Oracle AI Vector Search & Native AI: The Converged Database Era

Oracle has introduced a major shift in its database strategy with the release of Oracle 23ai, a converged database that natively supports AI capabilities. This new approach eliminates the need for separate AI systems and simplifies the integration of AI into applications. The integrated vector data type allows for efficient storage and manipulation of embeddings, while the VECTOR data type enables SQL queries to directly work with vector data without the need for complex parsing.

Oracle supports three vector formats: INT8, FLOAT32, and FLOAT64, with different memory optimizations based on the chosen format. Additionally, Oracle offers two vector index types: HNSW (Hierarchical Navigable Small World) for in-memory high accuracy searches and IVF (Inverted File Index) for disk-based massive datasets. The TARGET ACCURACY parameter simplifies the process by allowing users to specify a desired recall rate, and Oracle automatically optimizes the index accordingly.

Native SQL functions like VECTOR_DISTANCE enable natural language queries, offering a game-changing capability for developers and DBAs alike.