Record

Understand the record components that Oracle VecDB stores and retrieves during vector search.

A record is one row in a vector table. It represents one retrievable unit and consists of a record ID, a dense vector, and metadata.

Record ID

A record ID uniquely identifies a record. You can supply it or generate it automatically when you create the vector table with auto_generate_id set to true.

Dense Vector

A dense vector, sometimes called an embedding, is an ordered list of numeric values that captures the meaning of data. Similar items have vectors that are close together in a high-dimensional space. Every dimension stores a value, including zero values.

Oracle AI Database can generate the embedding with an integrated model, or your application can generate it with an external embedding model and upsert the resulting vector.

Metadata

Metadata is structured information stored alongside vectors. It supports the following:

• Filtering by attributes such as tenant, user, region, document type, or recency.
• More accurate and context-aware results by combining vector similarity with attribute conditions.
• Governance patterns, such as TTL and retention fields.

Oracle VecDB supports vector similarity queries and metadata filtering in the same retrieval flow.

Parent topic: How Oracle VecDB Works