feat(ui): replace hardcoded model lists with dynamic tokenlens API

Add GET /llm/model-suggestions endpoint that maps ProviderType to models.dev
provider keys and returns chat model IDs from the tokenlens catalog. Lazy-loads
catalog on first request to avoid empty results when engine hasn't started.

Frontend ModelCombobox now fetches suggestions via useQuery with 30min cache
instead of reading from hardcoded MODEL_SUGGESTIONS constant.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)

frontend/src/components/llm/ModelCombobox.tsx

@@ -1,7 +1,7 @@
 import { useState, useRef, useEffect } from 'react';
 import { Input } from '@/components/ui/input';
 import { useQuery } from '@tanstack/react-query';
-import { fetchModels, MODEL_SUGGESTIONS } from '@/services/llmProviderService';
+import { fetchModels, fetchModelSuggestions } from '@/services/llmProviderService';
 import type { ProviderType } from '@/services/llmProviderService';
 
 interface ModelComboboxProps {
@@ -42,9 +42,16 @@ export function ModelCombobox({
     staleTime: 5 * 60 * 1000,
   });
 
+  // Fetch dynamic model suggestions from backend (powered by models.dev)
+  const { data: suggestions = {} } = useQuery({
+    queryKey: ['llm-model-suggestions'],
+    queryFn: fetchModelSuggestions,
+    staleTime: 30 * 60 * 1000, // 30 min cache
+  });
+
   // Build tagged model list: API > suggestions > custom input
   const useApiFetched = fetchedModels.length > 0;
-  const suggestionModels = providerType ? MODEL_SUGGESTIONS[providerType] || [] : [];
+  const suggestionModels = providerType ? suggestions[providerType] || [] : [];
 
   type TaggedModel = { name: string; tag: 'API' | '推荐' | '自定义' };
 

frontend/src/components/llm/__tests__/ModelCombobox.test.tsx

@@ -11,6 +11,12 @@ vi.mock('@/services/llmProviderService', async () => {
   return {
     ...actual,
     fetchModels: vi.fn(),
+    fetchModelSuggestions: vi.fn().mockResolvedValue({
+      openai_compatible: ['gpt-4o', 'gpt-4o-mini', 'deepseek-chat'],
+      openai_responses: ['gpt-4o', 'gpt-4o-mini', 'o3-mini'],
+      anthropic: ['claude-sonnet-4-20250514', 'claude-3-5-haiku-20241022'],
+      gemini: ['gemini-2.5-pro', 'gemini-2.5-flash', 'gemini-2.0-flash'],
+    }),
   };
 });
 

frontend/src/services/llmProviderService.ts

@@ -31,13 +31,23 @@ export interface TestResult {
   error?: string;
 }
 
-export const MODEL_SUGGESTIONS: Record<ProviderType, string[]> = {
-  openai_compatible: ['gpt-4o', 'gpt-4o-mini', 'gpt-4-turbo', 'deepseek-chat', 'qwen-plus'],
-  openai_responses: ['gpt-4o', 'gpt-4o-mini', 'gpt-4.1', 'gpt-4.1-mini', 'o3-mini'],
-  anthropic: ['claude-sonnet-4-20250514', 'claude-3-5-haiku-20241022', 'claude-opus-4-20250514'],
+/** Fallback suggestions when API is unavailable (e.g. catalog not loaded yet). */
+const FALLBACK_SUGGESTIONS: Record<ProviderType, string[]> = {
+  openai_compatible: ['gpt-4o', 'gpt-4o-mini', 'deepseek-chat'],
+  openai_responses: ['gpt-4o', 'gpt-4o-mini', 'o3-mini'],
+  anthropic: ['claude-sonnet-4-20250514', 'claude-3-5-haiku-20241022'],
   gemini: ['gemini-2.5-pro', 'gemini-2.5-flash', 'gemini-2.0-flash'],
 };
 
+export const fetchModelSuggestions = async (): Promise<Record<string, string[]>> => {
+  try {
+    const response = await api.get<Record<string, string[]>>('/llm/model-suggestions');
+    return response.data;
+  } catch {
+    return FALLBACK_SUGGESTIONS;
+  }
+};
+
 export const fetchProviders = async (): Promise<ProviderDto[]> => {
   const response = await api.get<ProviderDto[]>('/llm/providers');
   return response.data;

src/controllers/llm-config.ts

@@ -14,6 +14,7 @@ import { secretRepo } from '../db/repositories/secret-repo';
 import { settingsRepo } from '../db/repositories/settings-repo';
 import { llmGateway } from '../llm/gateway';
 import { MODEL_ROLES } from '../llm/types';
+import { tokenCounter } from '../review/context/token-counter';
 
 export const llmConfigRouter = new Hono();
 
@@ -289,6 +290,39 @@ llmConfigRouter.post('/providers/:id/test', async (c) => {
   }
 });
 
+// ── Model Suggestions (from models.dev via tokenlens) ───────────────────
+
+/**
+ * Map our ProviderType to models.dev provider keys.
+ * openai_compatible is special: it aggregates multiple providers since
+ * users often point compatible endpoints at DeepSeek, Qwen, etc.
+ */
+const PROVIDER_TYPE_TO_CATALOG_KEYS: Record<string, string[]> = {
+  openai_compatible: ['openai', 'deepseek', 'qwen'],
+  openai_responses: ['openai'],
+  anthropic: ['anthropic'],
+  gemini: ['google'],
+};
+
+llmConfigRouter.get('/model-suggestions', async (c) => {
+  // Ensure catalog is loaded (lazy init on first request)
+  if (!tokenCounter.hasCatalog) {
+    await tokenCounter.refreshCatalog();
+  }
+
+  const result: Record<string, string[]> = {};
+
+  for (const [providerType, catalogKeys] of Object.entries(PROVIDER_TYPE_TO_CATALOG_KEYS)) {
+    const models: string[] = [];
+    for (const key of catalogKeys) {
+      models.push(...tokenCounter.getModelSuggestions(key));
+    }
+    result[providerType] = models;
+  }
+
+  return c.json(result);
+});
+
 // ── System Settings ─────────────────────────────────────────────────────
 
 llmConfigRouter.get('/settings', (c) => {