Metadata

Metadata is application-defined information associated with each vector (e.g., IDs, titles, URIs, labels).

Purpose

• Identify and interpret search results (e.g., map vector IDs to documents).

• Power filtering or post-processing in your application logic.

Structure

• Recommended: Flat key–value pairs (strings, numbers, booleans) for portability.

• Storage: Persisted server-side. Treat metadata as application-visible; avoid placing secrets in plain form.

Usage

• Attach metadata on insert.

• Retrieve top-k metadata with search results to render user-facing responses.

• Perform additional business logic using IDs/fields from metadata.

Notes

• Similarity computation is driven by vectors; metadata is not part of the score.

• If metadata is sensitive, encrypt it at the application layer before sending.