feat: aihubmix support

packages/aiCore/src/core/options/xai.ts

@@ -1,7 +1,7 @@
 // copy from @ai-sdk/xai/xai-chat-options.ts
 // 如果@ai-sdk/xai暴露出了xaiProviderOptions就删除这个文件
 
-import { z } from 'zod'
+import * as z from 'zod/v4'
 
 const webSourceSchema = z.object({
   type: z.literal('web'),
@@ -25,7 +25,7 @@ const newsSourceSchema = z.object({
 
 const rssSourceSchema = z.object({
   type: z.literal('rss'),
-  links: z.array(z.string().url()).max(1) // currently only supports one RSS link
+  links: z.array(z.url()).max(1) // currently only supports one RSS link
 })
 
 const searchSourceSchema = z.discriminatedUnion('type', [

src/renderer/src/aiCore/index_new.ts

@@ -30,14 +30,11 @@ import { AiSdkMiddlewareConfig, buildAiSdkMiddlewares } from './middleware/aisdk
 import { CompletionsResult } from './middleware/schemas'
 import reasoningTimePlugin from './plugins/reasoningTimePlugin'
 import { getAiSdkProviderId } from './provider/factory'
+import { getProviderByModel } from '@renderer/services/AssistantService'
+import { createAihubmixProvider } from './provider/aihubmix'
 
-/**
- * 将 Provider 配置转换为新 AI SDK 格式
- */
-function providerToAiSdkConfig(provider: Provider): {
-  providerId: ProviderId | 'openai-compatible'
-  options: ProviderSettingsMap[keyof ProviderSettingsMap]
-} {
+function getActualProvider(model: Model): Provider {
+  const provider = getProviderByModel(model)
   // 如果是 vertexai 类型且没有 googleCredentials，转换为 VertexProvider
   let actualProvider = cloneDeep(provider)
   if (provider.type === 'vertexai' && !isVertexProvider(provider)) {
@@ -47,18 +44,25 @@ function providerToAiSdkConfig(provider: Provider): {
     actualProvider = createVertexProvider(provider)
   }
 
-  if (
-    actualProvider.type === 'openai' ||
-    actualProvider.type === 'anthropic' ||
-    actualProvider.type === 'openai-response'
-  ) {
-    actualProvider.apiHost = formatApiHost(actualProvider.apiHost)
+  if (provider.id === 'aihubmix') {
+    actualProvider = createAihubmixProvider(model, actualProvider)
   }
 
   if (actualProvider.type === 'gemini') {
     actualProvider.apiHost = formatApiHost(actualProvider.apiHost, 'v1beta')
+  } else {
+    actualProvider.apiHost = formatApiHost(actualProvider.apiHost)
   }
+  return actualProvider
+}
 
+/**
+ * 将 Provider 配置转换为新 AI SDK 格式
+ */
+function providerToAiSdkConfig(actualProvider: Provider): {
+  providerId: ProviderId | 'openai-compatible'
+  options: ProviderSettingsMap[keyof ProviderSettingsMap]
+} {
   const aiSdkProviderId = getAiSdkProviderId(actualProvider)
 
   // 如果provider是openai，则使用strict模式并且默认responses api
@@ -126,14 +130,18 @@ function isModernSdkSupported(provider: Provider, model?: Model): boolean {
 export default class ModernAiProvider {
   private legacyProvider: LegacyAiProvider
   private config: ReturnType<typeof providerToAiSdkConfig>
+  private actualProvider: Provider
 
-  constructor(provider: Provider) {
-    this.legacyProvider = new LegacyAiProvider(provider)
+  constructor(model: Model) {
+    this.actualProvider = getActualProvider(model)
+    this.legacyProvider = new LegacyAiProvider(this.actualProvider)
 
     // 只保存配置，不预先创建executor
-    this.config = providerToAiSdkConfig(provider)
+    this.config = providerToAiSdkConfig(this.actualProvider)
+  }
 
-    console.log('[Modern AI Provider] Creating executor with MCP Prompt plugin enabled')
+  public getActualProvider() {
+    return this.actualProvider
   }
 
