@lobehub/chat 1.96.7 → 1.96.9

package/CHANGELOG.md

@@ -2,6 +2,56 @@
 
 # Changelog
 
+### [Version 1.96.9](https://github.com/lobehub/lobe-chat/compare/v1.96.8...v1.96.9)
+
+<sup>Released on **2025-06-23**</sup>
+
+#### 🐛 Bug Fixes
+
+- **misc**: Google Gemini tools declarations.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **misc**: Google Gemini tools declarations, closes [#8256](https://github.com/lobehub/lobe-chat/issues/8256) ([08f5d73](https://github.com/lobehub/lobe-chat/commit/08f5d73))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
+### [Version 1.96.8](https://github.com/lobehub/lobe-chat/compare/v1.96.7...v1.96.8)
+
+<sup>Released on **2025-06-23**</sup>
+
+#### 💄 Styles
+
+- **misc**: Optimized Gemini thinkingBudget configuration.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### Styles
+
+- **misc**: Optimized Gemini thinkingBudget configuration, closes [#8224](https://github.com/lobehub/lobe-chat/issues/8224) ([03625e8](https://github.com/lobehub/lobe-chat/commit/03625e8))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ### [Version 1.96.7](https://github.com/lobehub/lobe-chat/compare/v1.96.6...v1.96.7)
 
 <sup>Released on **2025-06-23**</sup>

package/changelog/v1.json

@@ -1,4 +1,22 @@
 [
+  {
+    "children": {
+      "fixes": [
+        "Google Gemini tools declarations."
+      ]
+    },
+    "date": "2025-06-23",
+    "version": "1.96.9"
+  },
+  {
+    "children": {
+      "improvements": [
+        "Optimized Gemini thinkingBudget configuration."
+      ]
+    },
+    "date": "2025-06-23",
+    "version": "1.96.8"
+  },
   {
     "children": {
       "improvements": [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.96.7",
+  "version": "1.96.9",
   "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/google.ts

@@ -20,7 +20,7 @@ const googleChatModels: AIChatModelCard[] = [
     },
     releasedAt: '2025-06-17',
     settings: {
-      extendParams: ['enableReasoning', 'reasoningBudgetToken'],
+      extendParams: ['thinkingBudget'],
       searchImpl: 'params',
       searchProvider: 'google',
     },
@@ -45,7 +45,7 @@ const googleChatModels: AIChatModelCard[] = [
     },
     releasedAt: '2025-06-05',
     settings: {
-      extendParams: ['enableReasoning', 'reasoningBudgetToken'],
+      extendParams: ['thinkingBudget'],
       searchImpl: 'params',
       searchProvider: 'google',
     },
@@ -119,7 +119,7 @@ const googleChatModels: AIChatModelCard[] = [
     },
     releasedAt: '2025-06-17',
     settings: {
-      extendParams: ['enableReasoning', 'reasoningBudgetToken'],
+      extendParams: ['thinkingBudget'],
       searchImpl: 'params',
       searchProvider: 'google',
     },
@@ -143,7 +143,7 @@ const googleChatModels: AIChatModelCard[] = [
     },
     releasedAt: '2025-05-20',
     settings: {
-      extendParams: ['enableReasoning', 'reasoningBudgetToken'],
+      extendParams: ['thinkingBudget'],
       searchImpl: 'params',
       searchProvider: 'google',
     },
@@ -167,7 +167,7 @@ const googleChatModels: AIChatModelCard[] = [
     },
     releasedAt: '2025-04-17',
     settings: {
-      extendParams: ['enableReasoning', 'reasoningBudgetToken'],
+      extendParams: ['thinkingBudget'],
       searchImpl: 'params',
       searchProvider: 'google',
     },
@@ -215,7 +215,7 @@ const googleChatModels: AIChatModelCard[] = [
     },
     releasedAt: '2025-06-11',
     settings: {
-      extendParams: ['enableReasoning', 'reasoningBudgetToken'],
+      extendParams: ['thinkingBudget'],
       searchImpl: 'params',
       searchProvider: 'google',
     },

