Make connection checks exercise the selected model end to end

The settings validation button now uses the currently selected LLM model for a real chat-completion probe instead of only listing models/auth state. The frontend message was simplified to reflect validation success without conflating it with model-count discovery, which remains handled by the dedicated fetch-models action.

Constraint: Keep the change minimal and local; preserve the separate fetch-models workflow for listing available models.
Rejected: Keep auth-only validation | it would keep missing the failure mode reported in issue #136.
Rejected: Add a broader provider refactor | unnecessary for the narrow validation behavior change.
Confidence: high
Scope-risk: narrow
Directive: Validation should probe the selected model path; model enumeration belongs to fetch-models.
Tested: npm ci; npm run build; cargo test --lib
Not-tested: Live provider endpoint behavior against real credentials on the user’s machine

openless-all/app/src-tauri/src/commands.rs

@@ -10,6 +10,7 @@ use tauri::{AppHandle, State};
 use crate::coordinator::Coordinator;
 use crate::permissions::{self, PermissionStatus};
 use crate::persistence::{CredentialAccount, CredentialsSnapshot, CredentialsVault};
+use crate::polish::{OpenAICompatibleConfig, OpenAICompatibleLLMProvider};
 use crate::types::{
     CredentialsStatus, DictationSession, DictionaryEntry, HotkeyCapability, HotkeyStatus,
     PolishMode, QaHotkeyBinding, UserPreferences,
@@ -88,7 +89,6 @@ pub fn read_credential(account: String) -> Result<Option<String>, String> {
 #[serde(rename_all = "camelCase")]
 pub struct ProviderCheckResult {
     ok: bool,
-    model_count: usize,
 }
 
 #[derive(Serialize)]
@@ -98,13 +98,18 @@ pub struct ProviderModelsResult {
 
 #[tauri::command]
 pub async fn validate_provider_credentials(kind: String) -> Result<ProviderCheckResult, String> {
-    let config = read_openai_provider_config(&kind)?;
-    fetch_provider_models(&config)
-        .await
-        .map(|models| ProviderCheckResult {
-            ok: true,
-            model_count: models.len(),
-        })
+    match kind.as_str() {
+        "llm" => validate_llm_provider()
+            .await
+            .map(|()| ProviderCheckResult { ok: true }),
+        "asr" => {
+            let config = read_openai_provider_config(&kind)?;
+            fetch_provider_models(&config)
+                .await
+                .map(|_| ProviderCheckResult { ok: true })
+        }
+        _ => Err(format!("unknown provider kind: {kind}")),
+    }
 }
 
 #[tauri::command]
@@ -141,6 +146,26 @@ fn read_openai_provider_config(kind: &str) -> Result<ProviderConfig, String> {
     Ok(ProviderConfig { base_url, api_key })
 }
 
+async fn validate_llm_provider() -> Result<(), String> {
+    let config = read_openai_provider_config("llm")?;
+    let model = CredentialsVault::get(CredentialAccount::ArkModelId)
+        .map_err(|e| e.to_string())?
+        .filter(|s| !s.is_empty())
+        .unwrap_or_else(|| "deepseek-v3-2".to_string());
+    let provider = OpenAICompatibleLLMProvider::new(OpenAICompatibleConfig::new(
+        "ark",
+        "Doubao Ark",
+        config.base_url,
+        config.api_key,
+        model,
+    ));
+    provider
+        .polish("验证连接", PolishMode::Raw, &[], &[], None)
+        .await
+        .map(|_| ())
+        .map_err(|e| e.to_string())
+}
+
 async fn fetch_provider_models(config: &ProviderConfig) -> Result<Vec<String>, String> {
     let url = models_url(&config.base_url);
     log::info!("[provider-check] GET {url}");
@@ -454,10 +479,7 @@ pub fn get_qa_hotkey_label(coord: CoordinatorState<'_>) -> String {
 /// 传入 `None` 形式的字段不在这里支持——前端用 `binding == null` 时调下面的
 /// "disable" 写法（写 prefs.qa_hotkey = None）即可。
 #[tauri::command]
-pub fn set_qa_hotkey(
-    coord: CoordinatorState<'_>,
-    binding: QaHotkeyBinding,
-) -> Result<(), String> {
+pub fn set_qa_hotkey(coord: CoordinatorState<'_>, binding: QaHotkeyBinding) -> Result<(), String> {
     let mut prefs = coord.prefs().get();
     prefs.qa_hotkey = Some(binding);
     coord.prefs().set(prefs).map_err(|e| e.to_string())?;

