Merge branch 'main' into fix/check-api-key

* docs(providerConfig): improve jsdoc for formatProviderApiHost function

* refactor(aiCore): improve provider handling with adaptProvider function

Introduce adaptProvider to centralize provider transformations and replace direct usage of handleSpecialProviders and formatProviderApiHost. This improves maintainability and provides consistent behavior across all provider usage scenarios.

* refactor(ProviderSettings): simplify api host formatting logic by using adaptProvider

Replace multiple format functions with a single adaptProvider utility to centralize host formatting logic and improve maintainability

* feat(api): add withoutTrailingSharp utility and update formatApiHost

add utility function to remove trailing # from URLs and update formatApiHost to use it
add comprehensive tests for new functionality

* feat(ProviderSetting): add help tooltip for api url selector

Add HelpTooltip component next to host selector to provide additional guidance about API URL configuration

CLAUDE.md

@@ -12,7 +12,15 @@ This file provides guidance to AI coding assistants when working with code in th
 - **Always propose before executing**: Before making any changes, clearly explain your planned approach and wait for explicit user approval to ensure alignment and prevent unwanted modifications.
 - **Lint, test, and format before completion**: Coding tasks are only complete after running `yarn lint`, `yarn test`, and `yarn format` successfully.
 - **Write conventional commits**: Commit small, focused changes using Conventional Commit messages (e.g., `feat:`, `fix:`, `refactor:`, `docs:`).
-- **Follow PR template**: When submitting pull requests, follow the template in `.github/pull_request_template.md` to ensure complete context and documentation.
+
+## Pull Request Workflow (CRITICAL)
+
+When creating a Pull Request, you MUST:
+
+1. **Read the PR template first**: Always read `.github/pull_request_template.md` before creating the PR
+2. **Follow ALL template sections**: Structure the `--body` parameter to include every section from the template
+3. **Never skip sections**: Include all sections even if marking them as N/A or "None"
+4. **Use proper formatting**: Match the template's markdown structure exactly (headings, checkboxes, code blocks)
 
 ## Development Commands
 

electron-builder.yml

@@ -134,9 +134,9 @@ artifactBuildCompleted: scripts/artifact-build-completed.js
 releaseInfo:
   releaseNotes: |
     <!--LANG:en-->
-    A New Era of Intelligence with Cherry Studio 1.7.0
+    A New Era of Intelligence with Cherry Studio 1.7.1
 
-    Today we're releasing Cherry Studio 1.7.0 — our most ambitious update yet, introducing Agent: autonomous AI that thinks, plans, and acts.
+    Today we're releasing Cherry Studio 1.7.1 — our most ambitious update yet, introducing Agent: autonomous AI that thinks, plans, and acts.
 
     For years, AI assistants have been reactive — waiting for your commands, responding to your questions. With Agent, we're changing that. Now, AI can truly work alongside you: understanding complex goals, breaking them into steps, and executing them independently.
 
@@ -187,9 +187,9 @@ releaseInfo:
     The Agent Era is here. We can't wait to see what you'll create.
 
     <!--LANG:zh-CN-->
-    Cherry Studio 1.7.0：开启智能新纪元
+    Cherry Studio 1.7.1：开启智能新纪元
 
-    今天，我们正式发布 Cherry Studio 1.7.0 —— 迄今最具雄心的版本，带来全新的 Agent：能够自主思考、规划和行动的 AI。
+    今天，我们正式发布 Cherry Studio 1.7.1 —— 迄今最具雄心的版本，带来全新的 Agent：能够自主思考、规划和行动的 AI。
 
     多年来，AI 助手一直是被动的——等待你的指令，回应你的问题。Agent 改变了这一切。现在，AI 能够真正与你并肩工作：理解复杂目标，将其拆解为步骤，并独立执行。
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "CherryStudio",
-  "version": "1.7.0",
+  "version": "1.7.1",
   "private": true,
   "description": "A powerful AI assistant for producer.",
   "main": "./out/main/index.js",

packages/ai-sdk-provider/src/cherryin-provider.ts

@@ -69,6 +69,7 @@ export interface CherryInProviderSettings {
   headers?: HeadersInput
   /**
    * Optional endpoint type to distinguish different endpoint behaviors.
+   * "image-generation" is also openai endpoint, but specifically for image generation.
    */
   endpointType?: 'openai' | 'openai-response' | 'anthropic' | 'gemini' | 'image-generation' | 'jina-rerank'
 }

src/main/services/agents/services/claudecode/index.ts

@@ -1,6 +1,7 @@
 // src/main/services/agents/services/claudecode/index.ts
 import { EventEmitter } from 'node:events'
 import { createRequire } from 'node:module'
+import path from 'node:path'
 
 import type {
   CanUseTool,
@@ -121,7 +122,11 @@ class ClaudeCodeService implements AgentServiceInterface {
       // TODO: support set small model in UI
       ANTHROPIC_DEFAULT_HAIKU_MODEL: modelInfo.modelId,
       ELECTRON_RUN_AS_NODE: '1',
-      ELECTRON_NO_ATTACH_CONSOLE: '1'
+      ELECTRON_NO_ATTACH_CONSOLE: '1',
+      // Set CLAUDE_CONFIG_DIR to app's userData directory to avoid path encoding issues
+      // on Windows when the username contains non-ASCII characters (e.g., Chinese characters)
+      // This prevents the SDK from using the user's home directory which may have encoding problems
+      CLAUDE_CONFIG_DIR: path.join(app.getPath('userData'), '.claude')
     }
 
     const errorChunks: string[] = []

src/renderer/src/aiCore/index_new.ts

@@ -27,6 +27,7 @@ import { buildAiSdkMiddlewares } from './middleware/AiSdkMiddlewareBuilder'
 import { buildPlugins } from './plugins/PluginBuilder'
 import { createAiSdkProvider } from './provider/factory'
 import {
+  adaptProvider,
   getActualProvider,
   isModernSdkSupported,
   prepareSpecialProviderConfig,
@@ -64,12 +65,11 @@ export default class ModernAiProvider {
    *    - URL will be automatically formatted via `formatProviderApiHost`, adding version suffixes like `/v1`
    *
    * 2. When called with `(model, provider)`:
-   *    - **Directly uses the provided provider WITHOUT going through `getActualProvider`**
-   *    - **URL will NOT be automatically formatted, `/v1` suffix will NOT be added**
-   *    - This is legacy behavior kept for backward compatibility
+   *    - The provided provider will be adapted via `adaptProvider`
+   *    - URL formatting behavior depends on the adapted result
    *
    * 3. When called with `(provider)`:
-   *    - Directly uses the provider without requiring a model
+   *    - The provider will be adapted via `adaptProvider`
    *    - Used for operations that don't need a model (e.g., fetchModels)
    *
    * @example
@@ -77,7 +77,7 @@ export default class ModernAiProvider {
    * // Recommended: Auto-format URL
    * const ai = new ModernAiProvider(model)
    *
-   * // Not recommended: Skip URL formatting (only for special cases)
+   * // Provider will be adapted
    * const ai = new ModernAiProvider(model, customProvider)
    *
    * // For operations that don't need a model
@@ -91,12 +91,12 @@ export default class ModernAiProvider {
     if (this.isModel(modelOrProvider)) {
       // 传入的是 Model
       this.model = modelOrProvider
-      this.actualProvider = provider || getActualProvider(modelOrProvider)
+      this.actualProvider = provider ? adaptProvider({ provider }) : getActualProvider(modelOrProvider)
       // 只保存配置，不预先创建executor
       this.config = providerToAiSdkConfig(this.actualProvider, modelOrProvider)
     } else {
       // 传入的是 Provider
-      this.actualProvider = modelOrProvider
+      this.actualProvider = adaptProvider({ provider: modelOrProvider })
       // model为可选，某些操作（如fetchModels）不需要model
     }
 
@@ -120,9 +120,12 @@ export default class ModernAiProvider {
       throw new Error('Model is required for completions. Please use constructor with model parameter.')
     }
 
-    // 每次请求时重新生成配置以确保API key轮换生效
-    this.config = providerToAiSdkConfig(this.actualProvider, this.model)
-    logger.debug('Generated provider config for completions', this.config)
+    // Config is now set in constructor, ApiService handles key rotation before passing provider
+    if (!this.config) {
+      // If config wasn't set in constructor (when provider only), generate it now
+      this.config = providerToAiSdkConfig(this.actualProvider, this.model!)
+    }
+    logger.debug('Using provider config for completions', this.config)
 
     // 检查 config 是否存在
     if (!this.config) {

src/renderer/src/aiCore/provider/providerConfig.ts

@@ -29,32 +29,6 @@ import { azureAnthropicProviderCreator } from './config/azure-anthropic'
 import { COPILOT_DEFAULT_HEADERS } from './constants'
 import { getAiSdkProviderId } from './factory'
 
-/**
- * 获取轮询的API key
- * 复用legacy架构的多key轮询逻辑
- */
-function getRotatedApiKey(provider: Provider): string {
-  const keys = provider.apiKey.split(',').map((key) => key.trim())
-  const keyName = `provider:${provider.id}:last_used_key`
-
-  if (keys.length === 1) {
-    return keys[0]
-  }
-
-  const lastUsedKey = window.keyv.get(keyName)
-  if (!lastUsedKey) {
-    window.keyv.set(keyName, keys[0])
-    return keys[0]
-  }
-
-  const currentIndex = keys.indexOf(lastUsedKey)
-  const nextIndex = (currentIndex + 1) % keys.length
-  const nextKey = keys[nextIndex]
-  window.keyv.set(keyName, nextKey)
-
-  return nextKey
-}
-
 /**
  * 处理特殊provider的转换逻辑
  */
@@ -78,11 +52,13 @@ function handleSpecialProviders(model: Model, provider: Provider): Provider {
 }
 
 /**
- * 主要用来对齐AISdk的BaseURL格式
- * @param provider
- * @returns
+ * Format and normalize the API host URL for a provider.
+ * Handles provider-specific URL formatting rules (e.g., appending version paths, Azure formatting).
+ *
+ * @param provider - The provider whose API host is to be formatted.
+ * @returns A new provider instance with the formatted API host.
  */
-function formatProviderApiHost(provider: Provider): Provider {
+export function formatProviderApiHost(provider: Provider): Provider {
   const formatted = { ...provider }
   if (formatted.anthropicApiHost) {
     formatted.anthropicApiHost = formatApiHost(formatted.anthropicApiHost)
@@ -114,18 +90,38 @@ function formatProviderApiHost(provider: Provider): Provider {
 }
 
 /**
- * 获取实际的Provider配置
- * 简化版：将逻辑分解为小函数
+ * Retrieve the effective Provider configuration for the given model.
+ * Applies all necessary transformations (special-provider handling, URL formatting, etc.).
+ *
+ * @param model - The model whose provider is to be resolved.
+ * @returns A new Provider instance with all adaptations applied.
  */
 export function getActualProvider(model: Model): Provider {
   const baseProvider = getProviderByModel(model)
 
-  // 按顺序处理各种转换
-  let actualProvider = cloneDeep(baseProvider)
-  actualProvider = handleSpecialProviders(model, actualProvider)
-  actualProvider = formatProviderApiHost(actualProvider)
+  return adaptProvider({ provider: baseProvider, model })
+}
 
-  return actualProvider
+/**
+ * Transforms a provider configuration by applying model-specific adaptations and normalizing its API host.
+ * The transformations are applied in the following order:
+ * 1. Model-specific provider handling (e.g., New-API, system providers, Azure OpenAI)
+ * 2. API host formatting (provider-specific URL normalization)
+ *
+ * @param provider - The base provider configuration to transform.
+ * @param model - The model associated with the provider; optional but required for special-provider handling.
+ * @returns A new Provider instance with all transformations applied.
+ */
+export function adaptProvider({ provider, model }: { provider: Provider; model?: Model }): Provider {
+  let adaptedProvider = cloneDeep(provider)
+
+  // Apply transformations in order
+  if (model) {
+    adaptedProvider = handleSpecialProviders(model, adaptedProvider)
+  }
+  adaptedProvider = formatProviderApiHost(adaptedProvider)
+
+  return adaptedProvider
 }
 