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

@@ -13,6 +13,7 @@ import { aiModelSelectors, useAiInfraStore } from '@/store/aiInfra';
 import ContextCachingSwitch from './ContextCachingSwitch';
 import ReasoningEffortSlider from './ReasoningEffortSlider';
 import ReasoningTokenSlider from './ReasoningTokenSlider';
+import ThinkingBudgetSlider from './ThinkingBudgetSlider';
 
 const ControlsForm = memo(() => {
   const { t } = useTranslation('chat');
@@ -93,6 +94,17 @@ const ControlsForm = memo(() => {
         paddingBottom: 0,
       },
     },
+    {
+      children: <ThinkingBudgetSlider />,
+      label: t('extendParams.reasoningBudgetToken.title'),
+      layout: 'vertical',
+      minWidth: 500,
+      name: 'thinkingBudget',
+      style: {
+        paddingBottom: 0,
+      },
+      tag: 'thinkingBudget',
+    },
   ].filter(Boolean) as FormItemProps[];
 
   return (

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

@@ -0,0 +1,133 @@
+import { InputNumber } from '@lobehub/ui';
+import { Slider } from 'antd';
+import { memo, useMemo } from 'react';
+import { Flexbox } from 'react-layout-kit';
+import useMergeState from 'use-merge-value';
+
+// 定义特殊值映射
+const SPECIAL_VALUES = {
+  AUTO: -1,
+  OFF: 0,
+};
+
+// 定义滑块位置到实际值的映射
+const SLIDER_TO_VALUE_MAP = [
+  SPECIAL_VALUES.AUTO, // 位置 0 -> -1 (Auto)
+  SPECIAL_VALUES.OFF, // 位置 1 -> 0 (OFF)
+  128, // 位置 2 -> 128
+  512, // 位置 3 -> 512
+  1024, // 位置 4 -> 1024
+  2048, // 位置 5 -> 2048
+  4096, // 位置 6 -> 4096
+  8192, // 位置 7 -> 8192
+  16_384, // 位置 8 -> 16384
+  24_576, // 位置 9 -> 24576
+  32_768, // 位置 10 -> 32768
+];
+
+// 从实际值获取滑块位置
+const getSliderPosition = (value: number): number => {
+  const index = SLIDER_TO_VALUE_MAP.indexOf(value);
+  return index === -1 ? 0 : index;
+};
+
+// 从滑块位置获取实际值（修复：0 不再被当作 falsy）
+const getValueFromPosition = (position: number): number => {
+  const v = SLIDER_TO_VALUE_MAP[position];
+  return v === undefined ? SPECIAL_VALUES.AUTO : v;
+};
+
+interface ThinkingBudgetSliderProps {
+  defaultValue?: number;
+  onChange?: (value: number) => void;
+  value?: number;
+}
+
+const ThinkingBudgetSlider = memo<ThinkingBudgetSliderProps>(
+  ({ value, onChange, defaultValue }) => {
+    // 首先确定初始的 budget 值
+    const initialBudget = value ?? defaultValue ?? SPECIAL_VALUES.AUTO;
+
+    const [budget, setBudget] = useMergeState(initialBudget, {
+      defaultValue,
+      onChange,
+      value,
+    });
+
+    const sliderPosition = getSliderPosition(budget);
+
+    const updateWithSliderPosition = (position: number) => {
+      const newValue = getValueFromPosition(position);
+      setBudget(newValue);
+    };
+
+    const updateWithRealValue = (value: number) => {
+      setBudget(value);
+    };
+
+    const marks = useMemo(() => {
+      return {
+        0: 'Auto',
+        1: 'OFF',
+        2: '128',
+        3: '512',
+        4: '1K',
+        5: '2K',
+        6: '4K',
+        7: '8K',
+        8: '16K',
+        9: '24K',
+        // eslint-disable-next-line sort-keys-fix/sort-keys-fix
+        10: '32K',
+      };
+    }, []);
+
+    return (
+      <Flexbox align={'center'} gap={12} horizontal paddingInline={'4px 0'}>
+        <Flexbox flex={1}>
+          <Slider
+            marks={marks}
+            max={10}
+            min={0}
+            onChange={updateWithSliderPosition}
+            step={null}
+            tooltip={{ open: false }}
+            value={sliderPosition}
+          />
+        </Flexbox>
+        <div>
+          <InputNumber
+            // eslint-disable-next-line @typescript-eslint/no-unused-vars
+            formatter={(value, _info) => {
+              if (value === SPECIAL_VALUES.AUTO) return 'Auto';
+              if (value === SPECIAL_VALUES.OFF) return 'OFF';
+              return `${value}`;
+            }}
+            max={32_768}
+            min={-1}
+            onChange={(e) => {
+              if (e === null || e === undefined) return;
+              updateWithRealValue(e as number);
+            }}
+            parser={(value) => {
+              if (typeof value === 'string') {
+                if (value.toLowerCase() === 'auto') return SPECIAL_VALUES.AUTO;
+                if (value.toLowerCase() === 'off') return SPECIAL_VALUES.OFF;
+                return parseInt(value.replaceAll(/[^\d-]/g, ''), 10) || 0;
+              }
+              if (typeof value === 'number') {
+                return value;
+              }
+              return SPECIAL_VALUES.AUTO;
+            }}
+            step={128}
+            style={{ width: 80 }}
+            value={budget}
+          />
+        </div>
+      </Flexbox>
+    );
+  },
+);
+
+export default ThinkingBudgetSlider;

