Vector Search 2.0

Preview

This feature is subject to the "Pre-GA Offerings Terms" in the General Service Terms section of the Service Specific Terms. Pre-GA features are available "as is" and might have limited support. For more information, see the launch stage descriptions.

Vector Search 2.0 is a Google Cloud product designed from the ground up as a self-tuning, fully managed vector database. While Google Cloud's existing Vector Search is a powerful approximate nearest neighbor (ANN) index-as-a-service system, Vector Search 2.0 evolves this concept into a comprehensive storage and retrieval system. Instead of managing indexes as the primary resource, you'll work with Collections of Data Objects.

The vector database architecture provides a replicated, scalable storage engine, making Vector Search 2.0 a single, unified data source for your AI applications and removing the need for auxiliary data storage.

Key benefits of this new architecture include:

• Developer Friendly: Get started quickly with intuitive client libraries that require minimal code. The system is auto-tuned to maintain high performance, abstracting away the underlying infrastructure so you don't have to configure VMs or replicas.

• Fast Onboarding & Evaluation: Create Collections, add your data, and start searching quickly.

• Unified Data Storage: Store, retrieve, and filter your documents by vector similarity and payload data, all in one place.

• Powerful Features: Automatically populate embedding fields using built-in models, explore your data with rich query capabilities, bring your own embeddings (BYOE), and quickly create indexes to scale performance.

• Simplified Pricing: Adaptable pricing features two models: usage-based for smaller workloads and resource-based for tuned performance.

Vector Search 2.0 maintains the high performance and massive scalability available in Vector Search 1.0, making it seamless to get started and scale.

Concepts

Before you begin, it's helpful to understand the following Vector Search 2.0 concepts:

• Collection: A container for a set of related JSON objects. This is similar to a table in a relational database. You can create many Collections within a single database.

• Data Object: An individual JSON object stored within a Collection.

• Collection: schema: Defines the structure and constraints of the Data Objects within a Collection. It can be configured for both strict and relaxed schema validations.

• Collection Index: Enables efficient approximate nearest neighbor (ANN) search across Data Objects within a Collection. A Collection can have multiple Indexes, such as one for each vector field in your Data Objects.

What's next?

• Learn how to create a Collection.