MariaDB Vector DocumentStore Integration for Haystack

[UmaisMu]

Title: RFC: MariaDB Vector DocumentStore Integration for Haystack

##Summary I'm participating in the MariaDB Python Hackathon and planning to build a MariaDB Vector DocumentStore integration for Haystack 2.x. I'd appreciate feedback from the community.

Background

• MariaDB 11.7+ now supports vector operations with VECTOR datatype and HNSW indexing
• Currently no Haystack DocumentStore implementation exists for MariaDB
• This would complete MariaDB's AI framework ecosystem alongside existing LangChain/LlamaIndex integrations

Proposed Implementation

• Full DocumentStore protocol compliance (write_documents, filter_documents, delete_documents, count_documents)
• Vector similarity search using MariaDB's VEC_DISTANCE functions
• Metadata filtering combining SQL queries with vector search
• Production-ready features: connection pooling, error handling, batch operations

Questions for Community

1. Would this integration be valuable for Haystack users?
2. Any specific implementation requirements or preferences?
3. Should this be part of haystack-integrations or separate package?
4. Any guidance on testing or documentation standards?

Timeline

October 2025 development as part of MariaDB hackathon, targeting production-ready PyPI package.

Looking forward to your thoughts and guidance!