Customization

The embedding engine can be extended at two points today:

• Writing a Custom Embedding Provider: talk to an endpoint whose request or response shape the shipped provider does not cover, or add your own preprocessing, authentication or batching rules.
• Registering a Custom Embedding Usage: add a search capability of your own to the registry that backs usages.<name>, the Usages tab, and the config validator.

Everything else — which model embeds what, how text is composed, chunk sizes, storage and compression — is configuration rather than code. See the Configuration reference.