 /**
@@ -139,7 +135,7 @@ export function providerToAiSdkConfig(actualProvider: Provider, model: Model): A
   const { baseURL, endpoint } = routeToEndpoint(actualProvider.apiHost)
   const baseConfig = {
     baseURL: baseURL,
-    apiKey: getRotatedApiKey(actualProvider)
+    apiKey: actualProvider.apiKey
   }
 
   const isCopilotProvider = actualProvider.id === SystemProviderIds.copilot

src/renderer/src/aiCore/trace/AiSdkSpanAdapter.ts

@@ -245,8 +245,8 @@ export class AiSdkSpanAdapter {
       'gen_ai.usage.output_tokens'
     ]
 
-    const completionTokens = attributes[inputsTokenKeys.find((key) => attributes[key]) || '']
-    const promptTokens = attributes[outputTokenKeys.find((key) => attributes[key]) || '']
+    const promptTokens = attributes[inputsTokenKeys.find((key) => attributes[key]) || '']
+    const completionTokens = attributes[outputTokenKeys.find((key) => attributes[key]) || '']
 
     if (completionTokens !== undefined || promptTokens !== undefined) {
       const usage: TokenUsage = {

src/renderer/src/aiCore/trace/__tests__/AiSdkSpanAdapter.test.ts

@@ -0,0 +1,53 @@
+import type { Span } from '@opentelemetry/api'
+import { SpanKind, SpanStatusCode } from '@opentelemetry/api'
+import { describe, expect, it, vi } from 'vitest'
+
+import { AiSdkSpanAdapter } from '../AiSdkSpanAdapter'
+
+vi.mock('@logger', () => ({
+  loggerService: {
+    withContext: () => ({
+      debug: vi.fn(),
+      error: vi.fn(),
+      info: vi.fn(),
+      warn: vi.fn()
+    })
+  }
+}))
+
+describe('AiSdkSpanAdapter', () => {
+  const createMockSpan = (attributes: Record<string, unknown>): Span => {
+    const span = {
+      spanContext: () => ({
+        traceId: 'trace-id',
+        spanId: 'span-id'
+      }),
+      _attributes: attributes,
+      _events: [],
+      name: 'test span',
+      status: { code: SpanStatusCode.OK },
+      kind: SpanKind.CLIENT,
+      startTime: [0, 0] as [number, number],
+      endTime: [0, 1] as [number, number],
+      ended: true,
+      parentSpanId: '',
+      links: []
+    }
+    return span as unknown as Span
+  }
+
+  it('maps prompt and completion usage tokens to the correct fields', () => {
+    const attributes = {
+      'ai.usage.promptTokens': 321,
+      'ai.usage.completionTokens': 654
+    }
+
+    const span = createMockSpan(attributes)
+    const result = AiSdkSpanAdapter.convertToSpanEntity({ span })
+
+    expect(result.usage).toBeDefined()
+    expect(result.usage?.prompt_tokens).toBe(321)
+    expect(result.usage?.completion_tokens).toBe(654)
+    expect(result.usage?.total_tokens).toBe(975)
+  })
+})

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

@@ -144,7 +144,7 @@ describe('reasoning utils', () => {
       expect(result).toEqual({})
     })
 
-    it('should disable reasoning for OpenRouter when no reasoning effort set', async () => {
+    it('should not override reasoning for OpenRouter when reasoning effort undefined', async () => {
       const { isReasoningModel } = await import('@renderer/config/models')
 
       vi.mocked(isReasoningModel).mockReturnValue(true)
@@ -161,6 +161,29 @@ describe('reasoning utils', () => {
         settings: {}
       } as Assistant
 
+      const result = getReasoningEffort(assistant, model)
+      expect(result).toEqual({})
+    })
+
+    it('should disable reasoning for OpenRouter when reasoning effort explicitly none', async () => {
+      const { isReasoningModel } = await import('@renderer/config/models')
+
+      vi.mocked(isReasoningModel).mockReturnValue(true)
+
+      const model: Model = {
+        id: 'anthropic/claude-sonnet-4',
+        name: 'Claude Sonnet 4',
+        provider: SystemProviderIds.openrouter
+      } as Model
+
+      const assistant: Assistant = {
+        id: 'test',
+        name: 'Test',
+        settings: {
+          reasoning_effort: 'none'
+        }
+      } as Assistant
+
       const result = getReasoningEffort(assistant, model)
       expect(result).toEqual({ reasoning: { enabled: false, exclude: true } })
     })
@@ -269,7 +292,9 @@ describe('reasoning utils', () => {
       const assistant: Assistant = {
         id: 'test',
         name: 'Test',
-        settings: {}
+        settings: {
+          reasoning_effort: 'none'
+        }
       } as Assistant
 
       const result = getReasoningEffort(assistant, model)

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

@@ -16,10 +16,8 @@ import {
   isGPT5SeriesModel,
   isGPT51SeriesModel,
   isGrok4FastReasoningModel,
-  isGrokReasoningModel,
   isOpenAIDeepResearchModel,
   isOpenAIModel,
-  isOpenAIReasoningModel,
   isQwenAlwaysThinkModel,
   isQwenReasoningModel,
   isReasoningModel,
@@ -64,30 +62,22 @@ export function getReasoningEffort(assistant: Assistant, model: Model): Reasonin
   }
   const reasoningEffort = assistant?.settings?.reasoning_effort
 
-  // Handle undefined and 'none' reasoningEffort.
-  // TODO: They should be separated.
-  if (!reasoningEffort || reasoningEffort === 'none') {
+  // reasoningEffort is not set, no extra reasoning setting
+  // Generally, for every model which supports reasoning control, the reasoning effort won't be undefined.
+  // It's for some reasoning models that don't support reasoning control, such as deepseek reasoner.
+  if (!reasoningEffort) {
+    return {}
+  }
+
+  // Handle 'none' reasoningEffort. It's explicitly off.
+  if (reasoningEffort === 'none') {
     // openrouter: use reasoning
     if (model.provider === SystemProviderIds.openrouter) {
-      // Don't disable reasoning for Gemini models that support thinking tokens
-      if (isSupportedThinkingTokenGeminiModel(model) && !GEMINI_FLASH_MODEL_REGEX.test(model.id)) {
-        return {}
-      }
       // 'none' is not an available value for effort for now.
       // I think they should resolve this issue soon, so I'll just go ahead and use this value.
       if (isGPT51SeriesModel(model) && reasoningEffort === 'none') {
         return { reasoning: { effort: 'none' } }
       }
-      // Don't disable reasoning for models that require it
-      if (
-        isGrokReasoningModel(model) ||
-        isOpenAIReasoningModel(model) ||
-        isQwenAlwaysThinkModel(model) ||
-        model.id.includes('seed-oss') ||
-        model.id.includes('minimax-m2')
-      ) {
-        return {}
-      }
       return { reasoning: { enabled: false, exclude: true } }
     }
 
@@ -101,11 +91,6 @@ export function getReasoningEffort(assistant: Assistant, model: Model): Reasonin
       return { enable_thinking: false }
     }
 
-    // claude
-    if (isSupportedThinkingTokenClaudeModel(model)) {
-      return {}
-    }
-
     // gemini
     if (isSupportedThinkingTokenGeminiModel(model)) {
       if (GEMINI_FLASH_MODEL_REGEX.test(model.id)) {
@@ -118,8 +103,10 @@ export function getReasoningEffort(assistant: Assistant, model: Model): Reasonin
             }
           }
         }
+      } else {
+        logger.warn(`Model ${model.id} cannot disable reasoning. Fallback to empty reasoning param.`)
+        return {}
       }
-      return {}
     }
 
     // use thinking, doubao, zhipu, etc.
@@ -139,6 +126,7 @@ export function getReasoningEffort(assistant: Assistant, model: Model): Reasonin
       }
     }
 
+    logger.warn(`Model ${model.id} doesn't match any disable reasoning behavior. Fallback to empty reasoning param.`)
     return {}
   }
 
@@ -293,6 +281,7 @@ export function getReasoningEffort(assistant: Assistant, model: Model): Reasonin
   }
 
   // OpenRouter models, use reasoning
+  // FIXME: duplicated openrouter handling. remove one
   if (model.provider === SystemProviderIds.openrouter) {
     if (isSupportedReasoningEffortModel(model) || isSupportedThinkingTokenModel(model)) {
       return {

src/renderer/src/assets/styles/ant.css

@@ -215,6 +215,10 @@
   border-top: none !important;
 }
 
+.ant-collapse-header-text {
+  overflow-x: hidden;
+}
+
 .ant-slider .ant-slider-handle::after {
   box-shadow: 0 1px 4px 0px rgb(128 128 128 / 50%) !important;
 }

src/renderer/src/config/models/reasoning.ts

@@ -460,16 +460,19 @@ export const isSupportedThinkingTokenZhipuModel = (model: Model): boolean => {
 }
 
 export const isDeepSeekHybridInferenceModel = (model: Model) => {
-  const modelId = getLowerBaseModelName(model.id)
-  // deepseek官方使用chat和reasoner做推理控制，其他provider需要单独判断，id可能会有所差别
-  // openrouter: deepseek/deepseek-chat-v3.1 不知道会不会有其他provider仿照ds官方分出一个同id的作为非思考模式的模型，这里有风险
-  // Matches: "deepseek-v3" followed by ".digit" or "-digit".
-  // Optionally, this can be followed by ".alphanumeric_sequence" or "-alphanumeric_sequence"
-  // until the end of the string.
-  // Examples: deepseek-v3.1, deepseek-v3-1, deepseek-v3.1.2, deepseek-v3.1-alpha
-  // Does NOT match: deepseek-v3.123 (missing separator after '1'), deepseek-v3.x (x isn't a digit)
-  // TODO: move to utils and add test cases
-  return /deepseek-v3(?:\.\d|-\d)(?:(\.|-)\w+)?$/.test(modelId) || modelId.includes('deepseek-chat-v3.1')
+  const { idResult, nameResult } = withModelIdAndNameAsId(model, (model) => {
+    const modelId = getLowerBaseModelName(model.id)
+    // deepseek官方使用chat和reasoner做推理控制，其他provider需要单独判断，id可能会有所差别
+    // openrouter: deepseek/deepseek-chat-v3.1 不知道会不会有其他provider仿照ds官方分出一个同id的作为非思考模式的模型，这里有风险
+    // Matches: "deepseek-v3" followed by ".digit" or "-digit".
+    // Optionally, this can be followed by ".alphanumeric_sequence" or "-alphanumeric_sequence"
+    // until the end of the string.
+    // Examples: deepseek-v3.1, deepseek-v3-1, deepseek-v3.1.2, deepseek-v3.1-alpha
+    // Does NOT match: deepseek-v3.123 (missing separator after '1'), deepseek-v3.x (x isn't a digit)
+    // TODO: move to utils and add test cases
+    return /deepseek-v3(?:\.\d|-\d)(?:(\.|-)\w+)?$/.test(modelId) || modelId.includes('deepseek-chat-v3.1')
+  })
+  return idResult || nameResult
 }
 
 export const isLingReasoningModel = (model?: Model): boolean => {
@@ -523,7 +526,6 @@ export function isReasoningModel(model?: Model): boolean {
       REASONING_REGEX.test(model.name) ||
       isSupportedThinkingTokenDoubaoModel(model) ||
       isDeepSeekHybridInferenceModel(model) ||
-      isDeepSeekHybridInferenceModel({ ...model, id: model.name }) ||
       false
     )
   }

src/renderer/src/hooks/useScrollPosition.ts

@@ -1,5 +1,5 @@
 import { throttle } from 'lodash'
-import { useEffect, useRef } from 'react'
+import { useEffect, useMemo, useRef } from 'react'
 
 import { useTimer } from './useTimer'
 
@@ -12,13 +12,18 @@ import { useTimer } from './useTimer'
  */
 export default function useScrollPosition(key: string, throttleWait?: number) {
   const containerRef = useRef<HTMLDivElement>(null)
-  const scrollKey = `scroll:${key}`
+  const scrollKey = useMemo(() => `scroll:${key}`, [key])
+  const scrollKeyRef = useRef(scrollKey)
   const { setTimeoutTimer } = useTimer()
 
+  useEffect(() => {
+    scrollKeyRef.current = scrollKey
+  }, [scrollKey])
+
   const handleScroll = throttle(() => {
     const position = containerRef.current?.scrollTop ?? 0
     window.requestAnimationFrame(() => {
-      window.keyv.set(scrollKey, position)
+      window.keyv.set(scrollKeyRef.current, position)
     })
   }, throttleWait ?? 100)
 
