TiDB Vector Search Public Beta

At PingCAP, we aim to make TiDB your go-to unified database, handling any workload seamlessly. We’ve already excelled in managing operational and transactional workloads. Now, we’re taking a significant step towards making TiDB the AI database of choice by launching built-in vector search in Public Beta. This allows you to develop AI applications directly with TiDB, eliminating the need for additional databases or technical stacks.

Role of Vector Search in AI Applications

Since the advent of GenAI and LLM technologies, organizations are eager to leverage AI’s potential. AI applications require high-dimensional data to capture and represent complex patterns and relationships effectively. Traditional keyword-based searches lack the ability to understand context and nuances in data. Vector search, however, excels in handling high-dimensional data, enabling advanced AI capabilities such as semantic search, recommendation systems, and image recognition.

Semantic Search: Vector search forms the foundation for semantic search. By understanding the inherent meaning behind data, it allows AI applications to retrieve information based on relevance, not just exact keywords

Retrieval-Augmented Generation (RAG): Vector search helps retrieve relevant context from vast amounts of text data, which the generative model then uses to craft its response.

Recommendation Engines: Vector search enables AI to analyze a user’s past interactions and purchases and recommend similar items based on the underlying user preferences captured in vector form.

Built-in Vector Search within TiDB

Historically, specialized databases have emerged to meet specific data storage and retrieval needs, such as NoSQL for unstructured data and graph databases for interconnected data. This trend led to the creation of vector databases for efficiently handling high-dimensional data crucial for AI applications. However, the proliferation of these specialized databases has resulted in database sprawl and increased complexity. By integrating vector search within TiDB, we offer a unified solution that combines traditional SQL functionalities with advanced vector capabilities, simplifying your architecture and meeting diverse needs more efficiently.

Try TiDB Serverless

Benefits of TiDB Vector Search

The benefits of using vector search within TiDB are substantial and distinct. Almost every database now integrates vector search, but TiDB stands out due to several key advantages. By innovatively introducing the vector data type into our distinctive storage engines and implementing similarity search indexes and algorithms like Hierarchical Navigable Small World (HNSW), we have enabled efficient storage, indexing, and retrieval of vector data in TiDB.

Dynamic Scalability to handle AI Demands: TiDB offers elastic scalability for the dynamic and unpredictable data requirements of AI applications. It can effortlessly handle billions of vectors without any degradation in performance. This guarantees both efficiency and economical operation and supports extensive data processing, essential for large language models and AI applications.

Unified Database: In addition to supporting Hybrid Transactional/Analytical Processing (HTAP), TiDB can now be used as a vector store. With this unification, TiDB can now handle GenAI applications’ document, graph and messaging specific scenarios by utilizing our SQL and Vector stores.

SQL Compatible: Developers can leverage the familiar SQL environment to effortlessly join, index, and query both operational and vector data. This capability enables advanced semantic searches, combining the power of vector search with the reliability and ease of MySQL.

In addition to these benefits, TiDB Serverless significantly reduces operational overhead and costs. Users pay only for the resources they actually use, making it a cost-effective solution for managing large vector datasets. This pay-as-you-go pricing model, combined with the system’s dynamic scalability, ensures that developers can efficiently manage their vector data without incurring unnecessary expenses.

The Journey Ahead

As we look to the future, TiDB aims to evolve from vector search to becoming your full-fledged AI infrastructure partner. This journey will involve native integrations with popular AI vendors and frameworks, enabling comprehensive Retrieval-Augmented Generation (RAG) processes—from retrieval and chunking to indexing and beyond. Our ultimate goal is to make the database invisible to developers, providing a seamless API layer that empowers you to focus on innovation. Be part of this exciting transformation by trying out our public beta at https://www.pingcap.com/ai. We can’t wait to see what groundbreaking solutions you’ll build with TiDB’s vector search capabilities.