, CitationItem, ModelTokensUsage } from '@/types/message';
 
 import { AgentRuntimeErrorType, ILobeAgentRuntimeErrorType } from '../../error';
 import { ChatStreamCallbacks } from '../../types';
@@ -18,6 +18,22 @@ import {
   generateToolCallId,
 } from './protocol';
 
+const convertUsage = (usage: OpenAI.Completions.CompletionUsage): ModelTokensUsage => {
+  return {
+    acceptedPredictionTokens: usage.completion_tokens_details?.accepted_prediction_tokens,
+    cachedTokens:
+      (usage as any).prompt_cache_hit_tokens || usage.prompt_tokens_details?.cached_tokens,
+    inputAudioTokens: usage.prompt_tokens_details?.audio_tokens,
+    inputCacheMissTokens: (usage as any).prompt_cache_miss_tokens,
+    inputTokens: usage.prompt_tokens,
+    outputAudioTokens: usage.completion_tokens_details?.audio_tokens,
+    outputTokens: usage.completion_tokens,
+    reasoningTokens: usage.completion_tokens_details?.reasoning_tokens,
+    rejectedPredictionTokens: usage.completion_tokens_details?.rejected_prediction_tokens,
+    totalTokens: usage.total_tokens,
+  };
+};
+
 export const transformOpenAIStream = (
   chunk: OpenAI.ChatCompletionChunk,
   streamContext: StreamContext,
@@ -41,11 +57,16 @@ export const transformOpenAIStream = (
     // maybe need another structure to add support for multiple choices
     const item = chunk.choices[0];
     if (!item) {
+      if (chunk.usage) {
+        const usage = chunk.usage;
+        return { data: convertUsage(usage), id: chunk.id, type: 'usage' };
+      }
+
       return { data: chunk, id: chunk.id, type: 'data' };
     }
 
-    // tools calling
-    if (typeof item.delta?.tool_calls === 'object' && item.delta.tool_calls?.length > 0) {
+    if (item && typeof item.delta?.tool_calls === 'object' && item.delta.tool_calls?.length > 0) {
+      // tools calling
       const tool_calls = item.delta.tool_calls.filter(
         (value) => value.index >= 0 || typeof value.index === 'undefined',
       );
@@ -97,6 +118,11 @@ export const transformOpenAIStream = (
         return { data: item.delta.content, id: chunk.id, type: 'text' };
       }
 
+      if (chunk.usage) {
+        const usage = chunk.usage;
+        return { data: convertUsage(usage), id: chunk.id, type: 'usage' };
+      }
+
       return { data: item.finish_reason, id: chunk.id, type: 'stop' };
     }
 
@@ -147,7 +173,7 @@ export const transformOpenAIStream = (
                       ({
                         title: typeof item === 'string' ? item : item.title,
                         url: typeof item === 'string' ? item : item.url,
-                      }) as CitationItem
+                      }) as CitationItem,
                   ),
                 },
                 id: chunk.id,

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

@@ -1,4 +1,5 @@
 import { ChatStreamCallbacks } from '@/libs/agent-runtime';
+import { ModelTokensUsage } from '@/types/message';
 
 import { AgentRuntimeErrorType } from '../../error';
 
@@ -23,6 +24,7 @@ export interface StreamContext {
     name: string;
   };
   toolIndex?: number;
+  usage?: ModelTokensUsage;
 }
 
 export interface StreamProtocolChunk {
@@ -44,6 +46,8 @@ export interface StreamProtocolChunk {
     | 'stop'
     // Error
     | 'error'
+    // token usage
+    | 'usage'
     // unknown data result
     | 'data';
 }

