AIProviderConnect Developer Manual¶

One .NET interface for talking to AI providers — chat, streaming, tool calling, and runtime model discovery behind a single IAIProvider.

AIProviderConnect ships a catalog of 35+ provider definitions (embedded JSON) and a small set of protocol-driven provider implementations. You configure a provider by ID, and the library speaks the right wire protocol for you.

Feature Overview¶

Capability	Entry point	Description
Chat completions	`IAIProvider.ChatAsync`	Single request/response chat over any registered provider
Streaming	`IStreamingChatProvider.StreamAsync`	Server-sent events as `IAsyncEnumerable<StreamingChatChunk>`
Tool calling	`ChatCompletionRequest.Tools`	Function definitions in, `ToolCall` list out
Model discovery	`IModelDiscoveryProvider.GetModelsAsync`	List models available from a provider at runtime
Provider catalog	`ProviderCatalog`	35+ embedded JSON provider definitions with metadata
Dependency injection	`AddAiProviders()`	One-line registration of every catalog provider

Architecture at a Glance¶

All runtime work goes through IAIProvider; metadata comes from the catalog. Five concrete providers cover every wire protocol in the catalog:

graph LR
    A[IAIProvider] --> B[AIProviderBase]
    B --> C[OpenAICompatibleProviderBase]
    B --> D[MessagesApiProvider]
    B --> E[KeyQueryProvider]
    C --> F[OpenAICompatibleProvider]
    C --> H[ModelCatalogProvider]
    I[ProviderCatalog] --> F
    I --> H
    I --> D
    I --> E

Every failure surfaces as an AiException with a stable error code — see Error Handling.

Where to Start¶

New to the library? Follow Installation and the Quick Start.
Wondering how providers are wired up? Read Wire Protocols and Dependency Injection.
Looking for a specific type? Check the API Reference.

License¶

MIT — see the LICENSE.txt in the repository.