@@ -28,5 +33,9 @@ export default function useScrollPosition(key: string, throttleWait?: number) {
     setTimeoutTimer('scrollEffect', scroll, 50)
   }, [scrollKey, setTimeoutTimer])
 
+  useEffect(() => {
+    return () => handleScroll.cancel()
+  }, [handleScroll])
+
   return { containerRef, handleScroll }
 }

src/renderer/src/hooks/useTimer.ts

@@ -1,4 +1,4 @@
-import { useEffect, useRef } from 'react'
+import { useCallback, useEffect, useRef } from 'react'
 
 /**
  * 定时器管理 Hook，用于管理 setTimeout 和 setInterval 定时器，支持通过 key 来标识不同的定时器
@@ -43,10 +43,38 @@ export const useTimer = () => {
   const timeoutMapRef = useRef(new Map<string, NodeJS.Timeout>())
   const intervalMapRef = useRef(new Map<string, NodeJS.Timeout>())
 
+  /**
+   * 清除指定 key 的 setTimeout 定时器
+   * @param key - 定时器标识符
+   */
+  const clearTimeoutTimer = useCallback((key: string) => {
+    clearTimeout(timeoutMapRef.current.get(key))
+    timeoutMapRef.current.delete(key)
+  }, [])
+
+  /**
+   * 清除指定 key 的 setInterval 定时器
+   * @param key - 定时器标识符
+   */
+  const clearIntervalTimer = useCallback((key: string) => {
+    clearInterval(intervalMapRef.current.get(key))
+    intervalMapRef.current.delete(key)
+  }, [])
+
+  /**
+   * 清除所有定时器，包括 setTimeout 和 setInterval
+   */
+  const clearAllTimers = useCallback(() => {
+    timeoutMapRef.current.forEach((timer) => clearTimeout(timer))
+    intervalMapRef.current.forEach((timer) => clearInterval(timer))
+    timeoutMapRef.current.clear()
+    intervalMapRef.current.clear()
+  }, [])
+
   // 组件卸载时自动清理所有定时器
   useEffect(() => {
     return () => clearAllTimers()
-  }, [])
+  }, [clearAllTimers])
 
   /**
    * 设置一个 setTimeout 定时器
@@ -65,12 +93,15 @@ export const useTimer = () => {
    * cleanup();
    * ```
    */
-  const setTimeoutTimer = (key: string, ...args: Parameters<typeof setTimeout>) => {
-    clearTimeout(timeoutMapRef.current.get(key))
-    const timer = setTimeout(...args)
-    timeoutMapRef.current.set(key, timer)
-    return () => clearTimeoutTimer(key)
-  }
+  const setTimeoutTimer = useCallback(
+    (key: string, ...args: Parameters<typeof setTimeout>) => {
+      clearTimeout(timeoutMapRef.current.get(key))
+      const timer = setTimeout(...args)
+      timeoutMapRef.current.set(key, timer)
+      return () => clearTimeoutTimer(key)
+    },
+    [clearTimeoutTimer]
+  )
 
   /**
    * 设置一个 setInterval 定时器
@@ -89,56 +120,31 @@ export const useTimer = () => {
    * cleanup();
    * ```
    */
-  const setIntervalTimer = (key: string, ...args: Parameters<typeof setInterval>) => {
-    clearInterval(intervalMapRef.current.get(key))
-    const timer = setInterval(...args)
-    intervalMapRef.current.set(key, timer)
-    return () => clearIntervalTimer(key)
-  }
-
-  /**
-   * 清除指定 key 的 setTimeout 定时器
-   * @param key - 定时器标识符
-   */
-  const clearTimeoutTimer = (key: string) => {
-    clearTimeout(timeoutMapRef.current.get(key))
-    timeoutMapRef.current.delete(key)
-  }
-
-  /**
-   * 清除指定 key 的 setInterval 定时器
-   * @param key - 定时器标识符
-   */
-  const clearIntervalTimer = (key: string) => {
-    clearInterval(intervalMapRef.current.get(key))
-    intervalMapRef.current.delete(key)
-  }
+  const setIntervalTimer = useCallback(
+    (key: string, ...args: Parameters<typeof setInterval>) => {
+      clearInterval(intervalMapRef.current.get(key))
+      const timer = setInterval(...args)
+      intervalMapRef.current.set(key, timer)
+      return () => clearIntervalTimer(key)
+    },
+    [clearIntervalTimer]
+  )
 
   /**
    * 清除所有 setTimeout 定时器
    */
-  const clearAllTimeoutTimers = () => {
+  const clearAllTimeoutTimers = useCallback(() => {
     timeoutMapRef.current.forEach((timer) => clearTimeout(timer))
     timeoutMapRef.current.clear()
-  }
+  }, [])
 
   /**
    * 清除所有 setInterval 定时器
    */
-  const clearAllIntervalTimers = () => {
+  const clearAllIntervalTimers = useCallback(() => {
     intervalMapRef.current.forEach((timer) => clearInterval(timer))
     intervalMapRef.current.clear()
-  }
-
-  /**
-   * 清除所有定时器，包括 setTimeout 和 setInterval
-   */
-  const clearAllTimers = () => {
-    timeoutMapRef.current.forEach((timer) => clearTimeout(timer))
-    intervalMapRef.current.forEach((timer) => clearInterval(timer))
-    timeoutMapRef.current.clear()
-    intervalMapRef.current.clear()
-  }
+  }, [])
 
   return {
     setTimeoutTimer,

src/renderer/src/i18n/locales/en-us.json

@@ -280,6 +280,7 @@
         "denied": "Tool request was denied.",
         "timeout": "Tool request timed out before receiving approval."
       },
+      "toolPendingFallback": "Tool",
       "waiting": "Waiting for tool permission decision..."
     },
     "type": {
@@ -1208,7 +1209,7 @@
   "endpoint_type": {
     "anthropic": "Anthropic",
     "gemini": "Gemini",
-    "image-generation": "Image Generation",
+    "image-generation": "Image Generation (OpenAI)",
     "jina-rerank": "Jina Rerank",
     "openai": "OpenAI",
     "openai-response": "OpenAI-Response"
@@ -4371,7 +4372,7 @@
         "url": {
           "preview": "Preview: {{url}}",
           "reset": "Reset",
-          "tip": "ending with # forces use of input address"
+          "tip": "Add # at the end to disable the automatically appended API version."
         }
       },
       "api_host": "API Host",

src/renderer/src/i18n/locales/zh-cn.json

@@ -280,6 +280,7 @@
         "denied": "工具请求已被拒绝。",
         "timeout": "工具请求在收到批准前超时。"
       },
+      "toolPendingFallback": "工具",
       "waiting": "等待工具权限决定..."
     },
     "type": {
@@ -1208,7 +1209,7 @@
   "endpoint_type": {
     "anthropic": "Anthropic",
     "gemini": "Gemini",
-    "image-generation": "图片生成",
+    "image-generation": "图像生成 (OpenAI)",
     "jina-rerank": "Jina 重排序",
     "openai": "OpenAI",
     "openai-response": "OpenAI-Response"
@@ -4371,7 +4372,7 @@
         "url": {
           "preview": "预览: {{url}}",
           "reset": "重置",
-          "tip": "# 结尾强制使用输入地址"
+          "tip": "在末尾添加 # 以禁用自动附加的API版本。"
         }
       },
       "api_host": "API 地址",

src/renderer/src/i18n/locales/zh-tw.json

@@ -280,6 +280,7 @@
         "denied": "工具請求已被拒絕。",
         "timeout": "工具請求在收到核准前逾時。"
       },
+      "toolPendingFallback": "工具",
       "waiting": "等待工具權限決定..."
     },
     "type": {
@@ -1208,7 +1209,7 @@
   "endpoint_type": {
     "anthropic": "Anthropic",
     "gemini": "Gemini",
-    "image-generation": "圖片生成",
+    "image-generation": "圖像生成 (OpenAI)",
     "jina-rerank": "Jina Rerank",
     "openai": "OpenAI",
     "openai-response": "OpenAI-Response"
@@ -4371,7 +4372,7 @@
         "url": {
           "preview": "預覽：{{url}}",
           "reset": "重設",
-          "tip": "# 結尾強制使用輸入位址"
+          "tip": "在末尾添加 # 以停用自動附加的 API 版本。"
         }
       },
       "api_host": "API 主機地址",

src/renderer/src/i18n/translate/de-de.json

@@ -280,6 +280,7 @@
         "denied": "Tool-Anfrage wurde abgelehnt.",
         "timeout": "Tool-Anfrage ist abgelaufen, bevor eine Genehmigung eingegangen ist."
       },
+      "toolPendingFallback": "Werkzeug",
       "waiting": "Warten auf Entscheidung über Tool-Berechtigung..."
     },
     "type": {
@@ -1208,7 +1209,7 @@
   "endpoint_type": {
     "anthropic": "Anthropic",
     "gemini": "Gemini",
-    "image-generation": "Bildgenerierung",
+    "image-generation": "Bilderzeugung (OpenAI)",
     "jina-rerank": "Jina Reranking",
     "openai": "OpenAI",
     "openai-response": "OpenAI-Response"
@@ -4371,7 +4372,7 @@
         "url": {
           "preview": "Vorschau: {{url}}",
           "reset": "Zurücksetzen",
-          "tip": "# am Ende erzwingt die Verwendung der Eingabe-Adresse"
+          "tip": "Fügen Sie am Ende ein # hinzu, um die automatisch angehängte API-Version zu deaktivieren."
         }
       },
       "api_host": "API-Adresse",

src/renderer/src/i18n/translate/el-gr.json

@@ -280,6 +280,7 @@
         "denied": "Το αίτημα για εργαλείο απορρίφθηκε.",
         "timeout": "Το αίτημα για το εργαλείο έληξε πριν λάβει έγκριση."
       },
+      "toolPendingFallback": "Εργαλείο",
       "waiting": "Αναμονή για απόφαση άδειας εργαλείου..."
     },
     "type": {
@@ -1208,7 +1209,7 @@
   "endpoint_type": {
     "anthropic": "Anthropic",
     "gemini": "Gemini",
-    "image-generation": "Δημιουργία Εικόνας",
+    "image-generation": "Δημιουργία Εικόνων (OpenAI)",
     "jina-rerank": "Επαναταξινόμηση Jina",
     "openai": "OpenAI",
     "openai-response": "Απάντηση OpenAI"
@@ -4371,7 +4372,7 @@
         "url": {
           "preview": "Προεπισκόπηση: {{url}}",
           "reset": "Επαναφορά",
-          "tip": "#τέλος ενδεχόμενη χρήση της εισαγωγής διευθύνσεως"
+          "tip": "Προσθέστε το σύμβολο # στο τέλος για να απενεργοποιήσετε την αυτόματα προστιθέμενη έκδοση API."
         }
       },
       "api_host": "Διεύθυνση API",

src/renderer/src/i18n/translate/es-es.json

@@ -280,6 +280,7 @@
         "denied": "La solicitud de herramienta fue denegada.",
         "timeout": "La solicitud de herramienta expiró antes de recibir la aprobación."
       },
+      "toolPendingFallback": "Herramienta",
       "waiting": "Esperando la decisión de permiso de la herramienta..."
     },
     "type": {
@@ -1208,7 +1209,7 @@
   "endpoint_type": {
     "anthropic": "Anthropic",
     "gemini": "Gemini",
-    "image-generation": "Generación de imágenes",
+    "image-generation": "Generación de Imágenes (OpenAI)",
     "jina-rerank": "Reordenamiento Jina",
     "openai": "OpenAI",
     "openai-response": "Respuesta de OpenAI"
@@ -4371,7 +4372,7 @@
         "url": {
           "preview": "Vista previa: {{url}}",
           "reset": "Restablecer",
-          "tip": "forzar uso de dirección de entrada con # al final"
+          "tip": "Añada # al final para deshabilitar la versión de la API que se añade automáticamente."
         }
       },
       "api_host": "Dirección API",

