src/renderer/src/i18n/zh-TW/settings.json (1)

104-160: Same newline-escape caveat as zh-HK file.
See previous comment for details.

src/renderer/src/i18n/ru-RU/settings.json (1)

104-160: Same newline-escape caveat as zh-HK file.

src/renderer/src/i18n/fa-IR/settings.json (1)

104-160: Same newline-escape caveat as zh-HK file.

src/shared/model.ts (1)

8-9: Add a trailing comma to avoid noisy future diffs

Every other enum member ends with a comma; keep the pattern for the last entry as well.

-  ImageGeneration = 'imageGeneration'
+  ImageGeneration = 'imageGeneration',

src/renderer/src/components/settings/ModelProviderSettingsDetail.vue (1)

348-352: Consider throttling initData() to prevent excessive refreshes

config-changed may fire repeatedly while a user tweaks multiple fields in the dialog.
Re-running initData() on every emission can cause unnecessary network / store calls.

-const handleConfigChanged = async () => {
-  await initData()
-}
+import { throttle } from 'lodash-es'
+
+const handleConfigChanged = throttle(async () => {
+  await initData()
+}, 500)

A 500 ms throttle (or debounce) keeps the UI in sync without spamming the store.

src/renderer/src/components/settings/ProviderModelList.vue (1)

111-119: Same note for provider models – duplication of inline arrow fn.

Nice symmetry with the custom-model block.
Minor: @enabled-change="(enabled) => handleModelEnabledChange(model, enabled)" plus the
new @config-changed arrow recreates two closures per row on every render.
If repaint frequency becomes a concern, consider hoisting those handlers
(handleConfigChanged) outside the template.

src/renderer/src/i18n/zh-HK/settings.json (1)

103-160: Large localisation block added – only one nitpick.

JSON includes explicit line-breaks (\n) within the resetConfirm.message.
If your i18n extraction tooling strips escapes, the newline may render as \n
instead of an actual break in some UI libraries.
Consider using a real line break or separate paragraphs.

src/renderer/src/i18n/en-US/settings.json (1)

125-133: Refine wording & drop superfluous repetition
Sentence reads a bit awkwardly (“The higher the higher the randomness.”). Consider tightening:

-        "description": "Control the randomness of the output. Most models are 0-1, and some support between 0-2. The higher the higher the randomness.",
+        "description": "Controls output randomness. Typical range is 0-1 (some models allow 0-2); higher values yield more random results.",

src/renderer/src/components/settings/ModelConfigItem.vue (2)

42-50: Great UX touch – settings button added

Button wiring, i18n title and icon all check out.
Small nitpick: lucide:settings icon colour could respond to hover: to match other action buttons.

-<Icon icon="lucide:settings" class="w-4 h-4 text-muted-foreground" />
+<Icon icon="lucide:settings" class="w-4 h-4 text-muted-foreground hover:text-foreground" />

---

104-114: Emit payload for clarity

configChanged currently emits no data. Passing at least the modelId (and maybe the updated config) helps parents update state without having to refetch.

-const onConfigSaved = () => {
-  emit('configChanged')
-}
+const onConfigSaved = (payload?: unknown) => {
+  emit('configChanged', { modelId, ...payload })
+}

You’ll then need to extend defineEmits accordingly.

src/main/presenter/syncPresenter/index.ts (7)

163-169: Remove unnecessary empty lines.

The empty lines break the flow of backup operations. Consider removing them to maintain consistency with the rest of the backup logic.

      // 备份模型配置文件
      if (fs.existsSync(this.MODEL_CONFIG_PATH)) {
        fs.copyFileSync(this.MODEL_CONFIG_PATH, tempModelConfigPath)
      }
-
-
-
      // 如果 provider_models 目录存在，备份整个目录

---

224-231: Remove unnecessary empty lines for consistency.

