Skip to main content
The VectorAI DB architecture consists of multiple layers that work together to provide high-performance vector operations. Client applications interact with the system through SDKs, which communicate with the Vector Data Engine (VDE) via gRPC. The VDE coordinates between the index manager, storage manager, ANN algorithms, and the underlying database engine for efficient vector storage and retrieval. VectorAI DB architecture diagram

​
Architecture components

1

Cortex SDKs

Client libraries (such as Python) communicate with the Cortex Server via gRPC to create collections, insert vectors, and run similarity searches. The SDK abstracts the complexity of gRPC communication and provides an intuitive API for developers.
2

Vector data engine (VDE) API layer

Receives gRPC requests from SDKs, validates and processes them, then routes them to the appropriate internal components in the VDE core. This layer serves as the gateway between client applications and the vector processing engine.
3

Storage manager

Handles persistent storage and retrieval of vectors and metadata while coordinating with the underlying database engine for durability and consistency. This component ensures data integrity across all vector operations.
4

Index manager

Manages vector indexes by building, updating, and querying them for efficient similarity search. The index manager coordinates with ANN algorithms to optimize search performance.
5

Approximate nearest neighbor (ANN)

Uses an in-process vector index optimized for fast similarity search over high-dimensional embeddings. This component trades exact precision for performance, delivering sub-millisecond query times on large datasets.
6

Database engine

Stores vector data, metadata, and collection information durably on disk. This underlying database engine provides ACID compliance and reliability for all vector operations.