openless-all/app/src/i18n/en.ts

@@ -298,7 +298,7 @@ export const en: typeof zhCN = {
       accessKeyLabel: 'Access Key',
       resourceIdLabel: 'Resource ID',
       toolsLabel: 'Connection check',
-      toolsDesc: 'Save the fields above, then validate credentials or fetch models. Manual model input remains available if fetching fails.',
+      toolsDesc: 'Save the fields above, then validate the selected model or fetch models. Manual model input remains available if fetching fails.',
       validate: 'Validate',
       validating: 'Validating…',
       fetchModels: 'Fetch models',
@@ -307,7 +307,7 @@ export const en: typeof zhCN = {
       modelsLoaded: 'Fetched {{count}} models.',
       selectModel: 'Select a model to fill the field above',
       modelSaved: 'Saved model {{model}}.',
-      validateSuccess: 'Credentials are valid. {{count}} models available.',
+      validateSuccess: 'Connection check passed.',
       providerHttpStatus: 'Provider returned HTTP {{status}}. Check the API key permissions or endpoint.',
       apiKeyMissing: 'API Key is empty.',
       endpointMissing: 'Endpoint is empty.',

openless-all/app/src/i18n/zh-CN.ts

@@ -296,7 +296,7 @@ export const zhCN = {
       accessKeyLabel: 'Access Key',
       resourceIdLabel: '资源 ID',
       toolsLabel: '连接检查',
-      toolsDesc: '先保存上方配置，再验证鉴权或拉取模型；失败时仍可手动填写模型 ID。',
+      toolsDesc: '先保存上方配置，再验证当前模型连通性或拉取模型；失败时仍可手动填写模型 ID。',
       validate: '验证',
       validating: '验证中…',
       fetchModels: '拉取模型',
@@ -305,7 +305,7 @@ export const zhCN = {
       modelsLoaded: '已拉取 {{count}} 个模型。',
       selectModel: '选择一个模型写入上方字段',
       modelSaved: '已保存模型 {{model}}。',
-      validateSuccess: '鉴权成功，可用模型 {{count}} 个。',
+      validateSuccess: '连接检查通过。',
       providerHttpStatus: '供应商接口返回 {{status}}，请检查 API Key 权限或 Endpoint。',
       apiKeyMissing: 'API Key 为空。',
       endpointMissing: 'Endpoint 为空。',

openless-all/app/src/lib/ipc.ts

@@ -68,7 +68,6 @@ const mockCredentialsStatus: CredentialsStatus = {
 
 export interface ProviderCheckResult {
   ok: boolean;
-  modelCount: number;
 }
 
 export interface ProviderModelsResult {
@@ -144,7 +143,7 @@ export function readCredential(account: string): Promise<string | null> {
 }
 
 export function validateProviderCredentials(kind: 'llm' | 'asr'): Promise<ProviderCheckResult> {
-  return invokeOrMock('validate_provider_credentials', { kind }, () => ({ ok: true, modelCount: 2 }));
+  return invokeOrMock('validate_provider_credentials', { kind }, () => ({ ok: true }));
 }
 
 export function listProviderModels(kind: 'llm' | 'asr'): Promise<ProviderModelsResult> {

openless-all/app/src/pages/Settings.tsx

@@ -428,12 +428,7 @@ function ProviderTools({ kind, modelAccount, onModelSelected }: { kind: 'llm' |
     setResult('loading', t('settings.providers.validating'));
     try {
       const result = await validateProviderCredentials(kind);
-      setResult(
-        result.ok ? (result.modelCount === 0 ? 'empty' : 'success') : 'error',
-        result.modelCount === 0
-          ? t('settings.providers.modelsEmpty')
-          : t('settings.providers.validateSuccess', { count: result.modelCount }),
-      );
+      setResult(result.ok ? 'success' : 'error', t('settings.providers.validateSuccess'));
     } catch (error) {
       setResult('error', providerErrorMessage(error, t));
     }