        // 导入模型配置文件
        if (fs.existsSync(modelConfigBackupPath)) {
          fs.copyFileSync(modelConfigBackupPath, this.MODEL_CONFIG_PATH)
        }
-
-
-
        eventBus.send(SYNC_EVENTS.IMPORT_COMPLETED, SendTarget.ALL_WINDOWS)

---

256-262: Remove unnecessary empty lines for consistency.

        // 恢复模型配置文件
        if (fs.existsSync(tempModelConfigPath)) {
          fs.copyFileSync(tempModelConfigPath, this.MODEL_CONFIG_PATH)
        }
-
-
-
        eventBus.send(SYNC_EVENTS.IMPORT_ERROR, SendTarget.ALL_WINDOWS, (error as Error).message || 'sync.error.unknown')

---

334-336: Remove unnecessary empty line.

      const finalMcpSettingsBackupPath = path.join(syncFolderPath, 'mcp-settings.json')
      const finalModelConfigBackupPath = path.join(syncFolderPath, 'model-config.json')
-

      // 确保数据库文件存在

---

370-382: Fix formatting issues.

Remove unnecessary empty lines and maintain consistent formatting.

      }
-
      // 备份 MCP 设置
      if (fs.existsSync(this.MCP_SETTINGS_PATH)) {
        fs.copyFileSync(this.MCP_SETTINGS_PATH, tempMcpSettingsBackupPath)
      }
-
      // 备份模型配置文件
      if (fs.existsSync(this.MODEL_CONFIG_PATH)) {
        fs.copyFileSync(this.MODEL_CONFIG_PATH, tempModelConfigBackupPath)
      }
-
-
-
      // 备份 provider_models 目录

---

399-403: Remove unnecessary empty line.

      if (!fs.existsSync(tempAppSettingsBackupPath)) {
        throw new Error('sync.error.tempConfigFailed')
      }
-
      if (!fs.existsSync(tempMcpSettingsBackupPath)) {
        throw new Error('sync.error.tempMcpSettingsFailed')
      }

---

422-436: Fix comment indentation and remove empty lines.

The comment indentation is incorrect and there are unnecessary empty lines.

      }

-            // 清理之前的模型配置文件备份
+      // 清理之前的模型配置文件备份
      if (fs.existsSync(finalModelConfigBackupPath)) {
        fs.unlinkSync(finalModelConfigBackupPath)
      }
-
      // 确保临时文件存在后再执行重命名
      fs.renameSync(tempDbBackupPath, finalDbBackupPath)
      fs.renameSync(tempAppSettingsBackupPath, finalAppSettingsBackupPath)
      fs.renameSync(tempMcpSettingsBackupPath, finalMcpSettingsBackupPath)
-
-            // 重命名模型配置文件
+      // 重命名模型配置文件
      if (fs.existsSync(tempModelConfigBackupPath)) {
        fs.renameSync(tempModelConfigBackupPath, finalModelConfigBackupPath)
      }
-
      // 重命名 provider_models 临时目录

src/renderer/src/components/settings/ModelConfigDialog.vue (3)

13-31: Consider making max token limits configurable per model.

The hardcoded maximum value of 1,000,000 tokens may not be appropriate for all models. Some models support much higher limits (e.g., 100M tokens for certain context windows), while others have lower limits.

Consider deriving these limits from the model's actual capabilities or making them configurable per provider/model.

---

217-226: Consider loading model-specific defaults.

The hardcoded default values may not be appropriate for all models. Different models have different optimal defaults (e.g., embedding models don't need high maxTokens).

Consider initializing these values based on the model type or loading them from the model's default configuration:

 // 配置数据
-const config = ref<ModelConfig>({
-  maxTokens: 4096,
-  contextLength: 8192,
-  temperature: 0.7,
-  vision: false,
-  functionCall: false,
-  reasoning: false,
-  type: ModelType.Chat
-})
+const config = ref<ModelConfig>({
+  maxTokens: 0,
+  contextLength: 0,
+  temperature: 0,
+  vision: false,
+  functionCall: false,
+  reasoning: false,
+  type: ModelType.Chat
+})
+
+// Load defaults in loadConfig() based on model type

---

308-319: Potential duplicate config loading.

