Validate provider model, auth, and endpoint before save

docs/pr/provider-model-check-pr.md

@@ -0,0 +1,96 @@
+# Add provider model-check support
+
+## Summary
+
+This PR adds a lightweight **model check** workflow to the Add/Edit Provider UI so users can verify whether a provider endpoint, API key, and model ID are actually usable **before saving**.
+
+## Why
+
+Today, adding or editing a provider often requires trial-and-error:
+
+- users may save an invalid model ID
+- users may save a bad API key
+- users may save an unreachable / misconfigured endpoint
+- the actual failure is only discovered later when the provider is used
+
+This change adds an explicit pre-save validation step that performs a real request and reports a clear result.
+
+## What changed
+
+### UI
+
+- Added a **"检测模型" / model-check** action in both:
+  - Add Provider dialog
+  - Edit Provider panel
+- Added inline feedback text for:
+  - success
+  - degraded / slow response
+  - model not found
+  - auth failure
+  - timeout
+  - unsupported protocol
+
+### Frontend
+
+- Added `src/lib/provider-model-check.ts`
+- Wired `ProvidersSettings.tsx` to call the new API and surface results in the UI
+
+### Backend
+
+- Added `/api/provider-model-check`
+- Added `electron/services/providers/provider-model-check.ts`
+- The backend performs a real streaming request and treats the first returned chunk as proof that the model is usable
+
+### Supported protocol paths
+
+- `openai-completions`
+- `openai-responses`
+- `anthropic-messages`
+
+## Validation behavior
+
+The check intentionally stays lightweight:
+
+- uses a tiny prompt (`Hi`)
+- uses small token limits
+- uses streaming mode
+- considers the request successful once the first response chunk arrives
+
+This makes the check fast while still proving that:
+
+- the endpoint is reachable
+- auth works
+- the requested model exists / is accepted
+
+## Verification
+
+Validated locally on a clean upstream-based working tree with only the minimal feature patch applied.
+
+Commands used:
+
+```bash
+pnpm run ext:bridge
+pnpm exec tsc --noEmit
+pnpm exec vite build
+```
+
+All passed successfully.
+
+## Screenshot
+
+A local verification screenshot is included in this branch:
+
+- `docs/pr/provider-model-check-test.png`
+
+It shows the provider settings UI with the model-check entry point and successful validation state used during testing.
+
+## Scope / non-goals
+
+This PR intentionally avoids unrelated customization:
+
+- no branding changes
+- no provider list policy changes
+- no advanced-settings refactor
+- no packaging / release pipeline changes
+
+The goal is to keep this patch reviewable and upstream-friendly.

electron/api/routes/providers.ts

@@ -20,6 +20,7 @@ import {
 import { validateApiKeyWithProvider } from '../../services/providers/provider-validation';
 import { getProviderService } from '../../services/providers/provider-service';
 import { providerAccountToConfig } from '../../services/providers/provider-store';
+import { checkProviderModel, type ModelCheckInput } from '../../services/providers/provider-model-check';
 import type { ProviderAccount } from '../../shared/providers/types';
 import { logger } from '../../utils/logger';
 
@@ -50,6 +51,21 @@ export async function handleProviderRoutes(
     );
   };
 
+  if (url.pathname === '/api/provider-model-check' && req.method === 'POST') {
+    try {
+      const body = await parseJsonBody<ModelCheckInput>(req);
+      sendJson(res, 200, await checkProviderModel(body));
+    } catch (error) {
+      sendJson(res, 500, {
+        success: false,
+        status: 'failed',
+        message: String(error),
+        modelUsed: '',
+      });
+    }
+    return true;
+  }
+
   if (url.pathname === '/api/provider-vendors' && req.method === 'GET') {
     sendJson(res, 200, await providerService.listVendors());
     return true;