src/renderer/src/i18n/translate/fr-fr.json

@@ -280,6 +280,7 @@
         "denied": "La demande d'outil a été refusée.",
         "timeout": "La demande d'outil a expiré avant d'obtenir l'approbation."
       },
+      "toolPendingFallback": "Outil",
       "waiting": "En attente de la décision d'autorisation de l'outil..."
     },
     "type": {
@@ -1208,7 +1209,7 @@
   "endpoint_type": {
     "anthropic": "Anthropic",
     "gemini": "Gemini",
-    "image-generation": "Génération d'images",
+    "image-generation": "Génération d'images (OpenAI)",
     "jina-rerank": "Reclassement Jina",
     "openai": "OpenAI",
     "openai-response": "Réponse OpenAI"
@@ -4371,7 +4372,7 @@
         "url": {
           "preview": "Aperçu : {{url}}",
           "reset": "Réinitialiser",
-          "tip": "forcer l'utilisation de l'adresse d'entrée si terminé par #"
+          "tip": "Ajoutez # à la fin pour désactiver la version d'API ajoutée automatiquement."
         }
       },
       "api_host": "Adresse API",

src/renderer/src/i18n/translate/ja-jp.json

@@ -280,6 +280,7 @@
         "denied": "ツールリクエストは拒否されました。",
         "timeout": "ツールリクエストは承認を受ける前にタイムアウトしました。"
       },
+      "toolPendingFallback": "ツール",
       "waiting": "ツールの許可決定を待っています..."
     },
     "type": {
@@ -1208,7 +1209,7 @@
   "endpoint_type": {
     "anthropic": "Anthropic",
     "gemini": "Gemini",
-    "image-generation": "画像生成",
+    "image-generation": "画像生成 (OpenAI)",
     "jina-rerank": "Jina Rerank",
     "openai": "OpenAI",
     "openai-response": "OpenAI-Response"
@@ -4371,7 +4372,7 @@
         "url": {
           "preview": "プレビュー: {{url}}",
           "reset": "リセット",
-          "tip": "#で終わる場合、入力されたアドレスを強制的に使用します"
+          "tip": "自動的に付加されるAPIバージョンを無効にするには、末尾に#を追加します。"
         }
       },
       "api_host": "APIホスト",

src/renderer/src/i18n/translate/pt-pt.json

@@ -16,7 +16,7 @@
       "error": {
         "failed": "Falha ao excluir o agente"
       },
-      "title": "删除代理"
+      "title": "Excluir Agente"
     },
     "edit": {
       "title": "Agent Editor"
@@ -111,7 +111,7 @@
           "label": "Modo de permissão",
           "options": {
             "acceptEdits": "Aceitar edições automaticamente",
-            "bypassPermissions": "忽略检查 de permissão",
+            "bypassPermissions": "Ignorar verificações de permissão",
             "default": "Padrão (perguntar antes de continuar)",
             "plan": "Modo de planejamento (plano sujeito a aprovação)"
           },
@@ -150,7 +150,7 @@
         },
         "success": {
           "install": "Plugin instalado com sucesso",
-          "uninstall": "插件 desinstalado com sucesso"
+          "uninstall": "Plugin desinstalado com sucesso"
         },
         "tab": "plug-in",
         "type": {
@@ -280,6 +280,7 @@
         "denied": "Solicitação de ferramenta foi negada.",
         "timeout": "A solicitação da ferramenta expirou antes de receber aprovação."
       },
+      "toolPendingFallback": "Ferramenta",
       "waiting": "Aguardando decisão de permissão da ferramenta..."
     },
     "type": {
@@ -1134,7 +1135,7 @@
     "duplicate": "Duplicar",
     "edit": "Editar",
     "enabled": "Ativado",
-    "error": "错误",
+    "error": "Erro",
     "errors": {
       "create_message": "Falha ao criar mensagem",
       "validation": "Falha na verificação"
@@ -1208,7 +1209,7 @@
   "endpoint_type": {
     "anthropic": "Anthropic",
     "gemini": "Gemini",
-    "image-generation": "Geração de Imagem",
+    "image-generation": "Geração de Imagens (OpenAI)",
     "jina-rerank": "Jina Reordenar",
     "openai": "OpenAI",
     "openai-response": "Resposta OpenAI"
@@ -4371,7 +4372,7 @@
         "url": {
           "preview": "Pré-visualização: {{url}}",
           "reset": "Redefinir",
-          "tip": "e forçar o uso do endereço original quando terminar com '#'"
+          "tip": "Adicione # no final para desativar a versão da API adicionada automaticamente."
         }
       },
       "api_host": "Endereço API",

src/renderer/src/i18n/translate/ru-ru.json

@@ -280,6 +280,7 @@
         "denied": "Запрос на инструмент был отклонён.",
         "timeout": "Запрос на инструмент превысил время ожидания до получения подтверждения."
       },
+      "toolPendingFallback": "Инструмент",
       "waiting": "Ожидание решения о разрешении на использование инструмента..."
     },
     "type": {
@@ -1208,7 +1209,7 @@
   "endpoint_type": {
     "anthropic": "Anthropic",
     "gemini": "Gemini",
-    "image-generation": "Изображение",
+    "image-generation": "Генерация изображений (OpenAI)",
     "jina-rerank": "Jina Rerank",
     "openai": "OpenAI",
     "openai-response": "OpenAI-Response"
@@ -4371,7 +4372,7 @@
         "url": {
           "preview": "Предпросмотр: {{url}}",
           "reset": "Сброс",
-          "tip": "заканчивая на # принудительно использует введенный адрес"
+          "tip": "Добавьте # в конце, чтобы отключить автоматически добавляемую версию API."
         }
       },
       "api_host": "Хост API",

src/renderer/src/pages/home/Inputbar/AgentSessionInputbar.tsx

@@ -9,6 +9,7 @@ import { getModel } from '@renderer/hooks/useModel'
 import { useSettings } from '@renderer/hooks/useSettings'
 import { useTextareaResize } from '@renderer/hooks/useTextareaResize'
 import { useTimer } from '@renderer/hooks/useTimer'
+import { CacheService } from '@renderer/services/CacheService'
 import { pauseTrace } from '@renderer/services/SpanManagerService'
 import { estimateUserPromptUsage } from '@renderer/services/TokenService'
 import { useAppDispatch, useAppSelector } from '@renderer/store'
@@ -41,19 +42,10 @@ import { getInputbarConfig } from './registry'
 import { TopicType } from './types'
 
 const logger = loggerService.withContext('AgentSessionInputbar')
-const agentSessionDraftCache = new Map<string, string>()
 
-const readDraftFromCache = (key: string): string => {
-  return agentSessionDraftCache.get(key) ?? ''
-}
+const DRAFT_CACHE_TTL = 24 * 60 * 60 * 1000 // 24 hours
 
-const writeDraftToCache = (key: string, value: string) => {
-  if (!value) {
-    agentSessionDraftCache.delete(key)
-  } else {
-    agentSessionDraftCache.set(key, value)
-  }
-}
+const getAgentDraftCacheKey = (agentId: string) => `agent-session-draft-${agentId}`
 
 type Props = {
   agentId: string
@@ -170,16 +162,15 @@ const AgentSessionInputbarInner: FC<InnerProps> = ({ assistant, agentId, session
   const scope = TopicType.Session
   const config = getInputbarConfig(scope)
 
-  // Use shared hooks for text and textarea management
-  const initialDraft = useMemo(() => readDraftFromCache(agentId), [agentId])
-  const persistDraft = useCallback((next: string) => writeDraftToCache(agentId, next), [agentId])
+  // Use shared hooks for text and textarea management with draft persistence
+  const draftCacheKey = getAgentDraftCacheKey(agentId)
   const {
     text,
     setText,
     isEmpty: inputEmpty
   } = useInputText({
-    initialValue: initialDraft,
-    onChange: persistDraft
+    initialValue: CacheService.get<string>(draftCacheKey) ?? '',
+    onChange: (value) => CacheService.set(draftCacheKey, value, DRAFT_CACHE_TTL)
   })
   const {
     textareaRef,
@@ -431,6 +422,7 @@ const AgentSessionInputbarInner: FC<InnerProps> = ({ assistant, agentId, session
         })
       )
 
+      // Clear text after successful send (draft is cleared automatically via onChange)
       setText('')
       setTimeoutTimer('agentSession_sendMessage', () => setText(''), 500)
     } catch (error) {

src/renderer/src/pages/home/Inputbar/Inputbar.tsx

@@ -14,7 +14,6 @@ import { useInputText } from '@renderer/hooks/useInputText'
 import { useMessageOperations, useTopicLoading } from '@renderer/hooks/useMessageOperations'
 import { useSettings } from '@renderer/hooks/useSettings'
 import { useShortcut } from '@renderer/hooks/useShortcuts'
-import { useSidebarIconShow } from '@renderer/hooks/useSidebarIcon'
 import { useTextareaResize } from '@renderer/hooks/useTextareaResize'
 import { useTimer } from '@renderer/hooks/useTimer'
 import {
@@ -24,6 +23,7 @@ import {
   useInputbarToolsState
 } from '@renderer/pages/home/Inputbar/context/InputbarToolsProvider'
 import { getDefaultTopic } from '@renderer/services/AssistantService'
+import { CacheService } from '@renderer/services/CacheService'
 import { EVENT_NAMES, EventEmitter } from '@renderer/services/EventService'
 import FileManager from '@renderer/services/FileManager'
 import { checkRateLimit, getUserMessage } from '@renderer/services/MessagesService'
@@ -39,7 +39,7 @@ import { getSendMessageShortcutLabel } from '@renderer/utils/input'
 import { documentExts, imageExts, textExts } from '@shared/config/constant'
 import { debounce } from 'lodash'
 import type { FC } from 'react'
-import React, { useCallback, useEffect, useMemo, useRef, useState } from 'react'
+import React, { useCallback, useEffect, useEffectEvent, useMemo, useRef, useState } from 'react'
 import { useTranslation } from 'react-i18next'
 
 import { InputbarCore } from './components/InputbarCore'
@@ -51,6 +51,17 @@ import TokenCount from './TokenCount'
 
 const logger = loggerService.withContext('Inputbar')
 
+const INPUTBAR_DRAFT_CACHE_KEY = 'inputbar-draft'
+const DRAFT_CACHE_TTL = 24 * 60 * 60 * 1000 // 24 hours
+
+const getMentionedModelsCacheKey = (assistantId: string) => `inputbar-mentioned-models-${assistantId}`
+
+const getValidatedCachedModels = (assistantId: string): Model[] => {
+  const cached = CacheService.get<Model[]>(getMentionedModelsCacheKey(assistantId))
+  if (!Array.isArray(cached)) return []
+  return cached.filter((model) => model?.id && model?.name)
+}
+
 interface Props {
   assistant: Assistant
   setActiveTopic: (topic: Topic) => void
@@ -80,16 +91,18 @@ const Inputbar: FC<Props> = ({ assistant: initialAssistant, setActiveTopic, topi
     toggleExpanded: () => {}
   })
 
+  const [initialMentionedModels] = useState(() => getValidatedCachedModels(initialAssistant.id))
+
   const initialState = useMemo(
     () => ({
       files: [] as FileType[],
-      mentionedModels: [] as Model[],
+      mentionedModels: initialMentionedModels,
       selectedKnowledgeBases: initialAssistant.knowledge_bases ?? [],
       isExpanded: false,
       couldAddImageFile: false,
       extensions: [] as string[]
     }),
-    [initialAssistant.knowledge_bases]
+    [initialMentionedModels, initialAssistant.knowledge_bases]
   )
 
   return (
@@ -121,7 +134,10 @@ const InputbarInner: FC<InputbarInnerProps> = ({ assistant: initialAssistant, se
   const { setFiles, setMentionedModels, setSelectedKnowledgeBases } = useInputbarToolsDispatch()
   const { setCouldAddImageFile } = useInputbarToolsInternalDispatch()
 
-  const { text, setText } = useInputText()
+  const { text, setText } = useInputText({
+    initialValue: CacheService.get<string>(INPUTBAR_DRAFT_CACHE_KEY) ?? '',
+    onChange: (value) => CacheService.set(INPUTBAR_DRAFT_CACHE_KEY, value, DRAFT_CACHE_TTL)
+  })
   const {
     textareaRef,
     resize: resizeTextArea,
@@ -133,7 +149,6 @@ const InputbarInner: FC<InputbarInnerProps> = ({ assistant: initialAssistant, se
     minHeight: 30
   })
 
