Skip to main content

How VectorAI works

VectorAI stores and organizes embeddings for fast similarity searches. Unlike traditional SQL databases that use exact matches, VectorAI finds the nearest vectors to identify related content and power intelligent recommendations.
1

Embed data into vectors

Transform raw data (text, images, audio) into numerical embeddings using ML models like BERT or CLIP. Each embedding is a fixed-length vector that preserves semantic meaning. Similar items cluster together in the vector space.
2

Store vectors in the database

Store generated vectors with their IDs and metadata in indexed collections. VectorAI supports multiple vector types and dimensions. This structure enables fast retrieval and filtering.
3

Embed queries into vectors

Convert user queries into vectors using the same ML model as your data. This ensures queries and data exist in the same vector space. Embeddings are generated in real time.
4

Run similarity searches

Find nearest neighbor vectors using distance metrics (cosine similarity, Euclidean distance, dot product). VectorAI uses Approximate Nearest Neighbor (ANN) algorithms for speed. Results are ranked by similarity score.
5

Use ANN algorithms for speed

HNSW (Hierarchical Navigable Small World) provides ultra-fast k-NN queries. IVF (Inverted File Index) offers memory-efficient indexing for large datasets. These algorithms balance speed with accuracy trade-offs.

Differences from traditional databases

The table below compares VectorAI DB with traditional databases across key aspects including data models, query patterns, and use cases.
AspectTraditional DBsVectorAI
Data modelRows of scalar fields (text, numbers, dates)High-dimensional vectors with metadata
Query patternEquality, range, or pattern matchingSimilarity search (nearest neighbors)
Search typeExact matches or structured queriesSemantic and contextual similarity
PerformanceOptimized for OLTP/OLAP workloadsOptimized for ANN search algorithms
Use caseTransactional data, reportingAI/ML applications, recommendations
IndexingB-trees, hash indexesHNSW, IVF, PQ for vector spaces
Role in AIData storage for ML trainingExternal knowledge store for embeddings

Next steps

Quickstart

Get started with VectorAI DB in under five minutes

Local installation

Install VectorAI DB using Docker

Key features

Explore the key features of VectorAI DB

Architecture

Understand the VectorAI DB architecture

Use cases

Explore common VectorAI DB use cases