refactor(ai): replace manual reasoning switches with auto-inference

Remove the "Reasoning Model" and "Disable Thinking Mode" toggles from
the model config dialog. These switches were injecting `enable_thinking`
(a Qwen/DashScope-specific request parameter) into all providers,
causing 400 errors on OpenAI, Anthropic, Gemini, and other official
services.

Replace with automatic inference in buildPiModel:
- reasoning: true when the model's catalog capabilities include
  'reasoning'; falls back to a model-name heuristic for unlisted
  custom/self-hosted models (qwen3, qwq, deepseek-r, o1/o3, etc.).
- compat.thinkingFormat 'qwen': only set when reasoning is true AND
  the model is from the qwen provider or the model id contains
  "qwen"/"qwq". All other providers get compat: undefined and will
  never receive enable_thinking.

This means official reasoning models (o3, deepseek-reasoner, etc.) now
automatically get thinkingLevel:'medium', and self-hosted Qwen3
models continue to work as before — without any manual toggle.

Behavior change: models with 'reasoning' in their catalog capabilities
now default to thinkingLevel:'medium' (previously always 'off' unless
the manual switch was enabled).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

apps/cli/src/ai/llm-config.ts

@@ -25,8 +25,6 @@ export interface AIServiceConfig {
   baseUrl?: string
   maxTokens?: number
   apiFormat?: string
-  disableThinking?: boolean
-  isReasoningModel?: boolean
   customModels?: Array<{ id: string; name: string }>
   authProfile?: string
 }

apps/cli/src/http/routes/ai.ts

@@ -267,8 +267,6 @@ export function registerAiRoutes(
       baseUrl?: string
       maxTokens?: number
       apiFormat?: string
-      disableThinking?: boolean
-      isReasoningModel?: boolean
       customModels?: Array<{ id: string; name: string }>
     }
   }>('/_web/ai/llm/configs', async (request) => {
@@ -286,8 +284,6 @@ export function registerAiRoutes(
       baseUrl?: string
       maxTokens?: number
       apiFormat?: string
-      disableThinking?: boolean
-      isReasoningModel?: boolean
       customModels?: Array<{ id: string; name: string }>
     }
   }>('/_web/ai/llm/configs/:id', async (request) => {

apps/desktop/main/ai/llm/model-types.ts

@@ -33,8 +33,6 @@ export interface LLMConnectionConfig {
 }
 
 export interface LLMConnectionConfigCompat extends LLMConnectionConfig {
-  disableThinking?: boolean
-  isReasoningModel?: boolean
   customModels?: Array<{ id: string; name: string }>
 }
 

apps/desktop/main/ai/llm/types.ts

@@ -26,8 +26,6 @@ export interface AIServiceConfig {
   baseUrl?: string
   maxTokens?: number
   apiFormat?: string
-  disableThinking?: boolean
-  isReasoningModel?: boolean
   customModels?: Array<{ id: string; name: string }>
   createdAt: number
   updatedAt: number

apps/desktop/main/ai/rag/index.ts

@@ -43,8 +43,6 @@ function ensureInit(): void {
         baseUrl: cfg.baseUrl,
         maxTokens: cfg.maxTokens,
         apiFormat: cfg.apiFormat,
-        disableThinking: cfg.disableThinking,
-        isReasoningModel: cfg.isReasoningModel,
         apiKey: cfg.apiKey || undefined,
       }
     },

