K2HVEC

K2HVEC is a high-performance, KVS-based vector database built on top of the K2HASH Key-Value Store library.
It uses vector embeddings(floating-point arrays generated by AI or machine learning models) as keys to store arbitrary data, delivering high-speed and accurate vector similarity searches.

By fully inheriting the powerful features, performance, and attributes of K2HASH, K2HVEC provides a secure and feature-rich vector search environment equipped with capabilities like data encryption and expiration(TTL) management.

Background and Objectives

In recent years, while the use of AI(such as LLMs) has expanded rapidly, reducing response latency and managing rising costs associated with increased API calls have become critical challenges.

To address these challenges, we devised an approach to cache and store pairs of AI-generated vector embeddings (floating-point arrays) and their corresponding responses (arbitrary data) either temporarily or permanently. By utilizing fast vector similarity search, past responses can be quickly retrieved and reused.

To implement this system, the following requirements had to be met:

• High-Speed Processing of Large and Variable-Length Data:
  Because returned data(values) can be large and variable in size, the system requires the ability to retain large volumes of data while delivering high-speed read/write performance.
• Enterprise-Grade Data Management Features:
  Security(encryption) for stored data and mechanisms to automatically delete or refresh data after a specified period(Expiration / TTL) are essential.

Therefore, by leveraging the proven, high-performance KVS K2HASH as our foundational core, we developed K2HVEC (Vector Database and Search built on K2HASH). This K2HVEC is a vector database that satisfies these requirements by inheriting high-speed I/O for large data alongside built-in attribute management like encryption and TTL.