ModelService

Namespace: Daisi.Host.Core.Services

The ModelService manages local model files for self-hosted inference. It queries the Orc for required models, checks which are present on disk, downloads missing ones with progress reporting, and registers them in the host settings.

Constructor

ModelService is typically resolved via dependency injection. It requires an ISettingsService, a ModelClientFactory, and an ILogger.

// DI registration (done at app startup)
builder.Services.AddSingleton<Daisi.Host.Core.Services.Interfaces.ISettingsService, DesktopSettingsService>();
builder.Services.AddSingleton<ModelService>();

Properties

• ISettingsService SettingsService get set
  The host settings service, providing access to model folder path and model configuration.
• Settings Settings get
  Shortcut to SettingsService.Settings.
• ModelClient ModelsClient get
  The Orc ModelClient used to query required models.
• List<LocalModel> LocalModels get
  The list of loaded local models after LoadModels() has been called.
• LocalModel? Default get
  The default model. Returns the first model marked IsDefault, or the first enabled model, or the first model in the list.

Methods

• void LoadModels()
  Loads all models registered in Settings.Model.Models into LocalModels. Models with LoadAtStartup = true are loaded into memory immediately.
• Task DownloadRequiredModelsAsync()
  Queries the Orc for required models, checks which are missing from the model folder, downloads them, registers them in settings, and saves. This is the all-in-one orchestration method.
• Task DownloadRequiredModelAsync(AIModel model, Action<double>? percentageProgress = null)
  Downloads a single model file from its URL to the local model folder. The optional percentageProgress callback is invoked with values from 0 to 100 as the download proceeds.
• bool IsDownloaded(AIModel model)
  Returns true if the model is registered in the local settings.
• bool IsDefault(AIModel model)
  Returns true if the model is marked as the default.
• bool IsEnabled(AIModel model)
  Returns true if the model is enabled.

Downloading a Single Model with Progress

Use DownloadRequiredModelAsync with a progress callback to show download progress to the user. The callback receives a percentage value (0–100).

var modelService = host.Services.GetRequiredService<ModelService>();

// Load host settings first
var settingsService = host.Services.GetRequiredService<ISettingsService>();
await settingsService.LoadAsync();

// Get required models from Orc
var requiredModels = modelService.ModelsClient.GetRequiredModels().Models;

// Check which files are missing
string modelFolder = settingsService.Settings.Model.ModelFolderPath;
var existingFiles = Directory.GetFiles(modelFolder)
    .Select(Path.GetFileName)
    .ToHashSet(StringComparer.OrdinalIgnoreCase);

foreach (var model in requiredModels)
{
    if (!existingFiles.Contains(model.FileName))
    {
        Console.WriteLine($"Downloading {model.Name}...");

        await modelService.DownloadRequiredModelAsync(model, progress =>
        {
            Console.Write($"\r  {progress:F1}% complete");
        });

        Console.WriteLine($"\n  Done: {model.FileName}");
    }
}

// Load models into memory
modelService.LoadModels();

Auto-Download Pattern (Used by DaisiBot)

DaisiBot's TUI and MAUI apps implement an auto-download pattern at startup. The flow is:

1. Check if HostModeEnabled is true in user settings (this is the default).
2. Call ILocalInferenceService.GetRequiredDownloadsAsync() to get the list of missing models.
3. If models are missing, show a progress dialog.
4. For each missing model, call ILocalInferenceService.DownloadModelAsync() with a progress callback that updates the UI.
5. After all downloads complete, call ILocalInferenceService.InitializeAsync() to load settings, tools, and models.
6. Auto-close the dialog and proceed to the main screen.

Users can skip the download (Esc in TUI, Skip button in MAUI) and switch to DaisiNet cloud mode later via the host mode toggle.

// Simplified startup pattern
var localInference = host.Services.GetRequiredService<ILocalInferenceService>();

var missing = await localInference.GetRequiredDownloadsAsync();

if (missing.Count > 0)
{
    foreach (var model in missing)
    {
        Console.WriteLine($"Downloading {model.Name}...");
        await localInference.DownloadModelAsync(model, progress =>
        {
            Console.Write($"\r  {progress:F1}%");
        });
        Console.WriteLine();
    }
}

await localInference.InitializeAsync();
Console.WriteLine("Local inference ready.");