apps/desktop/preload/apis/ai.ts

@@ -202,8 +202,6 @@ export interface AIServiceConfigDisplay {
   baseUrl?: string
   maxTokens?: number
   apiFormat?: string
-  disableThinking?: boolean
-  isReasoningModel?: boolean
   customModels?: Array<{ id: string; name: string }>
   createdAt: number
   updatedAt: number
@@ -721,8 +719,6 @@ export const llmApi = {
     baseUrl?: string
     maxTokens?: number
     apiFormat?: string
-    disableThinking?: boolean
-    isReasoningModel?: boolean
     customModels?: Array<{ id: string; name: string }>
   }): Promise<{ success: boolean; config?: AIServiceConfigDisplay; error?: string }> => {
     return ipcRenderer.invoke('llm:addConfig', config)
@@ -741,8 +737,6 @@ export const llmApi = {
       baseUrl?: string
       maxTokens?: number
       apiFormat?: string
-      disableThinking?: boolean
-      isReasoningModel?: boolean
       customModels?: Array<{ id: string; name: string }>
     }
   ): Promise<{ success: boolean; error?: string }> => {

apps/desktop/preload/index.d.ts

@@ -559,8 +559,6 @@ interface AIServiceConfigDisplay {
   baseUrl?: string
   maxTokens?: number
   apiFormat?: string
-  disableThinking?: boolean
-  isReasoningModel?: boolean
   createdAt: number
   updatedAt: number
 }
@@ -620,8 +618,6 @@ interface LlmApi {
     baseUrl?: string
     maxTokens?: number
     apiFormat?: string
-    disableThinking?: boolean
-    isReasoningModel?: boolean
     customModels?: Array<{ id: string; name: string }>
   }) => Promise<{ success: boolean; config?: AIServiceConfigDisplay; error?: string }>
   updateConfig: (
@@ -634,8 +630,6 @@ interface LlmApi {
       baseUrl?: string
       maxTokens?: number
       apiFormat?: string
-      disableThinking?: boolean
-      isReasoningModel?: boolean
       customModels?: Array<{ id: string; name: string }>
     }
   ) => Promise<{ success: boolean; error?: string }>

packages/core/src/ai/model-types.ts

@@ -57,8 +57,6 @@ export interface LLMConnectionConfig {
 }
 
 export interface LLMConnectionConfigCompat extends LLMConnectionConfig {
-  disableThinking?: boolean
-  isReasoningModel?: boolean
   customModels?: Array<{ id: string; name: string }>
 }
 

packages/node-runtime/src/ai/__tests__/llm-builder.test.ts

@@ -87,24 +87,56 @@ describe('buildPiModel', () => {
     assert.deepEqual(model.headers, { 'X-Custom': 'value' })
   })
 
-  it('sets compat.thinkingFormat when disableThinking is true', () => {
+  it('auto-infers reasoning=true for catalog models with reasoning capability', () => {
+    // o3 is in the ChatLab catalog with capabilities: ['chat', 'reasoning', ...]
     const model = buildPiModel({
       provider: 'openai',
-      model: 'test',
+      model: 'o3',
       baseUrl: 'https://api.openai.com/v1',
-      disableThinking: true,
     })
+    assert.equal(model.reasoning, true)
+  })
+
+  it('auto-infers reasoning=false and compat=undefined for non-reasoning OpenAI model', () => {
+    // gpt-4o is not a reasoning model; no enable_thinking should ever be injected
+    const model = buildPiModel({
+      provider: 'openai',
+      model: 'gpt-4o',
+      baseUrl: 'https://api.openai.com/v1',
+    })
+    assert.equal(model.reasoning, false)
+    assert.equal(model.compat, undefined)
+  })
+
+  it('auto-infers reasoning=true and compat.thinkingFormat=qwen for official qwen reasoning model', () => {
+    // qwen3-max is in the qwen provider catalog with reasoning capability
+    const model = buildPiModel({
+      provider: 'qwen',
+      model: 'qwen3-max',
+      baseUrl: 'https://dashscope.aliyuncs.com/compatible-mode/v1',
+    })
+    assert.equal(model.reasoning, true)
     assert.deepEqual(model.compat, { thinkingFormat: 'qwen' })
   })
 