   /**

src/renderer/src/aiCore/provider/aihubmix.ts

@@ -0,0 +1,55 @@
+import { ProviderId } from '@cherrystudio/ai-core/types'
+import { isOpenAILLMModel } from '@renderer/config/models'
+import { Model, Provider } from '@renderer/types'
+
+export function getAiSdkProviderIdForAihubmix(model: Model): ProviderId | 'openai-compatible' {
+  const id = model.id.toLowerCase()
+
+  if (id.startsWith('claude')) {
+    return 'anthropic'
+  }
+
+  if ((id.startsWith('gemini') || id.startsWith('imagen')) && !id.endsWith('-nothink') && !id.endsWith('-search')) {
+    return 'google'
+  }
+
+  if (isOpenAILLMModel(model)) {
+    return 'openai'
+  }
+
+  return 'openai-compatible'
+}
+
+export function createAihubmixProvider(model: Model, provider: Provider): Provider {
+  const providerId = getAiSdkProviderIdForAihubmix(model)
+  provider = {
+    ...provider,
+    extra_headers: {
+      ...provider.extra_headers,
+      'APP-Code': 'MLTG2087'
+    }
+  }
+  if (providerId === 'google') {
+    return {
+      ...provider,
+      type: 'gemini',
+      apiHost: 'https://aihubmix.com/gemini'
+    }
+  }
+
+  if (providerId === 'openai') {
+    return {
+      ...provider,
+      type: 'openai'
+    }
+  }
+
+  if (providerId === 'anthropic') {
+    return {
+      ...provider,
+      type: 'anthropic'
+    }
+  }
+
+  return provider
+}

src/renderer/src/aiCore/transformParameters.ts

@@ -27,7 +27,7 @@ import {
   isWebSearchModel
 } from '@renderer/config/models'
 import { getAssistantSettings, getDefaultModel } from '@renderer/services/AssistantService'
-import type { Assistant, MCPTool, Message, Model } from '@renderer/types'
+import type { Assistant, MCPTool, Message, Model, Provider } from '@renderer/types'
 import { FileTypes } from '@renderer/types'
 import { FileMessageBlock, ImageMessageBlock, ThinkingMessageBlock } from '@renderer/types/newMessage'
 import {
@@ -241,6 +241,7 @@ export async function convertMessagesToSdkMessages(
 export async function buildStreamTextParams(
   sdkMessages: StreamTextParams['messages'],
   assistant: Assistant,
+  provider: Provider,
   options: {
     mcpTools?: MCPTool[]
     enableTools?: boolean
@@ -285,14 +286,8 @@ export async function buildStreamTextParams(
     enableToolUse: enableTools
   })
 
-  // Add web search tools if enabled
-  // if (enableWebSearch) {
-  //   const webSearchTools = getWebSearchTools(model)
-  //   tools = { ...tools, ...webSearchTools }
-  // }
-
   // 构建真正的 providerOptions
-  const providerOptions = buildProviderOptions(assistant, model, {
+  const providerOptions = buildProviderOptions(assistant, model, provider, {
     enableReasoning,
     enableWebSearch,
     enableGenerateImage
@@ -321,11 +316,12 @@ export async function buildStreamTextParams(
 export async function buildGenerateTextParams(
   messages: ModelMessage[],
   assistant: Assistant,
+  provider: Provider,
   options: {
     mcpTools?: MCPTool[]
     enableTools?: boolean
   } = {}
 ): Promise<any> {
   // 复用流式参数的构建逻辑
-  return await buildStreamTextParams(messages, assistant, options)
+  return await buildStreamTextParams(messages, assistant, provider, options)
 }

src/renderer/src/aiCore/utils/options.ts

@@ -1,5 +1,5 @@
-import { getProviderByModel } from '@renderer/services/AssistantService'
-import { Assistant, Model } from '@renderer/types'
+import { getDefaultModel, getProviderByModel } from '@renderer/services/AssistantService'
+import { Assistant, Model, Provider } from '@renderer/types'
 
 import { getAiSdkProviderId } from '../provider/factory'
 import {
@@ -7,8 +7,10 @@ import {
   getCustomParameters,
   getGeminiReasoningParams,
   getOpenAIReasoningParams,
-  getReasoningEffort
+  getReasoningEffort,
+  getXAIReasoningParams
 } from './reasoning'
+import { getWebSearchParams } from './websearch'
 
 /**
  * 构建 AI SDK 的 providerOptions
@@ -18,25 +20,22 @@ import {
 export function buildProviderOptions(
   assistant: Assistant,
   model: Model,
+  actualProvider: Provider,
   capabilities: {
     enableReasoning: boolean
     enableWebSearch: boolean
     enableGenerateImage: boolean
   }
 ): Record<string, any> {
-  const provider = getProviderByModel(model)
-  const providerId = getAiSdkProviderId(provider)
+  const providerId = getAiSdkProviderId(actualProvider)
 
   // 构建 provider 特定的选项
   let providerSpecificOptions: Record<string, any> = {}
 
-  console.log('buildProviderOptions', providerId)
-  console.log('buildProviderOptions', provider)
-
   // 根据 provider 类型分离构建逻辑
-  switch (provider.type) {
-    case 'openai-response':
-    case 'azure-openai':
+  switch (providerId) {
+    case 'openai':
+    case 'azure':
       providerSpecificOptions = buildOpenAIProviderOptions(assistant, model, capabilities)
       break
 
@@ -44,11 +43,15 @@ export function buildProviderOptions(
       providerSpecificOptions = buildAnthropicProviderOptions(assistant, model, capabilities)
       break
 