package/src/locales/default/chat.ts

@@ -81,6 +81,36 @@ export default {
     deleteDisabledByThreads: '存在子话题，不能删除',
     regenerate: '重新生成',
   },
+  messages: {
+    modelCard: {
+      credit: '积分',
+      creditPricing: '定价',
+      creditTooltip:
+        '为便于计数，我们将 1$ 折算为 1M 积分，例如 $3/M tokens 即可折算为 3积分/token',
+      pricing: {
+        inputCachedTokens: '缓存输入 {{amount}}/积分 · ${{amount}}/M',
+        inputCharts: '${{amount}}/M 字符',
+        inputMinutes: '${{amount}}/分钟',
+        inputTokens: '输入 {{amount}}/积分 · ${{amount}}/M',
+        outputTokens: '输出 {{amount}}/积分 · ${{amount}}/M',
+      },
+    },
+    tokenDetails: {
+      input: '输入',
+      inputAudio: '音频输入',
+      inputCached: '输入缓存',
+      inputText: '文本输入',
+      inputTitle: '输入明细',
+      inputUncached: '输入未缓存',
+      output: '输出',
+      outputAudio: '音频输出',
+      outputText: '文本输出',
+      outputTitle: '输出明细',
+      reasoning: '深度思考',
+      title: '生成明细',
+      total: '总计消耗',
+    },
+  },
   newAgent: '新建助手',
   pin: '置顶',
   pinOff: '取消置顶',
@@ -194,7 +224,6 @@ export default {
     action: '语音朗读',
     clear: '删除语音',
   },
-
   updateAgent: '更新助理信息',
   upload: {
     action: {

package/src/server/routers/tools/search.ts

@@ -27,7 +27,7 @@ export const searchRouter = router({
         async (url) => {
           return await crawler.crawl({ impls: input.impls, url });
         },
-        { concurrency: 10 },
+        { concurrency: 3 },
       );
 
       return { results };

package/src/store/aiInfra/slices/aiModel/initialState.ts

@@ -1,8 +1,9 @@
-import { AiProviderModelListItem } from '@/types/aiModel';
+import { AiProviderModelListItem, LobeDefaultAiModelListItem } from '@/types/aiModel';
 
 export interface AIModelsState {
   aiModelLoadingIds: string[];
   aiProviderModelList: AiProviderModelListItem[];
+  builtinAiModelList: LobeDefaultAiModelListItem[];
   isAiModelListInit?: boolean;
   modelSearchKeyword: string;
 }
@@ -10,5 +11,6 @@ export interface AIModelsState {
 export const initialAIModelState: AIModelsState = {
   aiModelLoadingIds: [],
   aiProviderModelList: [],
+  builtinAiModelList: [],
   modelSearchKeyword: '',
 };

package/src/store/aiInfra/slices/aiModel/selectors.test.ts

@@ -34,6 +34,7 @@ describe('aiModelSelectors', () => {
         displayName: 'Remote Model',
       },
     ],
+    builtinAiModelList: [],
     modelSearchKeyword: '',
     aiModelLoadingIds: ['model2'],
     enabledAiModels: [

package/src/store/aiInfra/slices/aiModel/selectors.ts

@@ -22,8 +22,12 @@ const filteredAiProviderModelList = (s: AIProviderStoreState) => {
 };
 
 const totalAiProviderModelList = (s: AIProviderStoreState) => s.aiProviderModelList.length;
+
 const isEmptyAiProviderModelList = (s: AIProviderStoreState) => totalAiProviderModelList(s) === 0;
 
+const getModelCard = (model: string, provider: string) => (s: AIProviderStoreState) =>
+  s.builtinAiModelList.find((item) => item.id === model && item.providerId === provider);
+
 const hasRemoteModels = (s: AIProviderStoreState) =>
   s.aiProviderModelList.some((m) => m.source === AiModelSourceEnum.Remote);
 
@@ -113,6 +117,7 @@ export const aiModelSelectors = {
   filteredAiProviderModelList,
   getAiModelById,
   getEnabledModelById,
+  getModelCard,
   hasRemoteModels,
   isEmptyAiProviderModelList,
   isModelEnabled,

package/src/store/aiInfra/slices/aiProvider/action.ts

@@ -184,7 +184,7 @@ export const createAiProviderSlice: StateCreator<
         };
       },
       {
-        onSuccess: (data) => {
+        onSuccess: async (data) => {
           if (!data) return;
 
           const getModelListByType = (providerId: string, type: string) => {
@@ -206,10 +206,12 @@ export const createAiProviderSlice: StateCreator<
             children: getModelListByType(provider.id, 'chat'),
             name: provider.name || provider.id,
           }));
+          const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
 
           set(
             {
               aiProviderRuntimeConfig: data.runtimeConfig,
+              builtinAiModelList: LOBE_DEFAULT_MODEL_LIST,
               enabledAiModels: data.enabledAiModels,
               enabledAiProviders: data.enabledAiProviders,
               enabledChatModelList,