The watch with immediate: true and the onMounted hook both call loadConfig() when the dialog opens, which could cause redundant API calls.

 // 监听props变化，重新加载配置
 watch(() => [props.modelId, props.providerId, props.open], () => {
   if (props.open) {
     loadConfig()
   }
-}, { immediate: true })
+})

 onMounted(() => {
   if (props.open) {
     loadConfig()
   }
 })

test/main/presenter/modelConfig.test.ts (1)

197-198: Remove debug console.log statements.

Debug logging should be removed from test files.

-      console.log('Default config maxTokens:', defaultConfig.maxTokens)
-      console.log('Provider config maxTokens:', providerConfig.maxTokens)

src/renderer/src/stores/settings.ts (1)

1391-1409: Good implementation of model config methods. Remove empty lines.

The model configuration methods are well-implemented and properly refresh model data after changes.

   const resetModelConfig = async (modelId: string, providerId: string): Promise<void> => {
     await configP.resetModelConfig(modelId, providerId)
     // 配置重置后刷新相关模型数据
     await refreshProviderModels(providerId)
   }
-
-
-
   return {

src/main/presenter/configPresenter/modelConfig.ts (1)

13-13: Consider using a more unique delimiter to avoid potential conflicts

The delimiter -_- could potentially exist in provider or model IDs. Consider using a more unique separator like :: or |:| that is less likely to appear in IDs.

-const SPECIAL_CONCAT_CHAR = '-_-'
+const SPECIAL_CONCAT_CHAR = '|:|'

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

src/main/presenter/llmProviderPresenter/providers/siliconcloudProvider.ts (1)

123-129: 🛠️ Refactor suggestion

parseFloat may yield NaN, bypassing the exhausted-quota guard

If totalBalance is "--" or any non-numeric string, parseFloat returns NaN.
Later you compare with <= 0, but NaN <= 0 is false, so an unusable key would be treated as valid. Consider guarding for Number.isNaN:

const totalBalance = parseFloat(keyResponse.data.totalBalance)

-if (keyStatus.remainNum !== undefined && keyStatus.remainNum <= 0) {
+if (
+  keyStatus.remainNum === undefined ||
+  Number.isNaN(keyStatus.remainNum) ||
+  keyStatus.remainNum <= 0
+) {

This prevents silent acceptance of malformed balances.

src/main/presenter/configPresenter/index.ts (1)

859-896: 🛠️ Refactor suggestion

Replace any with the correct interface (issue previously raised)

The return/parameter types still use Record<string, any> despite earlier feedback.
Switch to Record<string, IModelConfig> (or ModelConfig) to regain type safety.

-importModelConfigs(configs: Record<string, any>, overwrite: boolean = false): void {
+importModelConfigs(configs: Record<string, IModelConfig>, overwrite: boolean = false): void {

-exportModelConfigs(): Record<string, any> {
+exportModelConfigs(): Record<string, IModelConfig> {

-getAllModelConfigs(): Record<string, any> {
+getAllModelConfigs(): Record<string, IModelConfig> {

Remember to:

1. import { IModelConfig } from './modelConfig' (or from the correct module).
2. Update ModelConfigHelper API signatures accordingly.

This change prevents accidental misuse and improves autocompletion in consumers.

src/main/presenter/llmProviderPresenter/providers/siliconcloudProvider.ts (2)

105-107: Keep header-key quoting consistent & drop redundant Content-Type on GET

Mixing quoted and unquoted object keys is perfectly legal, but it creates an inconsistent style within the same literal (Authorization is now unquoted while Content-Type remains quoted). Either quote both or neither to avoid diff-noise in future style passes.