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

@@ -560,6 +560,58 @@ describe('LobeGoogleAI', () => {
           },
         ]);
       });
+
+      it('should correctly convert function response message', async () => {
+        const messages: OpenAIChatMessage[] = [
+          {
+            content: '',
+            role: 'assistant',
+            tool_calls: [
+              {
+                id: 'call_1',
+                function: {
+                  name: 'get_current_weather',
+                  arguments: JSON.stringify({ location: 'London', unit: 'celsius' }),
+                },
+                type: 'function',
+              },
+            ],
+          },
+          {
+            content: '{"success":true,"data":{"temperature":"14°C"}}',
+            name: 'get_current_weather',
+            role: 'tool',
+            tool_call_id: 'call_1',
+          },
+        ];
+
+        const contents = await instance['buildGoogleMessages'](messages);
+        expect(contents).toHaveLength(2);
+        expect(contents).toEqual([
+          {
+            parts: [
+              {
+                functionCall: {
+                  args: { location: 'London', unit: 'celsius' },
+                  name: 'get_current_weather',
+                },
+              },
+            ],
+            role: 'model',
+          },
+          {
+            parts: [
+              {
+                functionResponse: {
+                  name: 'get_current_weather',
+                  response: { result: '{"success":true,"data":{"temperature":"14°C"}}' },
+                },
+              },
+            ],
+            role: 'user',
+          },
+        ]);
+      });
     });
 
     describe('buildGoogleTools', () => {
@@ -690,7 +742,7 @@ describe('LobeGoogleAI', () => {
 
         const converted = await instance['convertOAIMessagesToGoogleMessage'](message);
         expect(converted).toEqual({
-          role: 'function',
+          role: 'model',
           parts: [
             {
               functionCall: {