-  it('sets reasoning flag from isReasoningModel', () => {
-    const model = buildPiModel({
-      provider: 'openai',
-      model: 'o1',
-      baseUrl: 'https://api.openai.com/v1',
-      isReasoningModel: true,
-    })
+  it('auto-infers reasoning=true and compat.thinkingFormat=qwen for custom self-hosted qwen3 model', () => {
+    // Custom model not in catalog — falls back to name heuristic
+    const model = buildPiModel(
+      { provider: 'openai-compatible', model: 'qwen3:8b', baseUrl: 'http://localhost:11434/v1' },
+      { findModelFn: () => null }
+    )
     assert.equal(model.reasoning, true)
+    assert.deepEqual(model.compat, { thinkingFormat: 'qwen' })
+  })
+
+  it('auto-infers reasoning=false and compat=undefined for non-reasoning Anthropic model', () => {
+    const model = buildPiModel({
+      provider: 'openai-compatible',
+      model: 'claude-3-5-sonnet',
+      baseUrl: 'https://api.anthropic.com/v1',
+    })
+    assert.equal(model.reasoning, false)
+    assert.equal(model.compat, undefined)
   })
 
   it('uses custom findModelFn for context window', () => {

packages/node-runtime/src/ai/llm-builder.ts

@@ -15,8 +15,6 @@ export interface PiModelConfig {
   baseUrl?: string
   maxTokens?: number
   apiFormat?: string
-  disableThinking?: boolean
-  isReasoningModel?: boolean
 }
 
 export interface BuildPiModelOptions {
@@ -65,6 +63,37 @@ const BUILTIN_PROVIDER_API: Record<string, PiApi> = {
   anthropic: 'anthropic-messages',
 }
 
+/**
+ * Infer whether a model should have reasoning enabled and which
+ * thinking format to use, based on the model catalog and provider.
+ *
+ * - reasoning: true when the catalog marks the model as 'reasoning' capable,
+ *   or (for custom/unlisted models) when the model id matches a known pattern.
+ * - compat.thinkingFormat 'qwen': only for Qwen-family endpoints (official
+ *   DashScope provider or model id containing "qwen"/"qwq"). Other providers
+ *   must never receive `enable_thinking` in the request body.
+ */
+function inferReasoning(
+  provider: string,
+  modelId: string,
+  modelDef: ModelDefinition | null,
+): { reasoning: boolean; compat: PiModel<PiApi>['compat'] } {
+  // Catalog is the authoritative source; fall back to heuristic for unlisted
+  // custom/self-hosted models where modelDef is null.
+  const reasoning = modelDef
+    ? modelDef.capabilities.includes('reasoning')
+    : /qwen3|qwq|deepseek-r|r1\b|o1\b|o3\b|thinking|reasoning/i.test(modelId)
+
+  if (!reasoning) return { reasoning: false, compat: undefined }
+
+  // Qwen-family: official DashScope provider id OR model name containing qwen/qwq.
+  const isQwen = provider === 'qwen' || /qwen|qwq/i.test(modelId)
+  return {
+    reasoning: true,
+    compat: isQwen ? { thinkingFormat: 'qwen' } : undefined,
+  }
+}
+
 export function buildPiModel(config: PiModelConfig, options?: BuildPiModelOptions): PiModel<PiApi> {
   const providerDef = BUILTIN_PROVIDERS.find((p) => p.id === config.provider)
   const baseUrl = config.baseUrl || providerDef?.defaultBaseUrl || ''
@@ -111,6 +140,8 @@ export function buildPiModel(config: PiModelConfig, options?: BuildPiModelOption
       ? normalizeOpenAICompatibleBaseUrl(baseUrl)
       : baseUrl
 
