NPipeline separates AI features by capability and provider boundary. Install only the package that owns the behavior you need.
This page helps you choose a package. The linked package pages contain the complete configuration, execution, error-handling, and performance references.
Choose a package
| Package | Capability | Main dependency |
|---|---|---|
NPipeline.Extensions.AI | Shared metadata, usage, and exception contracts | NPipeline |
NPipeline.Extensions.AI.Chat | Chat-model transformations, enrichment, and batching | Microsoft.Extensions.AI.IChatClient |
NPipeline.Extensions.AI.Decisions | Provider-neutral typed classification and routing | NPipeline.Extensions.AI |
NPipeline.Extensions.AI.Decisions.Jev | TypeSafe AI System One client and Jev Choice routing | NPipeline.Extensions.AI.Decisions |
Most applications install Chat, Decisions, or Decisions.Jev. NuGet resolves the shared foundation transitively.
Understand the boundaries
Use Chat when a generative model must create structured output, rewrite data, summarize text, or enrich an item. These APIs use the common IChatClient abstraction.
Use Decisions when a model or service returns a typed label, confidence, and probability distribution. Routing retains the original item and doesn't require temporary AI fields in your domain model.
Use Decisions.Jev for near-real-time judgments through TypeSafe AI's Jev model. Jev uses its native System One contract instead of pretending to be a chat model.