-    case 'gemini':
-    case 'vertexai':
+    case 'google':
+    case 'google-vertex':
       providerSpecificOptions = buildGeminiProviderOptions(assistant, model, capabilities)
       break
 
+    case 'xai':
+      providerSpecificOptions = buildXAIProviderOptions(assistant, model, capabilities)
+      break
+
     default:
       // 对于其他 provider，使用通用的构建逻辑
       providerSpecificOptions = buildGenericProviderOptions(assistant, model, capabilities)
@@ -79,7 +82,7 @@ function buildOpenAIProviderOptions(
     enableGenerateImage: boolean
   }
 ): Record<string, any> {
-  const { enableReasoning, enableWebSearch, enableGenerateImage } = capabilities
+  const { enableReasoning } = capabilities
   let providerOptions: Record<string, any> = {}
 
   // OpenAI 推理参数
@@ -91,15 +94,6 @@ function buildOpenAIProviderOptions(
     }
   }
 
-  // Web 搜索和图像生成暂时使用通用格式
-  if (enableWebSearch) {
-    providerOptions.webSearch = { enabled: true }
-  }
-
-  if (enableGenerateImage) {
-    providerOptions.generateImage = { enabled: true }
-  }
-
   return providerOptions
 }
 
@@ -115,7 +109,7 @@ function buildAnthropicProviderOptions(
     enableGenerateImage: boolean
   }
 ): Record<string, any> {
-  const { enableReasoning, enableWebSearch, enableGenerateImage } = capabilities
+  const { enableReasoning } = capabilities
   let providerOptions: Record<string, any> = {}
 
   // Anthropic 推理参数
@@ -127,14 +121,6 @@ function buildAnthropicProviderOptions(
     }
   }
 
-  if (enableWebSearch) {
-    providerOptions.webSearch = { enabled: true }
-  }
-
-  if (enableGenerateImage) {
-    providerOptions.generateImage = { enabled: true }
-  }
-
   return providerOptions
 }
 
@@ -150,21 +136,39 @@ function buildGeminiProviderOptions(
     enableGenerateImage: boolean
   }
 ): Record<string, any> {
-  const { enableReasoning, enableWebSearch, enableGenerateImage } = capabilities
-  const providerOptions: Record<string, any> = {}
+  const { enableReasoning } = capabilities
+  let providerOptions: Record<string, any> = {}
 
   // Gemini 推理参数
   if (enableReasoning) {
     const reasoningParams = getGeminiReasoningParams(assistant, model)
-    Object.assign(providerOptions, reasoningParams)
+    providerOptions = {
+      ...providerOptions,
+      ...reasoningParams
+    }
   }
 
-  if (enableWebSearch) {
-    providerOptions.webSearch = { enabled: true }
-  }
+  return providerOptions
+}
 
-  if (enableGenerateImage) {
-    providerOptions.generateImage = { enabled: true }
+function buildXAIProviderOptions(
+  assistant: Assistant,
+  model: Model,
+  capabilities: {
+    enableReasoning: boolean
+    enableWebSearch: boolean
+    enableGenerateImage: boolean
+  }
+): Record<string, any> {
+  const { enableReasoning } = capabilities
+  let providerOptions: Record<string, any> = {}
+
+  if (enableReasoning) {
+    const reasoningParams = getXAIReasoningParams(assistant, model)
+    providerOptions = {
+      ...providerOptions,
+      ...reasoningParams
+    }
   }
 
   return providerOptions
@@ -182,7 +186,7 @@ function buildGenericProviderOptions(
     enableGenerateImage: boolean
   }
 ): Record<string, any> {
-  const { enableWebSearch, enableGenerateImage } = capabilities
+  const { enableWebSearch } = capabilities
   let providerOptions: Record<string, any> = {}
 
   const reasoningParams = getReasoningEffort(assistant, model)
@@ -192,11 +196,11 @@ function buildGenericProviderOptions(
   }
 
   if (enableWebSearch) {
-    providerOptions.webSearch = { enabled: true }
-  }
-
-  if (enableGenerateImage) {
-    providerOptions.generateImage = { enabled: true }
+    const webSearchParams = getWebSearchParams(model)
+    providerOptions = {
+      ...providerOptions,
+      ...webSearchParams
+    }
   }
 
   return providerOptions

src/renderer/src/aiCore/utils/reasoning.ts

@@ -314,6 +314,18 @@ export function getGeminiReasoningParams(assistant: Assistant, model: Model): Re
   return {}
 }
 