+  const { reasoning, compat } = inferReasoning(config.provider, modelId, modelDef)
+
   return {
     id: modelId,
     name: modelId,
@@ -118,11 +149,11 @@ export function buildPiModel(config: PiModelConfig, options?: BuildPiModelOption
     provider: config.provider,
     baseUrl: resolvedBaseUrl,
     headers: options?.headers,
-    reasoning: config.isReasoningModel ?? false,
+    reasoning,
     input: ['text'],
     cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
     contextWindow,
     maxTokens: config.maxTokens ?? 4096,
-    compat: config.disableThinking ? { thinkingFormat: 'qwen' } : undefined,
+    compat,
   }
 }

packages/node-runtime/src/ai/llm-config-store.ts

@@ -17,8 +17,6 @@ export interface AIServiceConfig {
   baseUrl?: string
   maxTokens?: number
   apiFormat?: string
-  disableThinking?: boolean
-  isReasoningModel?: boolean
   customModels?: Array<{ id: string; name: string }>
   createdAt: number
   updatedAt: number

src/components/common/Settings/AI/AIModelEditModal.vue

@@ -526,35 +526,6 @@ function closeModal() {
               </div>
             </template>
 
-            <!-- ===== 通用：推理模型选项 ===== -->
-            <template v-if="formData.provider || !isPresetMode">
-              <div class="flex items-center justify-between rounded-lg bg-gray-50 p-3 dark:bg-gray-800">
-                <div>
-                  <p class="text-sm font-medium text-gray-700 dark:text-gray-300">
-                    {{ t('settings.aiConfig.modal.isReasoningModel') }}
-                  </p>
-                  <p class="text-xs text-gray-500 dark:text-gray-400">
-                    {{ t('settings.aiConfig.modal.isReasoningModelDesc') }}
-                  </p>
-                </div>
-                <USwitch v-model="formData.isReasoningModel" />
-              </div>
-
-              <div
-                v-if="formData.isReasoningModel"
-                class="flex items-center justify-between rounded-lg bg-gray-50 p-3 dark:bg-gray-800"
-              >
-                <div>
-                  <p class="text-sm font-medium text-gray-700 dark:text-gray-300">
-                    {{ t('settings.aiConfig.modal.disableThinking') }}
-                  </p>
-                  <p class="text-xs text-gray-500 dark:text-gray-400">
-                    {{ t('settings.aiConfig.modal.disableThinkingDesc') }}
-                  </p>
-                </div>
-                <USwitch v-model="formData.disableThinking" />
-              </div>
-            </template>
           </div>
         </div>
 

src/components/common/Settings/AI/useAIConfigForm.ts

@@ -15,8 +15,6 @@ export interface AIServiceConfig {
   model?: string
   baseUrl?: string
   apiFormat?: string
-  disableThinking?: boolean
-  isReasoningModel?: boolean
   customModels?: Array<{ id: string; name: string }>
   createdAt: number
   updatedAt: number
@@ -96,8 +94,6 @@ export function useAIConfigForm(props: {
     baseUrl: '',
     apiFormat: API_FORMAT_DEFAULT,
     customName: '',
-    disableThinking: true,
-    isReasoningModel: false,
   })
 
   const validationResult = ref<'idle' | 'valid' | 'invalid'>('idle')
@@ -261,8 +257,6 @@ export function useAIConfigForm(props: {
       baseUrl: defaultProviderDef?.defaultBaseUrl || '',
       apiFormat: API_FORMAT_DEFAULT,
       customName: '',
-      disableThinking: true,
-      isReasoningModel: false,
     }
     validationResult.value = 'idle'
     validationMessage.value = ''
@@ -292,8 +286,6 @@ export function useAIConfigForm(props: {
       baseUrl: config.baseUrl || providerDef?.defaultBaseUrl || '',
       apiFormat: config.apiFormat || API_FORMAT_DEFAULT,
       customName: config.name || '',
-      disableThinking: config.disableThinking ?? true,
-      isReasoningModel: config.isReasoningModel ?? false,
     }
     validationResult.value = 'idle'
     validationMessage.value = ''
@@ -637,7 +629,6 @@ export function useAIConfigForm(props: {
       }
       const finalName = formData.value.customName.trim() || generateName()
 
