Vald

A Highly Scalable Distributed Vector Search Engine

Free

Description

Vald is engineered as a highly scalable distributed vector search engine, optimized for finding fast approximate nearest neighbors within dense vector datasets. Built on cloud-native principles, it leverages the NGT algorithm for efficient neighbor searching. Vald is designed to handle billions of feature vectors through automatic vector indexing, index backup capabilities, and horizontal scaling.

This search engine provides robust features such as asynchronous auto-indexing without stop-the-world pauses, thanks to its distributed index graph. It includes customizable ingress/egress filtering adaptable to gRPC interfaces and supports index replication for high availability. Vald is easy to install and offers high customizability for parameters like vector dimensions and replica counts, supported by SDKs in multiple programming languages.

Key Features

Asynchronize Auto Indexing: Performs indexing using a distributed index graph without stopping operations.
Customizable Ingress/Egress Filtering: Offers configurable filtering compatible with gRPC interfaces.
Cloud-native Architecture: Enables horizontal scaling of memory and CPU based on demand.
Auto Indexing Backup: Supports automatic backups to Object Storage or Persistent Volume for disaster recovery.
Distributed Indexing: Spreads the vector index across multiple agents.
Index Replication: Stores index replicas on multiple agents with automatic rebalancing.
Easy to Use: Simple installation process.
Highly Customizable: Allows configuration of vector dimension, replica count, and other parameters.
Multi-language SDKs: Supports Golang, Java, Nodejs, and Python.