+export function getXAIReasoningParams(assistant: Assistant, model: Model): Record<string, any> {
+  if (!isSupportedReasoningEffortGrokModel(model)) {
+    return {}
+  }
+
+  const { reasoning_effort: reasoningEffort } = getAssistantSettings(assistant)
+
+  return {
+    reasoningEffort
+  }
+}
+
 /**
  * 获取自定义参数
  * 从 assistant 设置中提取自定义参数

src/renderer/src/aiCore/utils/websearch.ts

@@ -1,37 +1,31 @@
-// import { isWebSearchModel } from '@renderer/config/models'
-// import { Model } from '@renderer/types'
-// // import {} from '@cherrystudio/ai-core'
+import { isOpenAIWebSearchChatCompletionOnlyModel } from '@renderer/config/models'
+import { WEB_SEARCH_PROMPT_FOR_OPENROUTER } from '@renderer/config/prompts'
+import { Model } from '@renderer/types'
 
-// // The tool name for Gemini search can be arbitrary, but let's use a descriptive one.
-// const GEMINI_SEARCH_TOOL_NAME = 'google_search'
+export function getWebSearchParams(model: Model): Record<string, any> {
+  if (model.provider === 'hunyuan') {
+    return { enable_enhancement: true, citation: true, search_info: true }
+  }
 
-// export function getWebSearchTools(model: Model): Record<string, any> {
-//   if (!isWebSearchModel(model)) {
-//     return {}
-//   }
+  if (model.provider === 'dashscope') {
+    return {
+      enable_search: true,
+      search_options: {
+        forced_search: true
+      }
+    }
+  }
 
-//   // Use provider from model if available, otherwise fallback to parsing model id.
-//   const provider = model.provider || model.id.split('/')[0]
+  if (isOpenAIWebSearchChatCompletionOnlyModel(model)) {
+    return {
+      web_search_options: {}
+    }
+  }
 
-//   switch (provider) {
-//     case 'anthropic':
-//       return {
-//         web_search: {
-//           type: 'web_search_20250305',
-//           name: 'web_search',
-//           max_uses: 5
-//         }
-//       }
-//     case 'google':
-//     case 'gemini':
-//       return {
-//         [GEMINI_SEARCH_TOOL_NAME]: {
-//           googleSearch: {}
-//         }
-//       }
-//     default:
-//       // For OpenAI and others, web search is often a parameter, not a tool.
-//       // The logic is handled in `buildProviderOptions`.
-//       return {}
-//   }
-// }
+  if (model.provider === 'openrouter') {
+    return {
+      plugins: [{ id: 'web', search_prompts: WEB_SEARCH_PROMPT_FOR_OPENROUTER }]
+    }
+  }
+  return {}
+}

src/renderer/src/config/models.ts

@@ -2776,8 +2776,6 @@ export function getOpenAIWebSearchParams(model: Model, isEnableWebSearch?: boole
   return {
     tools: webSearchTools
   }
-
-  return {}
 }
 
 export function isGemmaModel(model?: Model): boolean {
@@ -2839,7 +2837,7 @@ export const THINKING_TOKEN_MAP: Record<string, { min: number; max: number }> =
   'gemini-.*-pro.*$': { min: 128, max: 32768 },
 
   // Qwen models
-  'qwen-plus-.*$': { min: 0, max: 38912 },
+  'qwen-plus(-.*)?$': { min: 0, max: 38912 },
   'qwen-turbo-.*$': { min: 0, max: 38912 },
   'qwen3-0\\.6b$': { min: 0, max: 30720 },
   'qwen3-1\\.7b$': { min: 0, max: 30720 },

src/renderer/src/services/ApiService.ts

@@ -300,8 +300,8 @@ export async function fetchChatCompletion({
   }
   onChunkReceived: (chunk: Chunk) => void
 }) {
-  const provider = getAssistantProvider(assistant)
-  const AI = new AiProviderNew(provider)
+  const AI = new AiProviderNew(assistant.model || getDefaultModel())
+  const provider = AI.getActualProvider()
 
   const mcpTools = await fetchMcpTools(assistant)
 
@@ -310,7 +310,7 @@ export async function fetchChatCompletion({
     params: aiSdkParams,
     modelId,
     capabilities
-  } = await buildStreamTextParams(messages, assistant, {
+  } = await buildStreamTextParams(messages, assistant, provider, {
     mcpTools: mcpTools,
     enableTools: isEnabledToolUse(assistant),
     requestOptions: options

src/renderer/src/types/index.ts

@@ -176,14 +176,7 @@ export interface VertexProvider extends BaseProvider {
   location: string
 }
 
-export type ProviderType =
-  | 'openai'
-  | 'openai-response'
-  | 'anthropic'
-  | 'gemini'
-  | 'qwenlm'
-  | 'azure-openai'
-  | 'vertexai'
+export type ProviderType = 'openai' | 'openai-response' | 'anthropic' | 'gemini' | 'azure-openai' | 'vertexai'
 
 export type ModelType = 'text' | 'vision' | 'embedding' | 'reasoning' | 'function_calling' | 'web_search'