-      const isReasoning = formData.value.isReasoningModel
       const persistCustomModels =
         isCompatMode.value && compatModels.value.length > 0
           ? compatModels.value.map((m) => ({ id: m.id, name: m.name }))
@@ -651,8 +642,6 @@ export function useAIConfigForm(props: {
           model: formData.value.model.trim() || undefined,
           baseUrl: formData.value.baseUrl.trim() || undefined,
           apiFormat: savedApiFormat,
-          disableThinking: isReasoning ? formData.value.disableThinking : undefined,
-          isReasoningModel: isReasoning || undefined,
           customModels: persistCustomModels,
         })
 
@@ -670,8 +659,6 @@ export function useAIConfigForm(props: {
           model: formData.value.model.trim() || undefined,
           baseUrl: formData.value.baseUrl.trim() || undefined,
           apiFormat: savedApiFormat,
-          disableThinking: isReasoning ? formData.value.disableThinking : undefined,
-          isReasoningModel: isReasoning || undefined,
           customModels: persistCustomModels,
         }
 

src/i18n/locales/en-US/settings.json

@@ -154,10 +154,6 @@
       "addAllModels": "Add all",
       "searchModels": "Search models...",
       "apiEndpointOverrideHint": "Default URL is prefilled. Replace if needed",
-      "disableThinking": "Disable Thinking Mode",
-      "disableThinkingDesc": "For models like Qwen3, DeepSeek-R1. Uses standard tool calling format when disabled.",
-      "isReasoningModel": "Reasoning Model",
-      "isReasoningModelDesc": "Extracts thinking process and disables tool calling (e.g. DeepSeek-R1, QwQ)",
       "advancedOptions": "Advanced Options",
       "customService": "Custom Service",
       "unnamedConfig": "Unnamed Configuration",

src/i18n/locales/ja-JP/settings.json

@@ -154,10 +154,6 @@
       "addAllModels": "すべて追加",
       "searchModels": "モデルを検索...",
       "apiEndpointOverrideHint": "デフォルトURLは入力済みです。必要に応じて変更してください",
-      "disableThinking": "思考モードをオフ",
-      "disableThinkingDesc": "Qwen3 や DeepSeek-R1 などのモデル向け。オフにすると標準のツール実行形式を使います",
-      "isReasoningModel": "推論モデル",
-      "isReasoningModelDesc": "有効にすると推論過程を抽出し、ツール実行を無効にします（DeepSeek-R1、QwQ など）",
       "advancedOptions": "詳細設定",
       "customService": "カスタムサービス",
       "unnamedConfig": "名称未設定",

src/i18n/locales/zh-CN/settings.json

@@ -154,10 +154,6 @@
       "addAllModels": "全部添加",
       "searchModels": "搜索模型...",
       "apiEndpointOverrideHint": "默认地址已填入，可按需改为完整 URL",
-      "disableThinking": "禁用思考模式",
-      "disableThinkingDesc": "针对 Qwen3、DeepSeek-R1 等模型，禁用后使用标准工具调用格式",
-      "isReasoningModel": "推理模型",
-      "isReasoningModelDesc": "启用后将提取思考过程并禁用工具调用（如 DeepSeek-R1、QwQ 等）",
       "advancedOptions": "高级选项",
       "customService": "自定义服务",
       "unnamedConfig": "未命名配置",

src/i18n/locales/zh-TW/settings.json

@@ -154,10 +154,6 @@
       "addAllModels": "全部添加",
       "searchModels": "搜尋模型...",
       "apiEndpointOverrideHint": "預設地址已填入，可按需改為完整 URL",
-      "disableThinking": "關閉思考模式",
-      "disableThinkingDesc": "針對 Qwen3、DeepSeek-R1 等模型，停用後使用標準工具呼叫格式",
-      "isReasoningModel": "推理模型",
-      "isReasoningModelDesc": "啟用後將提取思考過程並停用工具呼叫（如 DeepSeek-R1、QwQ 等）",
       "advancedOptions": "進階設定",
       "customService": "自訂服務",
       "unnamedConfig": "未命名設定",