-  const showKnowledgeIcon = useSidebarIconShow('knowledge')
   const { assistant, addTopic, model, setModel, updateAssistant } = useAssistant(initialAssistant.id)
   const { sendMessageShortcut, showInputEstimatedTokens, enableQuickPanelTriggers } = useSettings()
   const [estimateTokenCount, setEstimateTokenCount] = useState(0)
@@ -190,6 +205,15 @@ const InputbarInner: FC<InputbarInnerProps> = ({ assistant: initialAssistant, se
     setCouldAddImageFile(canAddImageFile)
   }, [canAddImageFile, setCouldAddImageFile])
 
+  const onUnmount = useEffectEvent((id: string) => {
+    CacheService.set(getMentionedModelsCacheKey(id), mentionedModels, DRAFT_CACHE_TTL)
+  })
+
+  useEffect(() => {
+    return () => onUnmount(assistant.id)
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [assistant.id])
+
   const placeholderText = enableQuickPanelTriggers
     ? t('chat.input.placeholder', { key: getSendMessageShortcutLabel(sendMessageShortcut) })
     : t('chat.input.placeholder_without_triggers', {
@@ -381,9 +405,10 @@ const InputbarInner: FC<InputbarInnerProps> = ({ assistant: initialAssistant, se
     focusTextarea
   ])
 
+  // TODO: Just use assistant.knowledge_bases as selectedKnowledgeBases. context state is overdesigned.
   useEffect(() => {
-    setSelectedKnowledgeBases(showKnowledgeIcon ? (assistant.knowledge_bases ?? []) : [])
-  }, [assistant.knowledge_bases, setSelectedKnowledgeBases, showKnowledgeIcon])
+    setSelectedKnowledgeBases(assistant.knowledge_bases ?? [])
+  }, [assistant.knowledge_bases, setSelectedKnowledgeBases])
 
   useEffect(() => {
     // Disable web search if model doesn't support it

src/renderer/src/pages/home/Inputbar/components/InputbarCore.tsx

@@ -156,11 +156,8 @@ export const InputbarCore: FC<InputbarCoreProps> = ({
 
   const setText = useCallback<React.Dispatch<React.SetStateAction<string>>>(
     (value) => {
-      if (typeof value === 'function') {
-        onTextChange(value(textRef.current))
-      } else {
-        onTextChange(value)
-      }
+      const newText = typeof value === 'function' ? value(textRef.current) : value
+      onTextChange(newText)
     },
     [onTextChange]
   )

src/renderer/src/pages/home/Inputbar/tools/knowledgeBaseTool.tsx

@@ -1,5 +1,4 @@
 import { useAssistant } from '@renderer/hooks/useAssistant'
-import { useSidebarIconShow } from '@renderer/hooks/useSidebarIcon'
 import { defineTool, registerTool, TopicType } from '@renderer/pages/home/Inputbar/types'
 import type { KnowledgeBase } from '@renderer/types'
 import { isPromptToolUse, isSupportedToolUse } from '@renderer/utils/mcp-tools'
@@ -30,7 +29,6 @@ const knowledgeBaseTool = defineTool({
   render: function KnowledgeBaseToolRender(context) {
     const { assistant, state, actions, quickPanel } = context
 
-    const knowledgeSidebarEnabled = useSidebarIconShow('knowledge')
     const { updateAssistant } = useAssistant(assistant.id)
 
     const handleSelect = useCallback(
@@ -41,10 +39,6 @@ const knowledgeBaseTool = defineTool({
       [updateAssistant, actions]
     )
 
-    if (!knowledgeSidebarEnabled) {
-      return null
-    }
-
     return (
       <KnowledgeBaseButton
         quickPanel={quickPanel}

src/renderer/src/pages/home/Messages/Blocks/ThinkingBlock.tsx

@@ -102,10 +102,12 @@ const ThinkingBlock: React.FC<Props> = ({ block }) => {
   )
 }
 
+const normalizeThinkingTime = (value?: number) => (typeof value === 'number' && Number.isFinite(value) ? value : 0)
+
 const ThinkingTimeSeconds = memo(
   ({ blockThinkingTime, isThinking }: { blockThinkingTime: number; isThinking: boolean }) => {
     const { t } = useTranslation()
-    const [displayTime, setDisplayTime] = useState(blockThinkingTime)
+    const [displayTime, setDisplayTime] = useState(normalizeThinkingTime(blockThinkingTime))
 
     const timer = useRef<NodeJS.Timeout | null>(null)
 
@@ -121,7 +123,7 @@ const ThinkingTimeSeconds = memo(
           clearInterval(timer.current)
           timer.current = null
         }
-        setDisplayTime(blockThinkingTime)
+        setDisplayTime(normalizeThinkingTime(blockThinkingTime))
       }
 
       return () => {
@@ -132,10 +134,10 @@ const ThinkingTimeSeconds = memo(
       }
     }, [isThinking, blockThinkingTime])
 
-    const thinkingTimeSeconds = useMemo(
-      () => ((displayTime < 1000 ? 100 : displayTime) / 1000).toFixed(1),
-      [displayTime]
-    )
+    const thinkingTimeSeconds = useMemo(() => {
+      const safeTime = normalizeThinkingTime(displayTime)
+      return ((safeTime < 1000 ? 100 : safeTime) / 1000).toFixed(1)
+    }, [displayTime])
 
     return isThinking
       ? t('chat.thinking', {

src/renderer/src/pages/home/Messages/Blocks/__tests__/ThinkingBlock.test.tsx

@@ -255,6 +255,20 @@ describe('ThinkingBlock', () => {
         unmount()
       })
     })
+
+    it('should clamp invalid thinking times to a safe default', () => {
+      const testCases = [undefined, Number.NaN, Number.POSITIVE_INFINITY]
+
+      testCases.forEach((thinking_millsec) => {
+        const block = createThinkingBlock({
+          thinking_millsec: thinking_millsec as any,
+          status: MessageBlockStatus.SUCCESS
+        })
+        const { unmount } = renderThinkingBlock(block)
+        expect(getThinkingTimeText()).toHaveTextContent('0.1s')
+        unmount()
+      })
+    })
   })
 
   describe('collapse behavior', () => {

src/renderer/src/pages/home/Messages/ChatNavigation.tsx

@@ -10,6 +10,7 @@ import { useSettings } from '@renderer/hooks/useSettings'
 import { useTimer } from '@renderer/hooks/useTimer'
 import type { RootState } from '@renderer/store'
 // import { selectCurrentTopicId } from '@renderer/store/newMessage'
+import { scrollIntoView } from '@renderer/utils/dom'
 import { Button, Drawer, Tooltip } from 'antd'
 import type { FC } from 'react'
 import { useCallback, useEffect, useRef, useState } from 'react'
@@ -118,7 +119,8 @@ const ChatNavigation: FC<ChatNavigationProps> = ({ containerId }) => {
   }
 
   const scrollToMessage = (element: HTMLElement) => {
-    element.scrollIntoView({ behavior: 'smooth', block: 'start' })
+    // Use container: 'nearest' to keep scroll within the chat pane (Chromium-only, see #11565, #11567)
+    scrollIntoView(element, { behavior: 'smooth', block: 'start', container: 'nearest' })
   }
 
   const scrollToTop = () => {

src/renderer/src/pages/home/Messages/Message.tsx

@@ -15,6 +15,7 @@ import { estimateMessageUsage } from '@renderer/services/TokenService'
 import type { Assistant, Topic } from '@renderer/types'
 import type { Message, MessageBlock } from '@renderer/types/newMessage'
 import { classNames, cn } from '@renderer/utils'
+import { scrollIntoView } from '@renderer/utils/dom'
 import { isMessageProcessing } from '@renderer/utils/messageUtils/is'
 import { Divider } from 'antd'
 import type { Dispatch, FC, SetStateAction } from 'react'
@@ -79,9 +80,10 @@ const MessageItem: FC<Props> = ({
 
   useEffect(() => {
     if (isEditing && messageContainerRef.current) {
-      messageContainerRef.current.scrollIntoView({
+      scrollIntoView(messageContainerRef.current, {
         behavior: 'smooth',
-        block: 'center'
+        block: 'center',
+        container: 'nearest'
       })
     }
   }, [isEditing])
@@ -124,7 +126,7 @@ const MessageItem: FC<Props> = ({
   const messageHighlightHandler = useCallback(
     (highlight: boolean = true) => {
       if (messageContainerRef.current) {
-        messageContainerRef.current.scrollIntoView({ behavior: 'smooth' })
+        scrollIntoView(messageContainerRef.current, { behavior: 'smooth', block: 'center', container: 'nearest' })
         if (highlight) {
           setTimeoutTimer(
             'messageHighlightHandler',

src/renderer/src/pages/home/Messages/MessageAnchorLine.tsx

@@ -12,6 +12,7 @@ import { newMessagesActions } from '@renderer/store/newMessage'
 // import { updateMessageThunk } from '@renderer/store/thunk/messageThunk'
 import type { Message } from '@renderer/types/newMessage'
 import { isEmoji, removeLeadingEmoji } from '@renderer/utils'
+import { scrollIntoView } from '@renderer/utils/dom'
 import { getMainTextContent } from '@renderer/utils/messageUtils/find'
 import { Avatar } from 'antd'
 import { CircleChevronDown } from 'lucide-react'
@@ -119,7 +120,7 @@ const MessageAnchorLine: FC<MessageLineProps> = ({ messages }) => {
           () => {
             const messageElement = document.getElementById(`message-${message.id}`)
             if (messageElement) {
-              messageElement.scrollIntoView({ behavior: 'auto', block: 'start' })
+              scrollIntoView(messageElement, { behavior: 'auto', block: 'start', container: 'nearest' })
             }
           },
           100
@@ -141,7 +142,7 @@ const MessageAnchorLine: FC<MessageLineProps> = ({ messages }) => {
         return
       }
 
-      messageElement.scrollIntoView({ behavior: 'smooth', block: 'start' })
+      scrollIntoView(messageElement, { behavior: 'smooth', block: 'start', container: 'nearest' })
     },
     [setSelectedMessage]
   )

src/renderer/src/pages/home/Messages/MessageGroup.tsx

@@ -10,6 +10,7 @@ import type { MultiModelMessageStyle } from '@renderer/store/settings'
 import type { Topic } from '@renderer/types'
 import type { Message } from '@renderer/types/newMessage'
 import { classNames } from '@renderer/utils'
+import { scrollIntoView } from '@renderer/utils/dom'
 import { Popover } from 'antd'
 import type { ComponentProps } from 'react'
 import { memo, useCallback, useEffect, useMemo, useState } from 'react'
@@ -73,7 +74,7 @@ const MessageGroup = ({ messages, topic, registerMessageElement }: Props) => {
         () => {
           const messageElement = document.getElementById(`message-${message.id}`)
           if (messageElement) {
-            messageElement.scrollIntoView({ behavior: 'smooth', block: 'start' })
+            scrollIntoView(messageElement, { behavior: 'smooth', block: 'start', container: 'nearest' })
           }
         },
         200
@@ -132,7 +133,7 @@ const MessageGroup = ({ messages, topic, registerMessageElement }: Props) => {
             setSelectedMessage(message)
           } else {
             // 直接滚动
-            element.scrollIntoView({ behavior: 'smooth', block: 'start' })
+            scrollIntoView(element, { behavior: 'smooth', block: 'start', container: 'nearest' })
           }
         }
       }

src/renderer/src/pages/home/Messages/MessageOutline.tsx

@@ -3,6 +3,7 @@ import type { RootState } from '@renderer/store'
 import { messageBlocksSelectors } from '@renderer/store/messageBlock'
 import type { Message } from '@renderer/types/newMessage'
 import { MessageBlockType } from '@renderer/types/newMessage'
+import { scrollIntoView } from '@renderer/utils/dom'
 import type { FC } from 'react'
 import React, { useMemo, useRef } from 'react'
 import { useSelector } from 'react-redux'
@@ -72,10 +73,10 @@ const MessageOutline: FC<MessageOutlineProps> = ({ message }) => {
     const parent = messageOutlineContainerRef.current?.parentElement
     const messageContentContainer = parent?.querySelector('.message-content-container')
     if (messageContentContainer) {
-      const headingElement = messageContentContainer.querySelector(`#${id}`)
+      const headingElement = messageContentContainer.querySelector<HTMLElement>(`#${id}`)
       if (headingElement) {
         const scrollBlock = ['horizontal', 'grid'].includes(message.multiModelMessageStyle ?? '') ? 'nearest' : 'start'
-        headingElement.scrollIntoView({ behavior: 'smooth', block: scrollBlock })
+        scrollIntoView(headingElement, { behavior: 'smooth', block: scrollBlock, container: 'nearest' })
       }
     }
   }

src/renderer/src/pages/home/Messages/Tools/MessageAgentTools/BashTool.tsx

@@ -5,8 +5,6 @@ import { Terminal } from 'lucide-react'
 import { ToolTitle } from './GenericTools'
 import type { BashToolInput as BashToolInputType, BashToolOutput as BashToolOutputType } from './types'
 
