One connection. Any model.

Configured once, reused everywhere

An endpoint carries the few settings a model needs — which model, how large its context window is, how many tokens it may return — and exposes AskAsync for inference and AskEmbeddingsAsync for vectors. A SmartHub points at one endpoint and every adapter on it inherits the connection.

• Model and EmbeddingModel select the models to call.
• ContextWindow and MaxOutputTokens bound each request.
• SystemPrompt and ToolsPrompt override the built-in prompts.
• UseNativeTools switches to a provider's native tool payload.
• Usage reports token metrics for the endpoint.

// Point a hub at OpenAI — adapters need no further setup
var endpoint = new OpenAIEndpoint {
    Model = "gpt-4o",
    EmbeddingModel = "text-embedding-3-small",
    ContextWindow = 128000,
    MaxOutputTokens = 4096
};

var hub = new SmartHub(this) { Endpoint = endpoint };

// Switching providers is one line:
hub.Endpoint = new AnthropicEndpoint { Model = "claude-sonnet" };