Feature Overview

The RAG artifacts generated by RHEL AI require a local vector database. For the initial use cases, the Milvus database has been selected as the internal vector database.

We aim to integrate Milvus Lite as a local vector database for RHEL AI, providing an efficient solution for local vector data management and search. This will be an internal implementation detail whose client API will not be exposed to the user.

Milvus Lite shares the same API with Milvus Standalone and Distributed, ensuring compatibility and ease of use. It covers essential features such as vector data persistence, CRUD operations, sparse and dense vector search, metadata filtering, and multi-vector and hybrid search. The API compatibility allows the future to support the BYO Milvus database so users can provide an external enterprise-grade instance they want to use by default.

Goals

• Provide a local vector database solution for RHEL AI users.
• Ensure compatibility with external enterprise Milvus APIs.
• Implement RAG artifacts with essential vector data management and search features.
• Target primary user type: RHEL AI developers and data scientists.

Requirements

To consider the feature complete, the following requirements must be met:

• Successful integration of Milvus Lite as a local vector database.
• Successful testing with local RAG for RHEL AI users.

Background

Milvus Lite is an open-source vector database built on Milvus, a popular open-source vector database, and offers a lightweight alternative for local use cases. Integrating Milvus Lite as a local vector database for RHEL AI will provide the required internal tools for creating, managing and searching vector data.

Done

To consider the feature delivered, the following checklist must be completed:

• Successful integration of Milvus Lite as a local vector database.
• Successful testing with local RAG for RHEL AI users.

Questions to Answer

• What are the specific requirements for integrating Milvus Lite as a local vector database?

Out of Scope

The following items are out of scope for this feature:

1. Integration with other vector databases or data sources.
2. Implementation of advanced analytics or machine learning features.
3. Development of a user interface for Milvus Lite.

Customer Considerations

When designing and delivering the feature, consider the following customer-specific considerations:

• Provide developers documentation for Milvus Lite as a local vector database for RHEL AI RAG pipelines
• Address any security or privacy concerns related to using Milvus Lite as a local vector database.