Overview

Vector Retrieval Service is developed based on the kernel of Alibaba Cloud’s vector engine Proxima. It provides efficient vector retrieval services that are fully managed on the cloud-native architecture, and can be horizontally and vertically scaled to cater to business needs. Vector Retrieval Service provides powerful vector management and retrieval capabilities through simple APIs and SDK, so you can easily integrate it with applications of various scenarios, such as intelligent Q&A and multimodal search.

Benefits

• Fully-Managed Service Quickly integrate fully-managed cloud-native vector retrieval service for your business with low O&M costs on the serverless architecture, and only pay for data consumption
• Designed for Vector Search Fulfill diverse search needs from various business scenarios with rich features that support conditional filtering, data partitions, different search types, multimodal data retrieval, etc.
• Scale-Performance Balance Adjust service capacity and QPS flexibly according to business needs by scaling clusters or search services in easy steps to achieve the balance of scale, accuracy, and performance
• Low Code Start vector data management and search services easily with easy-to-use APIs that require minimal coding and out-of-the-box SDK

Features

Fully Managed Vector Retrieval Services on the Cloud

Highly Accurate and Efficient Search Vector Retrieval Service integrates Alibaba Cloud's vector search engine Proxima that provides high-performance algorithms to achieve low-latency search of large-scale data. Low O&M Costs The fully managed, cloud-native vector search service with horizontal and vertical scalability reduces O&M costs, enabling you to focus only on business needs without concerns about the underlying architecture. Minimalist SDK Design Low-code APIs and easy-to-use SDK support rapid integration with AI applications of various business scenarios.

Real-Time Indexing of Vector Data

Indexing with Streaming Data Vector Retrieval Service adopts a flat index architecture and supports online indexing of large-scale streaming data from scratch. Real-time Online Update When vector data is added, deleted, or modified, the vector status takes effect instantly. Vector data is checked immediately after being added, and written in hard drives in real-time. Vector data status is updated in real-time too. Fast Indexing of Massive Data Vector Retrieval Service optimizes the index structure and loading method in various ways, and supports large-scale vector data import from 2 to 20,000 dimensions.

Filtered Search

Customizable Schema For filtered search, Vector Retrieval Service uses pre-defined fields to achieve higher retrieval speed and less computing power consumption. Filtered Search with Multiple Expressions You can perform combination search with comparison, logic, and string operators including "<", "<=", "=", "!=", ">=", ">", "and", "or", and "like".

Sparse Vector

Keyword Search and Hybrid Search You can perform keyword search, vector search, or hybrid search (keyword+vector) with Vector Retrieval Service, which supports both sparse vectors with dense vectors to balance semantics and keywords. Sparse Vector Generator We recommend using DashText from Vector Retrieval Service for sparse vector encoding. DashText uses BM25 algorithm to convert raw text into sparse vector data, greatly simplifying the process of keyword-based vector search.