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

@@ -120,28 +120,37 @@ export class LobeGoogleAI implements LobeRuntimeAI {
   async chat(rawPayload: ChatStreamPayload, options?: ChatMethodOptions) {
     try {
       const payload = this.buildPayload(rawPayload);
-      const { model, thinking } = payload;
+      const { model, thinkingBudget } = payload;
 
       const thinkingConfig: GoogleAIThinkingConfig = {
         includeThoughts:
-          thinking?.type === 'enabled' ||
-          (!thinking && model && (model.includes('-2.5-') || model.includes('thinking')))
+          !!thinkingBudget ||
+          (!thinkingBudget && model && (model.includes('-2.5-') || model.includes('thinking')))
             ? true
             : undefined,
-        thinkingBudget:
-          thinking?.type === 'enabled'
-            ? (() => {
-                const budget = thinking.budget_tokens;
-                if (model.includes('-2.5-flash')) {
-                  return Math.min(budget, 24_576);
-                } else if (model.includes('-2.5-pro')) {
-                  return Math.max(128, Math.min(budget, 32_768));
-                }
-                return Math.min(budget, 24_576);
-              })()
-            : thinking?.type === 'disabled'
-              ? model.includes('-2.5-pro') ? 128 : 0
-              : undefined,
+        // https://ai.google.dev/gemini-api/docs/thinking#set-budget
+        thinkingBudget: (() => {
+          if (thinkingBudget !== undefined && thinkingBudget !== null) {
+            if (model.includes('-2.5-flash-lite')) {
+              if (thinkingBudget === 0 || thinkingBudget === -1) {
+                return thinkingBudget;
+              }
+              return Math.max(512, Math.min(thinkingBudget, 24_576));
+            } else if (model.includes('-2.5-flash')) {
+              return Math.min(thinkingBudget, 24_576);
+            } else if (model.includes('-2.5-pro')) {
+              return Math.max(128, Math.min(thinkingBudget, 32_768));
+            }
+            return Math.min(thinkingBudget, 24_576);
+          }
+
+          if (model.includes('-2.5-pro') || model.includes('-2.5-flash')) {
+            return -1;
+          } else if (model.includes('-2.5-flash-lite')) {
+            return 0;
+          }
+          return undefined;
+        })(),
       };
 
       const contents = await this.buildGoogleMessages(payload.messages);
@@ -344,6 +353,7 @@ export class LobeGoogleAI implements LobeRuntimeAI {
       system: system_message?.content,
     };
   }
+
   private convertContentToGooglePart = async (
     content: UserMessageContentPart,
   ): Promise<Part | undefined> => {
@@ -390,6 +400,7 @@ export class LobeGoogleAI implements LobeRuntimeAI {
 
   private convertOAIMessagesToGoogleMessage = async (
     message: OpenAIChatMessage,
+    toolCallNameMap?: Map<string, string>,
   ): Promise<Content> => {
     const content = message.content as string | UserMessageContentPart[];
     if (!!message.tool_calls) {
@@ -400,10 +411,28 @@ export class LobeGoogleAI implements LobeRuntimeAI {
             name: tool.function.name,
           },
         })),
-        role: 'function',
+        role: 'model',
       };
     }
 
+    // 将 tool_call result 转成 functionResponse part
+    if (message.role === 'tool' && toolCallNameMap && message.tool_call_id) {
+      const functionName = toolCallNameMap.get(message.tool_call_id);
+      if (functionName) {
+        return {
+          parts: [
+            {
+              functionResponse: {
+                name: functionName,
+                response: { result: message.content },
+              },
+            },
+          ],
+          role: 'user',
+        };
+      }
+    }
+
     const getParts = async () => {
       if (typeof content === 'string') return [{ text: content }];
 