Additionally, Content-Type: application/json is unnecessary on a GET request with no body—some servers even treat it as a hint that a body is coming. Consider removing it altogether:

-      headers: {
-        Authorization: `Bearer ${this.provider.apiKey}`,
-        'Content-Type': 'application/json'
-      }
+      headers: {
+        Authorization: `Bearer ${this.provider.apiKey}`
+      }

---

112-114: Multi-line template string embeds hard newlines in the error message

The new formatting improves readability in code, but it injects \n characters into the thrown string, producing a multi-line error that can be awkward in logs or UI surfaces.
If single-line output is preferred, wrap the template or join parts explicitly:

-      throw new Error(
-        `SiliconCloud API key check failed: ${response.status} ${response.statusText} - ${errorText}`
-      )
+      throw new Error(
+        `SiliconCloud API key check failed: ${response.status} ${response.statusText} - ${errorText}`
+      )

(or use [].join(' ')).

Purely stylistic—feel free to keep as-is if multi-line output is desired.

src/main/presenter/llmProviderPresenter/providers/ppioProvider.ts (1)

87-90: Consider formatting the currency amount for readability

'¥' + keyResponse.credit_balance / 10000 returns long floating-point values such as ¥123.456789. A small formatting tweak improves UX:

-const remaining = '¥' + keyResponse.credit_balance / 10000
+const remaining = `¥${(keyResponse.credit_balance / 10000).toFixed(2)}`

If fractions of a cent matter, adjust precision accordingly.

test/main/presenter/llmProviderPresenter.test.ts (1)

403-448: Concurrent-stream test still runs streams sequentially

The loop awaits each generator before moving to the next, so at most one stream is active.
To exercise the concurrency limiter, start the handlers without awaiting them and await Promise.all later:

-      for (const { eventId, stream } of streams) {
-        try {
-          let count = 0
-          for await (const event of stream) {
-            ...
-          }
-        } catch (error) {
-          // 预期的错误
-        }
-      }
+      await Promise.all(
+        streams.map(async ({ eventId, stream }) => {
+          try {
+            let count = 0
+            for await (const event of stream) {
+              ...
+              if (count >= 2) {
+                await llmProviderPresenter.stopStream(eventId)
+                break
+              }
+            }
+          } catch {
+            /* expected */ 
+          }
+        })
+      )

This keeps the three streams truly concurrent, giving the limiter a chance to reject the third one.

src/main/presenter/configPresenter/index.ts (2)

310-320: getBatchModelStatus does unnecessary store look-ups; consider bulk access

The method loops over modelIds and calls this.getSetting for every single model key.
Because ElectronStore persists data synchronously, this results in as many synchronous disk reads as there are models. Fetching all keys once (e.g. this.store.store) and resolving the status in memory would reduce I/O overhead, especially for providers that expose dozens of models.

-    for (const modelId of modelIds) {
-      const statusKey = this.getModelStatusKey(providerId, modelId)
-      const status = this.getSetting<boolean>(statusKey)
-      result[modelId] = typeof status === 'boolean' ? status : true
-    }
+    const cache = this.store.store                 // in-memory snapshot
+    for (const modelId of modelIds) {
+      const statusKey = this.getModelStatusKey(providerId, modelId)
+      const raw      = cache[statusKey] as unknown
+      result[modelId] = typeof raw === 'boolean' ? raw : true
+    }

---

417-424: Minor: keep modelStatusMap creation outside map callback

getAllEnabledModels assembles a modelIds array only to call getBatchModelStatus, which in turn rebuilds the keys.
Pre-computing the map once per provider is fine, but placing it outside the map callback would avoid creating a large temporary array (modelIds) altogether.

test/main/eventbus/eventbus.test.ts (1)

186-188: Prefer fake timers over setTimeout for faster, deterministic tests

Using real timers (await new Promise((resolve) => setTimeout(resolve, 0))) slows the suite and can introduce flakiness on CI.
Switching to Vitest’s fake-timer utilities makes the flush instant and deterministic:

-beforeEach(() => {
-  vi.useFakeTimers()
-})
-
-// …
-await new Promise((resolve) => setTimeout(resolve, 0))
+await vi.runOnlyPendingTimersAsync()

Remember to pair vi.useFakeTimers() in the relevant beforeEach/afterEach.

