Lodestar.Extensions.AI 0.1.0. This page is frozen at that release. Read the current documentation for what main says now. A link to a decision or a migration page follows main, and leaves the archive.

One type, OnnxEmbeddingGenerator: it presents the ONNX embedding path as Microsoft.Extensions.AI's IEmbeddingGenerator<string, Embedding<float>>, so a consumer written against that abstraction — a Semantic Kernel pipeline, a Microsoft.Extensions.AI chain, anything that takes the interface — can hold Lodestar embeddings without knowing Lodestar.

This package adds no arithmetic. Every vector it returns comes from OnnxTextEmbedder.EmbedBatch, unchanged, so what it owes is the interface contract and not a numerical one. That is why it carries no oracle corpus of its own: its suite asserts that its output is identical to the embedder's, and the embedder's own frozen corpora say what those numbers are.

Weights are never committed to this repository. A running example would need a model of tens of megabytes, and decisions/0003 rules that out; tools/fetch_*.py pulls vocabularies against a pinned SHA-256 when they are needed, and weights are not among them.

So the fences on these pages compile against the packed package and are marked docs-run: skip, which is what that marker is for — the same arrangement Lodestar.Onnx makes, and for the same reason.

Lodestar.Extensions.AI is the second member of the satellite tier, after Lodestar.Onnx. It exists because a core package carries no external dependency and an external dependency earns its own package, named for it — decisions/0076. A caller who wants embeddings and not the Microsoft.Extensions.AI abstractions takes Lodestar.Onnx alone and restores nothing extra.

Type	What it is
OnnxEmbeddingGenerator	Presents an OnnxTextEmbedder as an IEmbeddingGenerator<string, Embedding<float>>.

• ONNX inference — the package this one adapts.
• Semantic search with embeddings — the chain it sits in.
• Python → C# equivalence.