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

@@ -455,7 +455,10 @@ export const generateAIChat: StateCreator<
         await messageService.updateMessageError(messageId, error);
         await refreshMessages();
       },
-      onFinish: async (content, { traceId, observationId, toolCalls, reasoning, grounding }) => {
+      onFinish: async (
+        content,
+        { traceId, observationId, toolCalls, reasoning, grounding, usage },
+      ) => {
         // if there is traceId, update it
         if (traceId) {
           msgTraceId = traceId;
@@ -474,6 +477,7 @@ export const generateAIChat: StateCreator<
           toolCalls,
           reasoning: !!reasoning ? { ...reasoning, duration } : undefined,
           search: !!grounding?.citations ? grounding : undefined,
+          metadata: usage,
         });
       },
       onMessageHandle: async (chunk) => {

package/src/store/chat/slices/message/action.ts

@@ -17,6 +17,7 @@ import {
   ChatMessageError,
   ChatMessagePluginError,
   CreateMessageParams,
+  MessageMetadata,
   MessageToolCall,
   ModelReasoning,
 } from '@/types/message';
@@ -79,6 +80,7 @@ export interface ChatMessageAction {
       toolCalls?: MessageToolCall[];
       reasoning?: ModelReasoning;
       search?: GroundingSearch;
+      metadata?: MessageMetadata;
     },
   ) => Promise<void>;
   /**
@@ -308,6 +310,7 @@ export const chatMessage: StateCreator<
       tools: extra?.toolCalls ? internal_transformToolCalls(extra?.toolCalls) : undefined,
       reasoning: extra?.reasoning,
       search: extra?.search,
+      metadata: extra?.metadata,
     });
     await refreshMessages();
   },

package/src/store/global/initialState.ts

@@ -50,6 +50,7 @@ export interface SystemStatus {
    * 应用初始化时不启用 PGLite，只有当用户手动开启时才启用
    */
   isEnablePglite?: boolean;
+  isShowCredit?: boolean;
   language?: LocaleMode;
   latestChangelogId?: string;
   mobileShowPortal?: boolean;

package/src/store/global/selectors/systemStatus.ts

@@ -16,6 +16,7 @@ const showChatSideBar = (s: GlobalStore) => !s.status.zenMode && s.status.showCh
 const showSessionPanel = (s: GlobalStore) => !s.status.zenMode && s.status.showSessionPanel;
 const showFilePanel = (s: GlobalStore) => s.status.showFilePanel;
 const hidePWAInstaller = (s: GlobalStore) => s.status.hidePWAInstaller;
+const isShowCredit = (s: GlobalStore) => s.status.isShowCredit;
 
 const showChatHeader = (s: GlobalStore) => !s.status.zenMode;
 const inZenMode = (s: GlobalStore) => s.status.zenMode;
@@ -58,6 +59,7 @@ export const systemStatusSelectors = {
   isPgliteInited,
   isPgliteNotEnabled,
   isPgliteNotInited,
+  isShowCredit,
   mobileShowPortal,
   mobileShowTopic,
   portalWidth,

package/src/types/message/base.ts

@@ -13,6 +13,23 @@ export interface ModelReasoning {
   signature?: string;
 }
 
+export interface ModelTokensUsage {
+  acceptedPredictionTokens?: number;
+  cachedTokens?: number;
+  inputAudioTokens?: number;
+  inputCacheMissTokens?: number;
+  inputTokens?: number;
+  outputAudioTokens?: number;
+  outputTokens?: number;
+  reasoningTokens?: number;
+  rejectedPredictionTokens?: number;
+  totalTokens?: number;
+}
+
+export interface MessageMetadata extends ModelTokensUsage {
+  tps?: number;
+}
+
 export type MessageRoleType = 'user' | 'system' | 'assistant' | 'tool';
 
 export interface MessageItem {
@@ -23,6 +40,7 @@ export interface MessageItem {
   error: any | null;
   favorite: boolean | null;
   id: string;
+  metadata?: MessageMetadata | null;
   model: string | null;
   observationId: string | null;
   parentId: string | null;

package/src/types/message/chat.ts

@@ -6,7 +6,7 @@ import { MetaData } from '@/types/meta';
 import { MessageSemanticSearchChunk } from '@/types/rag';
 import { GroundingSearch } from '@/types/search';
 
