Oracle AI Vector Search
Easily bring AI-powered similarity search to your business data without managing and integrating multiple databases or compromising functionality, security, and consistency. AI Vector Search enables searching both structured and unstructured data by semantics or meaning, and by values, enabling ultra-sophisticated AI search applications. Native AI vector search capabilities can also help large language models (LLMs) deliver more accurate and contextually relevant results for enterprise use cases using retrieval-augmented generation (RAG) on your business data.
General
Vector data type
With Oracle AI Database 26ai, you can use the new
native vector data type to store vectors directly within tables. Support for vectors
with different dimension counts and formats mean that you can use the vector embedding
model of your choice to simplify application development and deployment.
Create Tables Using the VECTOR Data Type
Flexible vector generation
Use the ONNX framework to import the
embedding models of your choice and use them to generate vectors for your data or use
database APIs to generate vectors from your preferred embedding service—including NVIDIA
NIM containers. You also have the option of importing vectors directly into the
database.
Import Pretrained Models in ONNX Format for Vector Generation Within the Database
Simple standard SQL for querying vectors
Use simple, intuitive SQL
to perform similarity search on vectors and freely combine vectors with relational,
text, JSON, and other data types within the same query.
Use SQL Functions for Vector Operations
Seamless combination of AI vector data with business data
Combine
sophisticated business data search with AI vector similarity search using simple,
intuitive SQL and the full power of a converged database—JSON, graph, text, relational,
spatial, and more—all within a single query.
JSON Compatibility with the VECTOR Data Type
Vector indexes
Accelerate similarity searches using highly accurate
approximate search indexes (vector indexes), such as the in-memory neighbor graph index
for maximum performance and neighbor partition indexes for massive data sets.
Manage the Different Categories of Vector Indexes
Hybrid vector indexes
Index and query documents using a combination
of full-text search and semantic vector search to improve the overall search experience
and provide users with more-accurate information.
Simple search accuracy specification
Specify target search
accuracy as a simple percentage instead of being required to specify advanced
algorithmic parameters. Define default accuracy during index creation and override in
search queries if needed.
Determine the accuracy of your vector indexes
Powering retrieval-augmented generation
Enhance large language model
(LLM) interactions by providing context-specific private data to improve the accuracy of
responses through a combination of similarity search and business data search. Further
enrich retrieval-augmented generation (RAG) using built-in business criteria, such as
security filters, business metrics, and business rules.
Use Retrieval-Augmented Generation to Complement LLMs
Industry-leading security
Oracle AI Vector Search integrates
seamlessly with Oracle's industry-leading database security features to reduce risk and
simplify compliance. By leveraging robust tools, such as encryption, data masking,
privileged user access controls, activity monitoring, and auditing, organizations can
secure their data while taking full advantage of advanced AI search capabilities.
Support for a full generative AI pipeline
Perform all aspects of the
generative AI pipeline using native database APIs from end to end, making it easier for
developers to build next-gen AI applications using their business data—all from directly
within the database.
Oracle AI Vector Search Integration with LlamaIndex
AI Vector Search with a full machine learning suite
Handle a wide
range of AI use cases involving machine learning actions (decisions, predictions,
classification, forecasts, and so on) combined with the power of AI-based vector search.
For instance, it is easy to combine inference and classification with Oracle AI Vector
Search within the same SQL query.
Exadata optimizations
Accelerate vector index creation and search
starting with Exadata System Software 24ai optimizations. Gain the high performance,
scale, and availability Exadata provides to enterprise databases.
Learn more about Exadata AI optimizations
Elastic Vector Memory
Hierarchical navigable small world (HNSW)
vector indexes reside in database memory. Creating an HNSW index depends on the memory
being adequately sized to accommodate it, and subsequent changes to the underlying base
table can change the size of the HNSW index. With Elastic Vector Memory, the database
memory automatically resizes to accommodate all HNSW indexes dynamically.
Learn more about Elastic Vector Memory
Custom vector distance functions
Vector search operations are often
based on standard distance metrics, such as Euclidean, Cosine, and dot product. However,
in some situations, domain-specific or proprietary metrics are required. User-defined
vector distance functions allow users to create their own custom metrics.
Custom vector distance functions
Support for sparse vectors
Sparse vectors are vectors that typically
have a large number of dimensions, but only a few of those dimensions have non-zero
values. Because sparse vectors only store non-zero values, their use can improve
efficiency and save storage space. Native support in PL/SQL allows sparse vectors to be
created and used directly from within PL/SQL. Sparse vectors can also be indexed using
hierarchical navigable small world (HNSW) vector indexes, providing all the benefits of
speed, accuracy, and space optimization combined into one unit.
Learn more about sparse vectors
AI Vector Search with Iceberg Tables
With Oracle AI Database and AI
Vector Search, you can run similarity search over vector embeddings stored
externally—without first loading the embeddings or their source text chunks into the
database. This reduces data duplication, shortens time-to-value, and keeps data closer
to its existing governance controls.