Skip to main content


Developing with AVS requires an understanding of three distinct components described here. This guide describes each component at a high level, with additional details about the search and storage layers provided later in this guide.


Application layer (not provided)​

The first component is the AI application layer, which is where embeddings are generated for your data. Note, this layer is not provided by Aerospike. You must create an application that uses a machine learning model to generate your embeddings. See Developing with Vector Search to learn more about using our Python client or check out our Example Apps.

Search layer (AVS nodes)

The search layer is a new, horizontally-scalable component that performs four basic functions. Each function is covered in more detail later in this guide.

  • Clustering - uses heartbeats to assemble a cluster across a shared network
  • Index Construction - creates an index for performing searches (see HNSW for more details)
  • Caching - builds a cache from the index based on each query
  • Query Execution - requires compute-intensive operations to perform distance calculations and index traversal

Storage layer (Aerospike DB nodes)

The storage layer is built on the Aerospike database and stores the following:

  • Records - these include both the vectors against which you are performing your searches and associated metadata
  • Index - the index used for performing searches

The Aerospike database excels with especially large datasets and you can tune Aerospike to optimize performance for your dataset. See Aerospike documentation for details about scaling and configuring your storage layer, and see Vector Data in Aerospike to learn about the AVS data model.