Also applies to: 200-202, 214-216, 281-283, 302-304, 319-321

test/main/presenter/modelConfig.test.ts (3)

197-199: Remove noisy console.log statements from tests

These logs clutter CI output and offer no assertion value.
Simply delete them or convert to comments.

-      console.log('Default config maxTokens:', defaultConfig.maxTokens)
-      console.log('Provider config maxTokens:', providerConfig.maxTokens)

---

118-127: Hard-coding default numbers couples tests to implementation details

Asserting exact defaults (4096, 8192, etc.) makes the suite brittle when defaults legitimately change.
Prefer semantic checks:

-      expect(defaultConfig).toEqual({
-        maxTokens: 4096,
-        contextLength: 8192,
-        …
-      })
+      expect(defaultConfig.maxTokens).toBeGreaterThan(0)
+      expect(defaultConfig.contextLength).toBeGreaterThan(defaultConfig.maxTokens)
+      // retain specific assertions only when they convey behaviour rather than constants

---

60-67: { ...value } is a shallow copy – consider deep-cloning for safety

If any stored config objects gain nested structures, mutations inside a test could leak to other tests despite the copy.
Using structuredClone (Node 18+) or JSON.parse(JSON.stringify(value)) ensures isolation.

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

.cursor/rules/development-setup.mdc (2)

7-11: Add quick install hints for OxLint & pnpm

The list states the required tools/versions but omits how to get them. A one-liner per item improves first-time onboarding, e.g.:

- 使用 OxLint 进行代码检查（`npm i -g @oxlint/cli`）
- 使用 pnpm 包管理（启用 corepack: `corepack enable`，然后 `corepack prepare pnpm@9 --activate`）

This tiny addition saves contributors a round-trip to external docs.

---

13-26: Group dev commands in a fenced block & include tests

Inline back-ticked commands are harder to copy en masse. Present them as a single shell block and add the missing test script for completeness:

```bash
pnpm install       # install deps
pnpm lint          # lint with OxLint
pnpm format        # run Prettier
pnpm dev           # start electron/vue dev server
pnpm test          # run unit & e2e tests
pnpm build         # build production bundles

Makes the section copy-paste friendly and surfaces the testing workflow.

</blockquote></details>

</blockquote></details>

<details>
<summary>📜 Review details</summary>

**Configuration used: CodeRabbit UI**
**Review profile: CHILL**
**Plan: Pro**


<details>
<summary>📥 Commits</summary>

Reviewing files that changed from the base of the PR and between f4cc34b6b2d1e1315543bb1261b518e3731be580 and 438fd3d3186be15d358c0e9edbff10c564b49714.

</details>

<details>
<summary>📒 Files selected for processing (5)</summary>

* `.cursor/rules/development-setup.mdc` (1 hunks)
* `src/main/presenter/configPresenter/index.ts` (7 hunks)
* `src/main/presenter/configPresenter/modelConfig.ts` (1 hunks)
* `src/renderer/src/components/settings/ModelConfigDialog.vue` (1 hunks)
* `src/shared/presenter.d.ts` (3 hunks)

</details>

<details>
<summary>🚧 Files skipped from review as they are similar to previous changes (4)</summary>

* src/renderer/src/components/settings/ModelConfigDialog.vue
* src/main/presenter/configPresenter/modelConfig.ts
* src/main/presenter/configPresenter/index.ts
* src/shared/presenter.d.ts

</details>

<details>
<summary>🔇 Additional comments (1)</summary><blockquote>

<details>
<summary>.cursor/rules/development-setup.mdc (1)</summary>

`4-4`: **Confirm `alwaysApply: true` is intentional**

Setting `alwaysApply` to `true` forces this rule on every project clone, including forks and CI jobs.  
Double-check that such global scope is desired; otherwise keep it `false` and let teams opt-in.

</details>

</blockquote></details>

</details>

<!-- This is an auto-generated comment by CodeRabbit for review status -->