-const MAX_TAG_LENGTH = 100
-
 export function BashTool({
   input,
   output
@@ -17,12 +15,10 @@ export function BashTool({
   // 如果有输出，计算输出行数
   const outputLines = output ? output.split('\n').length : 0
 
-  // 处理命令字符串的截断，添加空值检查
+  // 处理命令字符串，添加空值检查
   const command = input?.command ?? ''
-  const needsTruncate = command.length > MAX_TAG_LENGTH
-  const displayCommand = needsTruncate ? `${command.slice(0, MAX_TAG_LENGTH)}...` : command
 
-  const tagContent = <Tag className="whitespace-pre-wrap break-all font-mono">{displayCommand}</Tag>
+  const tagContent = <Tag className="!m-0 max-w-full truncate font-mono">{command}</Tag>
 
   return {
     key: 'tool',
@@ -34,16 +30,12 @@ export function BashTool({
           params={input?.description}
           stats={output ? `${outputLines} ${outputLines === 1 ? 'line' : 'lines'}` : undefined}
         />
-        <div className="mt-1">
-          {needsTruncate ? (
-            <Popover
-              content={<div className="max-w-xl whitespace-pre-wrap break-all font-mono">{command}</div>}
-              trigger="hover">
-              {tagContent}
-            </Popover>
-          ) : (
-            tagContent
-          )}
+        <div className="mt-1 max-w-full">
+          <Popover
+            content={<div className="max-w-xl whitespace-pre-wrap break-all font-mono text-xs">{command}</div>}
+            trigger="hover">
+            {tagContent}
+          </Popover>
         </div>
       </>
     ),

src/renderer/src/pages/home/Messages/Tools/MessageAgentTools/GenericTools.tsx

@@ -18,9 +18,9 @@ export function ToolTitle({
 }) {
   return (
     <div className={`flex items-center gap-1 ${className}`}>
-      {icon}
-      {label && <span className="font-medium text-sm">{label}</span>}
-      {params && <span className="flex-shrink-0 text-muted-foreground text-xs">{params}</span>}
+      {icon && <span className="flex flex-shrink-0">{icon}</span>}
+      {label && <span className="flex-shrink-0 font-medium text-sm">{label}</span>}
+      {params && <span className="min-w-0 truncate text-muted-foreground text-xs">{params}</span>}
       {stats && <span className="flex-shrink-0 text-muted-foreground text-xs">{stats}</span>}
     </div>
   )

src/renderer/src/pages/home/Messages/Tools/MessageAgentTools/index.tsx

@@ -1,7 +1,10 @@
 import { loggerService } from '@logger'
+import { useAppSelector } from '@renderer/store'
+import { selectPendingPermission } from '@renderer/store/toolPermissions'
 import type { NormalToolResponse } from '@renderer/types'
 import type { CollapseProps } from 'antd'
-import { Collapse } from 'antd'
+import { Collapse, Spin } from 'antd'
+import { useTranslation } from 'react-i18next'
 
 // 导出所有类型
 export * from './types'
@@ -83,17 +86,41 @@ function ToolContent({ toolName, input, output }: { toolName: AgentToolsType; in
 // 统一的组件渲染入口
 export function MessageAgentTools({ toolResponse }: { toolResponse: NormalToolResponse }) {
   const { arguments: args, response, tool, status } = toolResponse
-  logger.info('Rendering agent tool response', {
+  logger.debug('Rendering agent tool response', {
     tool: tool,
     arguments: args,
+    status,
     response
   })
 
+  const pendingPermission = useAppSelector((state) =>
+    selectPendingPermission(state.toolPermissions, toolResponse.toolCallId)
+  )
+
   if (status === 'pending') {
-    return <ToolPermissionRequestCard toolResponse={toolResponse} />
+    if (pendingPermission) {
+      return <ToolPermissionRequestCard toolResponse={toolResponse} />
+    }
+    return <ToolPendingIndicator toolName={tool?.name} description={tool?.description} />
   }
 
   return (
     <ToolContent toolName={tool.name as AgentToolsType} input={args as ToolInput} output={response as ToolOutput} />
   )
 }
+
+function ToolPendingIndicator({ toolName, description }: { toolName?: string; description?: string }) {
+  const { t } = useTranslation()
+  const label = toolName || t('agent.toolPermission.toolPendingFallback', 'Tool')
+  const detail = description?.trim() || t('agent.toolPermission.executing')
+
+  return (
+    <div className="flex w-full max-w-xl items-center gap-3 rounded-xl border border-default-200 bg-default-100 px-4 py-3 shadow-sm">
+      <Spin size="small" />
+      <div className="flex flex-col gap-1">
+        <span className="font-semibold text-default-700 text-sm">{label}</span>
+        <span className="text-default-500 text-xs">{detail}</span>
+      </div>
+    </div>
+  )
+}

src/renderer/src/pages/settings/ProviderSettings/ProviderSetting.tsx

@@ -1,8 +1,10 @@
+import { adaptProvider } from '@renderer/aiCore/provider/providerConfig'
 import OpenAIAlert from '@renderer/components/Alert/OpenAIAlert'
 import { LoadingIcon } from '@renderer/components/Icons'
 import { HStack } from '@renderer/components/Layout'
 import { ApiKeyListPopup } from '@renderer/components/Popups/ApiKeyListPopup'
 import Selector from '@renderer/components/Selector'
+import { HelpTooltip } from '@renderer/components/TooltipIcons'
 import { isEmbeddingModel, isRerankModel } from '@renderer/config/models'
 import { PROVIDER_URLS } from '@renderer/config/providers'
 import { useTheme } from '@renderer/context/ThemeProvider'
@@ -19,14 +21,7 @@ import type { SystemProviderId } from '@renderer/types'
 import { isSystemProvider, isSystemProviderId, SystemProviderIds } from '@renderer/types'
 import type { ApiKeyConnectivity } from '@renderer/types/healthCheck'
 import { HealthStatus } from '@renderer/types/healthCheck'
-import {
-  formatApiHost,
-  formatApiKeys,
-  formatAzureOpenAIApiHost,
-  formatVertexApiHost,
-  getFancyProviderName,
-  validateApiHost
-} from '@renderer/utils'
+import { formatApiHost, formatApiKeys, getFancyProviderName, validateApiHost } from '@renderer/utils'
 import { formatErrorMessage } from '@renderer/utils/error'
 import {
   isAIGatewayProvider,
@@ -36,7 +31,6 @@ import {
   isNewApiProvider,
   isOpenAICompatibleProvider,
   isOpenAIProvider,
-  isSupportAPIVersionProvider,
   isVertexProvider
 } from '@renderer/utils/provider'
 import { Button, Divider, Flex, Input, Select, Space, Switch, Tooltip } from 'antd'
@@ -281,12 +275,10 @@ const ProviderSetting: FC<Props> = ({ providerId }) => {
   }, [configuredApiHost, apiHost])
 
   const hostPreview = () => {
-    if (apiHost.endsWith('#')) {
-      return apiHost.replace('#', '')
-    }
+    const formattedApiHost = adaptProvider({ provider: { ...provider, apiHost } }).apiHost
 
     if (isOpenAICompatibleProvider(provider)) {
-      return formatApiHost(apiHost, isSupportAPIVersionProvider(provider)) + '/chat/completions'
+      return formattedApiHost + '/chat/completions'
     }
 
     if (isAzureOpenAIProvider(provider)) {
@@ -294,29 +286,26 @@ const ProviderSetting: FC<Props> = ({ providerId }) => {
       const path = !['preview', 'v1'].includes(apiVersion)
         ? `/v1/chat/completion?apiVersion=v1`
         : `/v1/responses?apiVersion=v1`
-      return formatAzureOpenAIApiHost(apiHost) + path
+      return formattedApiHost + path
     }
 
     if (isAnthropicProvider(provider)) {
-      // AI SDK uses the baseURL with /v1, then appends /messages
-      // formatApiHost adds /v1 automatically if not present
-      const normalizedHost = formatApiHost(apiHost)
-      return normalizedHost + '/messages'
+      return formattedApiHost + '/messages'
     }
 
     if (isGeminiProvider(provider)) {
-      return formatApiHost(apiHost, true, 'v1beta') + '/models'
+      return formattedApiHost + '/models'
     }
     if (isOpenAIProvider(provider)) {
-      return formatApiHost(apiHost) + '/responses'
+      return formattedApiHost + '/responses'
     }
     if (isVertexProvider(provider)) {
-      return formatVertexApiHost(provider) + '/publishers/google'
+      return formattedApiHost + '/publishers/google'
     }
     if (isAIGatewayProvider(provider)) {
-      return formatApiHost(apiHost) + '/language-model'
+      return formattedApiHost + '/language-model'
     }
-    return formatApiHost(apiHost)
+    return formattedApiHost
   }
 
   // API key 连通性检查状态指示器，目前仅在失败时显示
@@ -494,16 +483,21 @@ const ProviderSetting: FC<Props> = ({ providerId }) => {
           {!isDmxapi && (
             <>
               <SettingSubtitle style={{ display: 'flex', alignItems: 'center', justifyContent: 'space-between' }}>
-                <Tooltip title={hostSelectorTooltip} mouseEnterDelay={0.3}>
-                  <Selector
-                    size={14}
-                    value={activeHostField}
-                    onChange={(value) => setActiveHostField(value as HostField)}
-                    options={hostSelectorOptions}
-                    style={{ paddingLeft: 1, fontWeight: 'bold' }}
-                    placement="bottomLeft"
-                  />
-                </Tooltip>
+                <div className="flex items-center gap-1">
+                  <Tooltip title={hostSelectorTooltip} mouseEnterDelay={0.3}>
+                    <div>
+                      <Selector
+                        size={14}
+                        value={activeHostField}
+                        onChange={(value) => setActiveHostField(value as HostField)}
+                        options={hostSelectorOptions}
+                        style={{ paddingLeft: 1, fontWeight: 'bold' }}
+                        placement="bottomLeft"
+                      />
+                    </div>
+                  </Tooltip>
+                  <HelpTooltip title={t('settings.provider.api.url.tip')}></HelpTooltip>
+                </div>
                 <div style={{ display: 'flex', alignItems: 'center', gap: 4 }}>
                   <Button
                     type="text"

src/renderer/src/services/ApiService.ts

@@ -8,11 +8,11 @@ import { isDedicatedImageGenerationModel, isEmbeddingModel, isFunctionCallingMod
 import { getStoreSetting } from '@renderer/hooks/useSettings'
 import i18n from '@renderer/i18n'
 import store from '@renderer/store'
-import type { FetchChatCompletionParams } from '@renderer/types'
 import type { Assistant, MCPServer, MCPTool, Model, Provider } from '@renderer/types'
+import { type FetchChatCompletionParams, isSystemProvider } from '@renderer/types'
 import type { StreamTextParams } from '@renderer/types/aiCoreTypes'
 import { type Chunk, ChunkType } from '@renderer/types/chunk'
-import type { Message } from '@renderer/types/newMessage'
+import type { Message, ResponseError } from '@renderer/types/newMessage'
 import type { SdkModel } from '@renderer/types/sdk'
 import { removeSpecialCharactersForTopicName, uuid } from '@renderer/utils'
 import { abortCompletion, readyToAbort } from '@renderer/utils/abortController'
@@ -22,7 +22,8 @@ import { purifyMarkdownImages } from '@renderer/utils/markdown'
 import { isPromptToolUse, isSupportedToolUse } from '@renderer/utils/mcp-tools'
 import { findFileBlocks, getMainTextContent } from '@renderer/utils/messageUtils/find'
 import { containsSupportedVariables, replacePromptVariables } from '@renderer/utils/prompt'
-import { isEmpty, takeRight } from 'lodash'
+import { NOT_SUPPORT_API_KEY_PROVIDERS } from '@renderer/utils/provider'
+import { cloneDeep, isEmpty, takeRight } from 'lodash'
 
 import type { ModernAiProviderConfig } from '../aiCore/index_new'
 import AiProviderNew from '../aiCore/index_new'
@@ -43,6 +44,8 @@ import {
 // } from './MessagesService'
 // import WebSearchService from './WebSearchService'
 
