Building a Custom Provider

Implement IAIProvider to connect Reveal SDK AI to an LLM service that does not have a built-in provider. A provider is a connection adapter; register it by name and expose it through one or more profiles. Clients select profiles, never providers directly.

Implement the provider

Implement GetResponseAsync. Reveal resolves the selected profile before calling the provider, so request.Settings includes the model and call settings.

MyCustomProvider.cs

using Reveal.Sdk.AI;

public sealed class MyCustomProvider : IAIProvider
{
    public async Task<AIResponse> GetResponseAsync(
        AIRequest request,
        CancellationToken cancellationToken = default)
    {
        // Send request.Messages and request.Settings to your LLM service.
        var result = await _client.GetResponseAsync(
            request.Messages,
            request.Settings,
            cancellationToken);

        return new AIResponse
        {
            Text = result.Text,
            FinishReason = AIFinishReason.Stop,
            Model = request.Settings.Model
        };
    }
}

AIRequest property	Description
Messages	Conversation messages with their roles.
Settings	Fully resolved profile settings, including the model.
Intent	The SDK workflow that initiated the request.
ResponseFormat	Whether text or JSON output is requested.
UserContext	Calling user, or the system user for background work.

Return AIResponse with the generated Text; include FinishReason, Model, and AITokenUsage whenever the service reports them.

Register the connection and profile

Program.cs

builder.Services.AddRevealAI()
    .AddProvider<MyCustomProvider>("my-service")
    .AddProfile("my-service-default", profile =>
    {
        profile.Provider = "my-service";
        profile.Model = "my-model";
        profile.DisplayName = "My service";
        profile.Temperature = 0.2;
    })
    .SetDefaultProfile("my-service-default");

tip

Do not keep a model name inside the provider implementation. One provider can serve multiple profiles; read the resolved model and generation settings from request.Settings.

Migrating existing providers

SendPromptAsync(ProviderRequest) remains temporarily for compatibility. Implement GetResponseAsync(AIRequest) for new providers: it preserves message roles and receives profile-resolved settings. The legacy Name property is no longer used for selection; the name passed to AddProvider identifies the connection.