crestapps-core-ai-runtime

CrestApps.Core AI Runtime

The runtime is the provider-agnostic layer every other AI feature builds on. A connection holds credentials and an endpoint, a deployment names a model on a connection and declares what that model can do, and a slot says what the installation uses a deployment for.

Guidelines

• Prefer AddCrestAppsCore(...).AddAISuite(...) for application composition. Use AddCoreAIServices() only when the host composes raw service registrations itself.
• Program against IAIClientFactory, IAICompletionService, IAICompletionContextBuilder and IAIDeploymentManager. Never create a provider SDK client in business code.
• Every configured deployment declares its model capabilities through Features. A configured entry with no features and no legacy purpose is skipped at startup with the warning "declares no model capabilities".
• Pick deployments through slots (AIDeploymentSlotNames) and ResolveSlotAsync(...). There is no deployment purpose or type in 2.0.
• Use crestapps-core-ai-model-capabilities for registering features, parameters and slots, crestapps-core-ai-usage-metering for usage records, and the provider skills (crestapps-core-openai, crestapps-core-azure-openai, crestapps-core-ollama, crestapps-core-azure-ai-inference) for provider-specific settings.

Registration

using CrestApps.Core.AI;
using CrestApps.Core.AI.OpenAI;

builder.Services.AddCrestAppsCore(crestApps => crestApps
    .AddAISuite(ai => ai
        .AddOpenAI()
    )
);

AddAISuite(...) calls AddCoreServices(), AddCoreAIServices() and AddCoreAIOrchestration(). Without the builder:

builder.Services
    .AddCoreAIServices()
    .AddCoreAIOrchestration()
    .AddCoreAIOpenAI();

AddCoreAIServices() also chains AddCoreAITemplating() and AddCoreAIDeploymentCapabilities() (features, parameters and slots). IAIDeploymentManager is registered by AddCoreAIOrchestration(), so register orchestration whenever code resolves deployments.