@@ -421,9 +450,20 @@ export class LobeGoogleAI implements LobeRuntimeAI {
 
   // convert messages from the OpenAI format to Google GenAI SDK
   private buildGoogleMessages = async (messages: OpenAIChatMessage[]): Promise<Content[]> => {
+    const toolCallNameMap = new Map<string, string>();
+    messages.forEach((message) => {
+      if (message.role === 'assistant' && message.tool_calls) {
+        message.tool_calls.forEach((toolCall) => {
+          if (toolCall.type === 'function') {
+            toolCallNameMap.set(toolCall.id, toolCall.function.name);
+          }
+        });
+      }
+    });
+
     const pools = messages
       .filter((message) => message.role !== 'function')
-      .map(async (msg) => await this.convertOAIMessagesToGoogleMessage(msg));
+      .map(async (msg) => await this.convertOAIMessagesToGoogleMessage(msg, toolCallNameMap));
 
     return Promise.all(pools);
   };
@@ -484,12 +524,18 @@ export class LobeGoogleAI implements LobeRuntimeAI {
   ): GoogleFunctionCallTool[] | undefined {
     // 目前 Tools (例如 googleSearch) 无法与其他 FunctionCall 同时使用
     if (payload?.messages?.some((m) => m.tool_calls?.length)) {
-      return; // 若历史消息中已有 function calling，则不再注入任何 Tools
+      return this.buildFunctionDeclarations(tools);
     }
     if (payload?.enabledSearch) {
       return [{ googleSearch: {} } as GoogleSearchRetrievalTool];
     }
 
+    return this.buildFunctionDeclarations(tools);
+  }
+
+  private buildFunctionDeclarations(
+    tools: ChatCompletionTool[] | undefined,
+  ): GoogleFunctionCallTool[] | undefined {
     if (!tools || tools.length === 0) return;
 
     return [

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

@@ -126,6 +126,7 @@ export interface ChatStreamPayload {
     budget_tokens: number;
     type: 'enabled' | 'disabled';
   };
+  thinkingBudget?: number;
   tool_choice?: string;
   tools?: ChatCompletionTool[];
   /**

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

@@ -7,7 +7,7 @@ import { GoogleGenerativeAIStream } from './google-ai';
 
 describe('GoogleGenerativeAIStream', () => {
   it('should transform Google Generative AI stream to protocol stream', async () => {
-    vi.spyOn(uuidModule, 'nanoid').mockReturnValueOnce('1');
+    vi.spyOn(uuidModule, 'nanoid').mockReturnValueOnce('1').mockReturnValueOnce('abcd1234');
 
     const mockGenerateContentResponse = (text: string, functionCalls?: any[]) =>
       ({
@@ -59,7 +59,7 @@ describe('GoogleGenerativeAIStream', () => {
       // tool call
       'id: chat_1\n',
       'event: tool_calls\n',
-      `data: [{"function":{"arguments":"{\\"arg1\\":\\"value1\\"}","name":"testFunction"},"id":"testFunction_0","index":0,"type":"function"}]\n\n`,
+      `data: [{"function":{"arguments":"{\\"arg1\\":\\"value1\\"}","name":"testFunction"},"id":"testFunction_0_abcd1234","index":0,"type":"function"}]\n\n`,
 
       // text
       'id: chat_1\n',

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

@@ -30,7 +30,7 @@ describe('OllamaStream', () => {
       });
 
       const protocolStream = OllamaStream(mockOllamaStream);
-    
+
       const decoder = new TextDecoder();
       const chunks = [];
 
@@ -62,7 +62,7 @@ describe('OllamaStream', () => {
           'id: chat_2',
           'event: stop',
           `data: "finished"\n`,
-        ].map((line) => `${line}\n`)
+        ].map((line) => `${line}\n`),
       );
     });
 
