Vald A Highly Scalable Distributed Vector Search Engine

What is Vald?

Vald is a highly scalable distributed vector search engine designed for fast approximate nearest neighbor (ANN) searches on dense vectors. Built on a Cloud-Native architecture, it leverages the efficient NGT algorithm for its core search functionality.

Vald is engineered to manage massive datasets, capable of handling billions of feature vectors through features like automatic vector indexing, index backup to Object Storage or Persistent Volume for disaster recovery, and seamless horizontal scaling. Its distributed index graph allows continuous operation even during indexing, avoiding system pauses. Furthermore, Vald supports index replication across multiple agents, ensuring high availability and automatic rebalancing if an agent fails. It also offers customizable ingress/egress filtering and is designed for ease of use and high configurability.

Features

Asynchronize Auto Indexing: Continues to work during indexing using a distributed index graph, avoiding stop-the-world.
Distributed Indexing: Distributes vector index across multiple agents.
Index Replication: Stores each index in multiple agents for redundancy and automatic rebalancing.
Auto Indexing Backup: Supports automatic backups to Object Storage or Persistent Volume for disaster recovery.
Horizontal Scalability: Scales memory and CPU resources horizontally based on demand.
Customizable Ingress/Egress Filtering: Allows configuration of filters for gRPC interfaces.
Cloud-Native Architecture: Designed based on cloud-native principles.
Multi-Language SDK Support: Provides SDKs for Golang, Java, Node.js, and Python.