-import { MessageRoleType, ModelReasoning } from './base';
+import { MessageMetadata, MessageRoleType, ModelReasoning } from './base';
 import { ChatPluginPayload, ChatToolPayload } from './tools';
 import { Translate } from './translate';
 
@@ -82,15 +82,16 @@ export interface ChatMessage {
   imageList?: ChatImageItem[];
   meta: MetaData;
 
+  metadata?: MessageMetadata | null;
   /**
    * observation id
    */
   observationId?: string;
+
   /**
    * parent message id
    */
   parentId?: string;
-
   plugin?: ChatPluginPayload;
   pluginError?: any;
   pluginState?: any;
@@ -100,8 +101,8 @@ export interface ChatMessage {
   quotaId?: string;
   ragQuery?: string | null;
   ragQueryId?: string | null;
-  ragRawQuery?: string | null;
 
+  ragRawQuery?: string | null;
   reasoning?: ModelReasoning | null;
   /**
    * message role type

package/src/utils/fetch/fetchSSE.ts

@@ -10,6 +10,7 @@ import {
   MessageToolCallChunk,
   MessageToolCallSchema,
   ModelReasoning,
+  ModelTokensUsage,
 } from '@/types/message';
 import { GroundingSearch } from '@/types/search';
 
@@ -28,9 +29,15 @@ export type OnFinishHandler = (
     toolCalls?: MessageToolCall[];
     traceId?: string | null;
     type?: SSEFinishType;
+    usage?: ModelTokensUsage;
   },
 ) => Promise<void>;
 
+export interface MessageUsageChunk {
+  type: 'usage';
+  usage: ModelTokensUsage;
+}
+
 export interface MessageTextChunk {
   text: string;
   type: 'text';
@@ -59,7 +66,12 @@ export interface FetchSSEOptions {
   onErrorHandle?: (error: ChatMessageError) => void;
   onFinish?: OnFinishHandler;
   onMessageHandle?: (
-    chunk: MessageTextChunk | MessageToolCallsChunk | MessageReasoningChunk | MessageGroundingChunk,
+    chunk:
+      | MessageTextChunk
+      | MessageToolCallsChunk
+      | MessageReasoningChunk
+      | MessageGroundingChunk
+      | MessageUsageChunk,
   ) => void;
   smoothing?: SmoothingParams | boolean;
 }
@@ -317,6 +329,7 @@ export const fetchSSE = async (url: string, options: RequestInit & FetchSSEOptio
   });
 
   let grounding: GroundingSearch | undefined = undefined;
+  let usage: ModelTokensUsage | undefined = undefined;
   await fetchEventSource(url, {
     body: options.body,
     fetch: options?.fetcher,
@@ -377,6 +390,9 @@ export const fetchSSE = async (url: string, options: RequestInit & FetchSSEOptio
         }
 
         case 'text': {
+          // skip empty text
+          if (!data) break;
+
           if (textSmoothing) {
             textController.pushToQueue(data);
 
@@ -389,6 +405,12 @@ export const fetchSSE = async (url: string, options: RequestInit & FetchSSEOptio
           break;
         }
 
+        case 'usage': {
+          usage = data;
+          options.onMessageHandle?.({ type: 'usage', usage: data });
+          break;
+        }
+
         case 'grounding': {
           grounding = data;
           options.onMessageHandle?.({ grounding: data, type: 'grounding' });
@@ -475,6 +497,7 @@ export const fetchSSE = async (url: string, options: RequestInit & FetchSSEOptio
         toolCalls,
         traceId,
         type: finishedType,
+        usage,
       });
     }
   }