+// FIXME: 这里太多重复逻辑，需要重构
+
 const logger = loggerService.withContext('ApiService')
 
 export async function fetchMcpTools(assistant: Assistant) {
@@ -95,7 +98,15 @@ export async function fetchChatCompletion({
     modelId: assistant.model?.id,
     modelName: assistant.model?.name
   })
-  const AI = new AiProviderNew(assistant.model || getDefaultModel())
+
+  // Get base provider and apply API key rotation
+  const baseProvider = getProviderByModel(assistant.model || getDefaultModel())
+  const providerWithRotatedKey = {
+    ...cloneDeep(baseProvider),
+    apiKey: getRotatedApiKey(baseProvider)
+  }
+
+  const AI = new AiProviderNew(assistant.model || getDefaultModel(), providerWithRotatedKey)
   const provider = AI.getActualProvider()
 
   const mcpTools: MCPTool[] = []
@@ -172,7 +183,13 @@ export async function fetchMessagesSummary({ messages, assistant }: { messages:
     return null
   }
 
-  const AI = new AiProviderNew(model)
+  // Apply API key rotation
+  const providerWithRotatedKey = {
+    ...cloneDeep(provider),
+    apiKey: getRotatedApiKey(provider)
+  }
+
+  const AI = new AiProviderNew(model, providerWithRotatedKey)
 
   const topicId = messages?.find((message) => message.topicId)?.topicId || ''
 
@@ -271,7 +288,13 @@ export async function fetchNoteSummary({ content, assistant }: { content: string
     return null
   }
 
-  const AI = new AiProviderNew(model)
+  // Apply API key rotation
+  const providerWithRotatedKey = {
+    ...cloneDeep(provider),
+    apiKey: getRotatedApiKey(provider)
+  }
+
+  const AI = new AiProviderNew(model, providerWithRotatedKey)
 
   // only 2000 char and no images
   const truncatedContent = content.substring(0, 2000)
@@ -359,7 +382,13 @@ export async function fetchGenerate({
     return ''
   }
 
-  const AI = new AiProviderNew(model)
+  // Apply API key rotation
+  const providerWithRotatedKey = {
+    ...cloneDeep(provider),
+    apiKey: getRotatedApiKey(provider)
+  }
+
+  const AI = new AiProviderNew(model, providerWithRotatedKey)
 
   const assistant = getDefaultAssistant()
   assistant.model = model
@@ -404,28 +433,44 @@ export async function fetchGenerate({
 
 export function hasApiKey(provider: Provider) {
   if (!provider) return false
-  if (['ollama', 'lmstudio', 'vertexai', 'cherryai'].includes(provider.id)) return true
+  if (isSystemProvider(provider) && NOT_SUPPORT_API_KEY_PROVIDERS.includes(provider.id)) return true
   return !isEmpty(provider.apiKey)
 }
 
 /**
- * Get the first available embedding model from enabled providers
+ * 获取轮询的API key
+ * 复用legacy架构的多key轮询逻辑
  */
-// function getFirstEmbeddingModel() {
-//   const providers = store.getState().llm.providers.filter((p) => p.enabled)
+function getRotatedApiKey(provider: Provider): string {
+  const keys = provider.apiKey.split(',').map((key) => key.trim())
+  const keyName = `provider:${provider.id}:last_used_key`
 
-//   for (const provider of providers) {
-//     const embeddingModel = provider.models.find((model) => isEmbeddingModel(model))
-//     if (embeddingModel) {
-//       return embeddingModel
-//     }
-//   }
+  if (keys.length === 1) {
+    return keys[0]
+  }
 
-//   return undefined
-// }
+  const lastUsedKey = window.keyv.get(keyName)
+  if (!lastUsedKey) {
+    window.keyv.set(keyName, keys[0])
+    return keys[0]
+  }
+
+  const currentIndex = keys.indexOf(lastUsedKey)
+  const nextIndex = (currentIndex + 1) % keys.length
+  const nextKey = keys[nextIndex]
+  window.keyv.set(keyName, nextKey)
+
+  return nextKey
+}
 
 export async function fetchModels(provider: Provider): Promise<SdkModel[]> {
-  const AI = new AiProviderNew(provider)
+  // Apply API key rotation
+  const providerWithRotatedKey = {
+    ...cloneDeep(provider),
+    apiKey: getRotatedApiKey(provider)
+  }
+
+  const AI = new AiProviderNew(providerWithRotatedKey)
 
   try {
     return await AI.models()
@@ -435,12 +480,7 @@ export async function fetchModels(provider: Provider): Promise<SdkModel[]> {
 }
 
 export function checkApiProvider(provider: Provider): void {
-  if (
-    provider.id !== 'ollama' &&
-    provider.id !== 'lmstudio' &&
-    provider.type !== 'vertexai' &&
-    provider.id !== 'copilot'
-  ) {
+  if (isSystemProvider(provider) && !NOT_SUPPORT_API_KEY_PROVIDERS.includes(provider.id)) {
     if (!provider.apiKey) {
       window.toast.error(i18n.t('message.error.enter.api.label'))
       throw new Error(i18n.t('message.error.enter.api.label'))
@@ -461,8 +501,7 @@ export function checkApiProvider(provider: Provider): void {
 export async function checkApi(provider: Provider, model: Model, timeout = 15000): Promise<void> {
   checkApiProvider(provider)
 
-  // Don't pass in provider parameter. We need auto-format URL
-  const ai = new AiProviderNew(model)
+  const ai = new AiProviderNew(model, provider)
 
   const assistant = getDefaultAssistant()
   assistant.model = model
@@ -476,7 +515,7 @@ export async function checkApi(provider: Provider, model: Model, timeout = 15000
   } else {
     const abortId = uuid()
     const signal = readyToAbort(abortId)
-    let chunkError
+    let streamError: ResponseError | undefined
     const params: StreamTextParams = {
       system: assistant.prompt,
       prompt: 'hi',
@@ -495,19 +534,18 @@ export async function checkApi(provider: Provider, model: Model, timeout = 15000
       callType: 'check',
       onChunk: (chunk: Chunk) => {
         if (chunk.type === ChunkType.ERROR) {
-          chunkError = chunk.error
+          streamError = chunk.error
         } else {
           abortCompletion(abortId)
         }
       }
     }
 
-    // Try streaming check
     try {
       await ai.completions(model.id, params, config)
     } catch (e) {
-      if (!isAbortError(e) && !isAbortError(chunkError)) {
-        throw e
+      if (!isAbortError(e) && !isAbortError(streamError)) {
+        throw streamError ?? e
       }
     }
   }

src/renderer/src/types/index.ts

@@ -239,6 +239,7 @@ export type ModelType = 'text' | 'vision' | 'embedding' | 'reasoning' | 'functio
 
 export type ModelTag = Exclude<ModelType, 'text'> | 'free'
 
+// "image-generation" is also openai endpoint, but specifically for image generation.
 export type EndpointType = 'openai' | 'openai-response' | 'anthropic' | 'gemini' | 'image-generation' | 'jina-rerank'
 
 export type ModelPricing = {

src/renderer/src/types/newMessage.ts

@@ -234,6 +234,7 @@ export interface Response {
   error?: ResponseError
 }
 
+// FIXME: Weak type safety. It may be a specific class instance which inherits Error in runtime.
 export type ResponseError = Record<string, any>
 
 export interface MessageInputBaseParams {

src/renderer/src/utils/__tests__/api.test.ts

@@ -13,7 +13,8 @@ import {
   routeToEndpoint,
   splitApiKeyString,
   validateApiHost,
-  withoutTrailingApiVersion
+  withoutTrailingApiVersion,
+  withoutTrailingSharp
 } from '../api'
 
 vi.mock('@renderer/store', () => {
@@ -81,6 +82,27 @@ describe('api', () => {
     it('keeps host untouched when api version unsupported', () => {
       expect(formatApiHost('https://api.example.com', false)).toBe('https://api.example.com')
     })
+
+    it('removes trailing # and does not append api version when host ends with #', () => {
+      expect(formatApiHost('https://api.example.com#')).toBe('https://api.example.com')
+      expect(formatApiHost('http://localhost:5173/#')).toBe('http://localhost:5173/')
+      expect(formatApiHost(' https://api.openai.com/# ')).toBe('https://api.openai.com/')
+    })
+
+    it('handles trailing # with custom api version settings', () => {
+      expect(formatApiHost('https://api.example.com#', true, 'v2')).toBe('https://api.example.com')
+      expect(formatApiHost('https://api.example.com#', false, 'v2')).toBe('https://api.example.com')
+    })
+
+    it('handles host with both trailing # and existing api version', () => {
+      expect(formatApiHost('https://api.example.com/v2#')).toBe('https://api.example.com/v2')
+      expect(formatApiHost('https://api.example.com/v3beta#')).toBe('https://api.example.com/v3beta')
+    })
+
+    it('trims whitespace before processing trailing #', () => {
+      expect(formatApiHost('  https://api.example.com#  ')).toBe('https://api.example.com')
+      expect(formatApiHost('\thttps://api.example.com#\n')).toBe('https://api.example.com')
+    })
   })
 
   describe('hasAPIVersion', () => {
@@ -404,4 +426,56 @@ describe('api', () => {
       expect(withoutTrailingApiVersion('')).toBe('')
     })
   })
+
+  describe('withoutTrailingSharp', () => {
+    it('removes trailing # from URL', () => {
+      expect(withoutTrailingSharp('https://api.example.com#')).toBe('https://api.example.com')
+      expect(withoutTrailingSharp('http://localhost:3000#')).toBe('http://localhost:3000')
+    })
+
+    it('returns URL unchanged when no trailing #', () => {
+      expect(withoutTrailingSharp('https://api.example.com')).toBe('https://api.example.com')
+      expect(withoutTrailingSharp('http://localhost:3000')).toBe('http://localhost:3000')
+    })
+
+    it('handles URLs with multiple # characters but only removes trailing one', () => {
+      expect(withoutTrailingSharp('https://api.example.com#path#')).toBe('https://api.example.com#path')
+    })
+
+    it('handles URLs with # in the middle (not trailing)', () => {
+      expect(withoutTrailingSharp('https://api.example.com#section/path')).toBe('https://api.example.com#section/path')
+      expect(withoutTrailingSharp('https://api.example.com/v1/chat/completions#')).toBe(
+        'https://api.example.com/v1/chat/completions'
+      )
+    })
+
+    it('handles empty string', () => {
+      expect(withoutTrailingSharp('')).toBe('')
+    })
+
+    it('handles single character #', () => {
+      expect(withoutTrailingSharp('#')).toBe('')
+    })
+
+    it('preserves whitespace around the URL (pure function)', () => {
+      expect(withoutTrailingSharp('  https://api.example.com#  ')).toBe('  https://api.example.com#  ')
+      expect(withoutTrailingSharp('\thttps://api.example.com#\n')).toBe('\thttps://api.example.com#\n')
+    })
+
+    it('only removes exact trailing # character', () => {
+      expect(withoutTrailingSharp('https://api.example.com# ')).toBe('https://api.example.com# ')
+      expect(withoutTrailingSharp(' https://api.example.com#')).toBe(' https://api.example.com')
+      expect(withoutTrailingSharp('https://api.example.com#\t')).toBe('https://api.example.com#\t')
+    })
+
+    it('handles URLs ending with multiple # characters', () => {
+      expect(withoutTrailingSharp('https://api.example.com##')).toBe('https://api.example.com#')
+      expect(withoutTrailingSharp('https://api.example.com###')).toBe('https://api.example.com##')
+    })
+
+    it('preserves URL with trailing # and other content', () => {
+      expect(withoutTrailingSharp('https://api.example.com/v1#')).toBe('https://api.example.com/v1')
+      expect(withoutTrailingSharp('https://api.example.com/v2beta#')).toBe('https://api.example.com/v2beta')
+    })
+  })
 })

src/renderer/src/utils/api.ts

@@ -62,6 +62,23 @@ export function withoutTrailingSlash<T extends string>(url: T): T {
   return url.replace(/\/$/, '') as T
 }
 
