Introduction to Vector Support

This chapter introduces you to vector support and its supported features in Oracle Backend with Firebase APIs.

• Overview

• Supported Features

Overview

Vectors are how AI models represent meaning. A piece of text, an image, an audio clip - a model turns each one into a long list of numbers called an embedding. Two pieces of content with similar meaning end up with similar embeddings, so you can find related items by looking for vectors that sit close together in space. That’s what powers semantic search, recommendations, and Retrieval-Augmented Generation (RAG)-style chatbots.

With Oracle Backend with Firebase APIs, you can store your vectors right next to your existing app data, in the same database, using the same SDK you already use. You can run a similarity search the same way you’d run any other query without having to manage a separate vector based system.

The vector DBMS_BAAS service enables storage, indexing, and querying of high-dimensional vector embeddings within a managed backend-as-a-service environment.

It integrates vector capabilities into Oracle Backend with Firebase APIs, so users can:

• Store high-dimensional vectors alongside structured data

• Perform similarity searches (for example, cosine, Euclidean distance)

• Build AI-powered applications directly on database infrastructure

This service is usually designed to work with LLM pipelines, enabling tight coupling between data storage and AI inference workflows.

Supported Features

The following features are supported for working with vector data, from data type handling and embedding generation to indexing, data sources, and model support.

Vector Data Types

• Dense vectors

• Sparse vectors

Embedding Generation Modes

• BYOV (Bring Your Own Vector): You generate the embedding yourself and pass it to the SDK.

• AUTO_TRIGGER: The database generates the embedding synchronously, as part of your insert.

• Auto Scheduler: The database generates the embedding asynchronously, in the background.

Embedding Sources (When Not Using BYOV)

When generating embeddings, the service can use one of the following data sources:

• The full document

• A specific field within the document, addressed by a dot-separated path

• A URL pointing to a file in DBFS or OCI Object Storage

Embedding Formats

• INT8

• FLOAT32

• FLOAT64

• BINARY

Dimensions

• FLEX (flexible variable length)

• FIXED (specify the exact number of dimensions as detailed by the model)

Model Providers

Embedding generation supports the following model types:

• ONNX - load your own ONNX model into the database

• OCIGENAI - use Oracle Cloud’s hosted embedding models

Indexing

• HNSW (Hierarchical Navigable Small World)

• IVF (Inverted File)