> ## Documentation Index
> Fetch the complete documentation index at: https://handbook.aiql.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Embeddings

> Vectors for retrieval and schema matching

The **embedding** model turns text into vectors. Queries, schema names, and extracted instances are embedded so later steps can match meaning, not only identical strings.

Answer models use these vectors on the [Middleware API](/products/middleware). Schema refinement also uses them to find overlapping classes.

## Configuration

| Setting          | How it is configured                                                                                  |
| ---------------- | ----------------------------------------------------------------------------------------------------- |
| What is embedded | Queries, class and property names, and instance labels in the workspace.                              |
| Similarity       | Cosine similarity.                                                                                    |
| Retrieval        | Hits are ranked literal first, then lexical, then semantic. An embedding-only hit is last, not first. |
| Schema           | Near-duplicate classes and properties are proposed for merge using the same vectors.                  |
| Serving          | AiQL selects the checkpoint and vector size. Customers do not pick an embedding model.                |