+/**
+ * Removes the trailing '#' from a URL string if it exists.
+ *
+ * @template T - The string type to preserve type safety
+ * @param {T} url - The URL string to process
+ * @returns {T} The URL string without a trailing '#'
+ *
+ * @example
+ * ```ts
+ * withoutTrailingSharp('https://example.com#') // 'https://example.com'
+ * withoutTrailingSharp('https://example.com')  // 'https://example.com'
+ * ```
+ */
+export function withoutTrailingSharp<T extends string>(url: T): T {
+  return url.replace(/#$/, '') as T
+}
+
 /**
  * Formats an API host URL by normalizing it and optionally appending an API version.
  *
@@ -70,12 +87,12 @@ export function withoutTrailingSlash<T extends string>(url: T): T {
  * @param apiVersion - The API version to append if needed. Defaults to `'v1'`.
  *
  * @returns The formatted API host URL. If the host is empty after normalization, returns an empty string.
- *          If the host ends with '#', API version is not supported, or the host already contains a version, returns the normalized host as-is.
+ *          If the host ends with '#', API version is not supported, or the host already contains a version, returns the normalized host with trailing '#' removed.
  *          Otherwise, returns the host with the API version appended.
  *
  * @example
  * formatApiHost('https://api.example.com/') // Returns 'https://api.example.com/v1'
- * formatApiHost('https://api.example.com#') // Returns 'https://api.example.com#'
+ * formatApiHost('https://api.example.com#') // Returns 'https://api.example.com'
  * formatApiHost('https://api.example.com/v2', true, 'v1') // Returns 'https://api.example.com/v2'
  */
 export function formatApiHost(host?: string, supportApiVersion: boolean = true, apiVersion: string = 'v1'): string {
@@ -84,10 +101,13 @@ export function formatApiHost(host?: string, supportApiVersion: boolean = true,
     return ''
   }
 
-  if (normalizedHost.endsWith('#') || !supportApiVersion || hasAPIVersion(normalizedHost)) {
-    return normalizedHost
+  const shouldAppendApiVersion = !(normalizedHost.endsWith('#') || !supportApiVersion || hasAPIVersion(normalizedHost))
+
+  if (shouldAppendApiVersion) {
+    return `${normalizedHost}/${apiVersion}`
+  } else {
+    return withoutTrailingSharp(normalizedHost)
   }
-  return `${normalizedHost}/${apiVersion}`
 }
 
 /**

src/renderer/src/utils/dom.ts

@@ -1,3 +1,15 @@
+import { loggerService } from '@logger'
+
+const logger = loggerService.withContext('utils/dom')
+
+interface ChromiumScrollIntoViewOptions extends ScrollIntoViewOptions {
+  /**
+   * @see https://developer.mozilla.org/en-US/docs/Web/API/Element/scrollIntoView#container
+   * @see https://github.com/microsoft/TypeScript/issues/62803
+   */
+  container?: 'all' | 'nearest'
+}
+
 /**
  * Simple wrapper for scrollIntoView with common default options.
  * Provides a unified interface with sensible defaults.
@@ -5,7 +17,12 @@
  * @param element - The target element to scroll into view
  * @param options - Scroll options. If not provided, uses { behavior: 'smooth', block: 'center', inline: 'nearest' }
  */
-export function scrollIntoView(element: HTMLElement, options?: ScrollIntoViewOptions): void {
+export function scrollIntoView(element: HTMLElement, options?: ChromiumScrollIntoViewOptions): void {
+  if (!element) {
+    logger.warn('[scrollIntoView] Unexpected falsy element. Do nothing as fallback.')
+    return
+  }
+
   const defaultOptions: ScrollIntoViewOptions = {
     behavior: 'smooth',
     block: 'center',

src/renderer/src/utils/provider.ts

@@ -183,3 +183,11 @@ export const isSupportAPIVersionProvider = (provider: Provider) => {
   }
   return provider.apiOptions?.isNotSupportAPIVersion !== false
 }
+
+export const NOT_SUPPORT_API_KEY_PROVIDERS: readonly SystemProviderId[] = [
+  'ollama',
+  'lmstudio',
+  'vertexai',
+  'aws-bedrock',
+  'copilot'
+]

src/renderer/src/windows/mini/home/HomeWindow.tsx

@@ -254,6 +254,17 @@ const HomeWindow: FC<{ draggable?: boolean }> = ({ draggable = true }) => {
 
         let blockId: string | null = null
         let thinkingBlockId: string | null = null
+        let thinkingStartTime: number | null = null
+
+        const resolveThinkingDuration = (duration?: number) => {
+          if (typeof duration === 'number' && Number.isFinite(duration)) {
+            return duration
+          }
+          if (thinkingStartTime !== null) {
+            return Math.max(0, performance.now() - thinkingStartTime)
+          }
+          return 0
+        }
 
         setIsLoading(true)
         setIsOutputted(false)
@@ -291,6 +302,7 @@ const HomeWindow: FC<{ draggable?: boolean }> = ({ draggable = true }) => {
               case ChunkType.THINKING_START:
                 {
                   setIsOutputted(true)
+                  thinkingStartTime = performance.now()
                   if (thinkingBlockId) {
                     store.dispatch(
                       updateOneBlock({ id: thinkingBlockId, changes: { status: MessageBlockStatus.STREAMING } })
@@ -315,9 +327,13 @@ const HomeWindow: FC<{ draggable?: boolean }> = ({ draggable = true }) => {
                 {
                   setIsOutputted(true)
                   if (thinkingBlockId) {
+                    if (thinkingStartTime === null) {
+                      thinkingStartTime = performance.now()
+                    }
+                    const thinkingDuration = resolveThinkingDuration(chunk.thinking_millsec)
                     throttledBlockUpdate(thinkingBlockId, {
                       content: chunk.text,
-                      thinking_millsec: chunk.thinking_millsec
+                      thinking_millsec: thinkingDuration
                     })
                   }
                 }
@@ -325,14 +341,17 @@ const HomeWindow: FC<{ draggable?: boolean }> = ({ draggable = true }) => {
               case ChunkType.THINKING_COMPLETE:
                 {
                   if (thinkingBlockId) {
+                    const thinkingDuration = resolveThinkingDuration(chunk.thinking_millsec)
                     cancelThrottledBlockUpdate(thinkingBlockId)
                     store.dispatch(
                       updateOneBlock({
                         id: thinkingBlockId,
-                        changes: { status: MessageBlockStatus.SUCCESS, thinking_millsec: chunk.thinking_millsec }
+                        changes: { status: MessageBlockStatus.SUCCESS, thinking_millsec: thinkingDuration }
                       })
                     )
                   }
+                  thinkingStartTime = null
+                  thinkingBlockId = null
                 }
                 break
               case ChunkType.TEXT_START:
@@ -404,6 +423,8 @@ const HomeWindow: FC<{ draggable?: boolean }> = ({ draggable = true }) => {
                 if (!isAborted) {
                   throw new Error(chunk.error.message)
                 }
+                thinkingStartTime = null
+                thinkingBlockId = null
               }
               //fall through
               case ChunkType.BLOCK_COMPLETE:

src/renderer/src/windows/selection/action/components/ActionUtils.ts

@@ -41,8 +41,19 @@ export const processMessages = async (
 
     let textBlockId: string | null = null
     let thinkingBlockId: string | null = null
+    let thinkingStartTime: number | null = null
     let textBlockContent: string = ''
 
+    const resolveThinkingDuration = (duration?: number) => {
+      if (typeof duration === 'number' && Number.isFinite(duration)) {
+        return duration
+      }
+      if (thinkingStartTime !== null) {
+        return Math.max(0, performance.now() - thinkingStartTime)
+      }
+      return 0
+    }
+
     const assistantMessage = getAssistantMessage({
       assistant,
       topic
@@ -79,6 +90,7 @@ export const processMessages = async (
         switch (chunk.type) {
           case ChunkType.THINKING_START:
             {
+              thinkingStartTime = performance.now()
               if (thinkingBlockId) {
                 store.dispatch(
                   updateOneBlock({ id: thinkingBlockId, changes: { status: MessageBlockStatus.STREAMING } })
@@ -102,9 +114,13 @@ export const processMessages = async (
           case ChunkType.THINKING_DELTA:
             {
               if (thinkingBlockId) {
+                if (thinkingStartTime === null) {
+                  thinkingStartTime = performance.now()
+                }
+                const thinkingDuration = resolveThinkingDuration(chunk.thinking_millsec)
                 throttledBlockUpdate(thinkingBlockId, {
                   content: chunk.text,
-                  thinking_millsec: chunk.thinking_millsec
+                  thinking_millsec: thinkingDuration
                 })
               }
               onStream()
@@ -113,6 +129,7 @@ export const processMessages = async (
           case ChunkType.THINKING_COMPLETE:
             {
               if (thinkingBlockId) {
+                const thinkingDuration = resolveThinkingDuration(chunk.thinking_millsec)
                 cancelThrottledBlockUpdate(thinkingBlockId)
                 store.dispatch(
                   updateOneBlock({
@@ -120,12 +137,13 @@ export const processMessages = async (
                     changes: {
                       content: chunk.text,
                       status: MessageBlockStatus.SUCCESS,
-                      thinking_millsec: chunk.thinking_millsec
+                      thinking_millsec: thinkingDuration
                     }
                   })
                 )
                 thinkingBlockId = null
               }
+              thinkingStartTime = null
             }
             break
           case ChunkType.TEXT_START:
@@ -190,6 +208,7 @@ export const processMessages = async (
           case ChunkType.ERROR:
             {
               const blockId = textBlockId || thinkingBlockId
+              thinkingStartTime = null
               if (blockId) {
                 store.dispatch(
                   updateOneBlock({

src/renderer/src/windows/selection/action/components/__tests__/ActionUtils.test.ts

@@ -284,6 +284,54 @@ describe('processMessages', () => {
     })
   })
 
+  describe('thinking timer fallback', () => {
+    it('should use local timer when thinking_millsec is missing', async () => {
+      const nowValues = [1000, 1500, 2000]
+      let nowIndex = 0
+      const performanceSpy = vi.spyOn(performance, 'now').mockImplementation(() => {
+        const value = nowValues[Math.min(nowIndex, nowValues.length - 1)]
+        nowIndex += 1
+        return value
+      })
+
+      const mockChunks = [
+        { type: ChunkType.THINKING_START },
+        { type: ChunkType.THINKING_DELTA, text: 'Thinking...' },
+        { type: ChunkType.THINKING_COMPLETE, text: 'Done thinking' },
+        { type: ChunkType.TEXT_START },
+        { type: ChunkType.TEXT_COMPLETE, text: 'Final answer' },
+        { type: ChunkType.BLOCK_COMPLETE }
+      ]
+
+      vi.mocked(fetchChatCompletion).mockImplementation(async ({ onChunkReceived }: any) => {
+        for (const chunk of mockChunks) {
+          await onChunkReceived(chunk)
+        }
+      })
+
+      await processMessages(
+        mockAssistant,
+        mockTopic,
+        'test prompt',
+        mockSetAskId,
+        mockOnStream,
+        mockOnFinish,
+        mockOnError
+      )
+
+      const thinkingDeltaCall = vi.mocked(throttledBlockUpdate).mock.calls.find(([id]) => id === 'thinking-block-1')
+      const deltaPayload = thinkingDeltaCall?.[1] as { thinking_millsec?: number } | undefined
+      expect(deltaPayload?.thinking_millsec).toBe(500)
+
+      const thinkingCompleteUpdate = vi
+        .mocked(updateOneBlock)
+        .mock.calls.find(([payload]) => (payload as any)?.changes?.thinking_millsec !== undefined)
+      expect((thinkingCompleteUpdate?.[0] as any)?.changes?.thinking_millsec).toBe(1000)
+
+      performanceSpy.mockRestore()
+    })
+  })
+
   describe('stream with exceptions', () => {
     it('should handle error chunks properly', async () => {
       const mockError = new Error('Stream processing error')