feat: Dynamisches Modell-Laden für Nvidia & Pollinations Provider

- OpenAiService: FetchModels() generisch für beide Provider (OpenAI-Format & Pollinations-Array-Format)
- Pollinations Base-URL direkt auf text.pollinations.ai/openai (kein g4f.space Proxy → kein 402)
- Pollinations Modelle-URL: text.pollinations.ai/models
- ModelsEndpoint: /api/models/nvidia + /api/models/pollinations
- agent-page.js: loadProviderModels() generisch, Promise.all für beide Provider parallel
- index.html: statische Nvidia-Option entfernt, beide Optgroups mit 'Wird geladen...'

Services/OpenAiService.cs

@@ -25,7 +25,8 @@ public class OpenAiService
     // Provider-Definitionen
     private static readonly Dictionary<string, (string BaseUrl, string ApiKey)> Providers = new()
     {
-        ["nvidia"] = ("https://g4f.space/api/nvidia/v1", "")
+        ["nvidia"]      = ("https://g4f.space/api/nvidia/v1",      ""),
+        ["pollinations"] = ("https://text.pollinations.ai/openai", "")
     };
 
     // Tools die für OpenAI-kompatible Provider verfügbar sind
@@ -146,6 +147,11 @@ public class OpenAiService
         }
     };
 
+    public record ProviderModel(string Id, string DisplayName);
+
+    private List<ProviderModel> _nvidiaModels      = new List<ProviderModel>();
+    private List<ProviderModel> _pollinationsModels = new List<ProviderModel>();
+
     public OpenAiService(IHttpClientFactory httpFactory, DatabaseService db, ToolService toolService)
     {
         _httpFactory = httpFactory;
@@ -153,6 +159,61 @@ public class OpenAiService
         _toolService = toolService;
     }
 
+    /// <summary>
+    /// Lädt verfügbare Modelle aller Provider beim Server-Start und cached sie.
+    /// </summary>
+    public async Task LoadModelsAsync()
+    {
+        _nvidiaModels      = await FetchModels("https://g4f.space/api/nvidia/models",      "Nvidia");
+        _pollinationsModels = await FetchModels("https://text.pollinations.ai/models", "Pollinations");
+    }
+
+    private async Task<List<ProviderModel>> FetchModels(string url, string providerName)
+    {
+        try
+        {
+            var client = CreateClient("");
+            var resp = await client.GetAsync(url);
+            if (!resp.IsSuccessStatusCode) return new List<ProviderModel>();
+
+            var json = await resp.Content.ReadAsStringAsync();
+            var root = JsonSerializer.Deserialize<JsonElement>(json);
+
+            // Pollinations: root is array [{ name, ... }]
+            // Nvidia/OpenAI: root is { data: [{ id, ... }] }
+            JsonElement arr;
+            if (root.ValueKind == JsonValueKind.Array)
+                arr = root;
+            else if (root.TryGetProperty("data", out var data))
+                arr = data;
+            else
+                return new List<ProviderModel>();
+
+            var models = new List<ProviderModel>();
+            foreach (var item in arr.EnumerateArray())
+            {
+                // try "id" first (Nvidia), fallback to "name" (Pollinations)
+                string id = null;
+                if (item.TryGetProperty("id",   out var idProp)   && idProp.ValueKind   == JsonValueKind.String) id = idProp.GetString();
+                if (id == null && item.TryGetProperty("name", out var nameProp) && nameProp.ValueKind == JsonValueKind.String) id = nameProp.GetString();
+                if (string.IsNullOrEmpty(id)) continue;
+                var display = id.Contains('/') ? id[(id.LastIndexOf('/') + 1)..] : id;
+                models.Add(new ProviderModel(id, display));
+            }
+
+            Console.WriteLine($"[OpenAiService] {models.Count} {providerName}-Modelle geladen.");
+            return models;
+        }
+        catch (Exception ex)
+        {
+            Console.WriteLine($"[OpenAiService] Fehler beim Laden der {providerName}-Modelle: {ex.Message}");
+            return new List<ProviderModel>();
+        }
+    }
+
+    public List<ProviderModel> GetNvidiaModels()      => _nvidiaModels;
+    public List<ProviderModel> GetPollinationsModels() => _pollinationsModels;
+
     /// <summary>
     /// Gibt den Provider und das eigentliche Modell-ID aus einem kombinierten Model-String zurück.
     /// Format: "nvidia:moonshotai/kimi-k2.5" → ("nvidia", "moonshotai/kimi-k2.5")