@@ -116,7 +116,7 @@ describe('OllamaStream', () => {
     });
 
     it('tools use', async () => {
-      vi.spyOn(uuidModule, 'nanoid').mockReturnValueOnce('1');
+      vi.spyOn(uuidModule, 'nanoid').mockReturnValueOnce('1').mockReturnValueOnce('abcd1234');
 
       const mockOllamaStream = new ReadableStream<ChatResponse>({
         start(controller) {
@@ -178,7 +178,7 @@ describe('OllamaStream', () => {
         [
           'id: chat_1',
           'event: tool_calls',
-          `data: [{"function":{"arguments":"{\\"city\\":\\"杭州\\"}","name":"realtime-weather____fetchCurrentWeather"},"id":"realtime-weather____fetchCurrentWeather_0","index":0,"type":"function"}]\n`,
+          `data: [{"function":{"arguments":"{\\"city\\":\\"杭州\\"}","name":"realtime-weather____fetchCurrentWeather"},"id":"realtime-weather____fetchCurrentWeather_0_abcd1234","index":0,"type":"function"}]\n`,
           'id: chat_1',
           'event: stop',
           `data: "finished"\n`,

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

@@ -1,5 +1,6 @@
 import { CitationItem, ModelSpeed, ModelTokensUsage } from '@/types/message';
 import { safeParseJSON } from '@/utils/safeParseJSON';
+import { nanoid } from '@/utils/uuid';
 
 import { AgentRuntimeErrorType } from '../../error';
 import { parseToolCalls } from '../../helpers';
@@ -98,7 +99,7 @@ export interface StreamProtocolToolCallChunk {
 }
 
 export const generateToolCallId = (index: number, functionName?: string) =>
-  `${functionName || 'unknown_tool_call'}_${index}`;
+  `${functionName || 'unknown_tool_call'}_${index}_${nanoid()}`;
 
 const chatStreamable = async function* <T>(stream: AsyncIterable<T>) {
   for await (const response of stream) {

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

@@ -137,7 +137,7 @@ describe('VertexAIStream', () => {
   });
 
   it('tool_calls', async () => {
-    vi.spyOn(uuidModule, 'nanoid').mockReturnValueOnce('1');
+    vi.spyOn(uuidModule, 'nanoid').mockReturnValueOnce('1').mockReturnValueOnce('abcd1234');
 
     const rawChunks = [
       {
@@ -227,7 +227,7 @@ describe('VertexAIStream', () => {
       // text
       'id: chat_1\n',
       'event: tool_calls\n',
-      `data: [{"function":{"arguments":"{\\"city\\":\\"杭州\\"}","name":"realtime-weather____fetchCurrentWeather"},"id":"realtime-weather____fetchCurrentWeather_0","index":0,"type":"function"}]\n\n`,
+      `data: [{"function":{"arguments":"{\\"city\\":\\"杭州\\"}","name":"realtime-weather____fetchCurrentWeather"},"id":"realtime-weather____fetchCurrentWeather_0_abcd1234","index":0,"type":"function"}]\n\n`,
       'id: chat_1\n',
       'event: stop\n',
       'data: "STOP"\n\n',

package/src/services/chat.ts

@@ -263,6 +263,13 @@ class ChatService {
       if (modelExtendParams!.includes('reasoningEffort') && chatConfig.reasoningEffort) {
         extendParams.reasoning_effort = chatConfig.reasoningEffort;
       }
+
+      if (
+        modelExtendParams!.includes('thinkingBudget') &&
+        chatConfig.thinkingBudget !== undefined
+      ) {
+        extendParams.thinkingBudget = chatConfig.thinkingBudget;
+      }
     }
 
     return this.getChatCompletion(

package/src/types/agent/chatConfig.ts

@@ -26,7 +26,7 @@ export interface LobeAgentChatConfig {
   enableReasoningEffort?: boolean;
   reasoningBudgetToken?: number;
   reasoningEffort?: 'low' | 'medium' | 'high';
-
+  thinkingBudget?: number;
   /**
    * 禁用上下文缓存
    */

package/src/types/aiModel.ts

@@ -147,7 +147,8 @@ export type ExtendParamsType =
   | 'reasoningBudgetToken'
   | 'enableReasoning'
   | 'disableContextCaching'
-  | 'reasoningEffort';
+  | 'reasoningEffort'
+  | 'thinkingBudget';
 
 export interface AiModelSettings {
   extendParams?: ExtendParamsType[];