MariaDB AI RAG 1.1.0 Release Notes

Release Date: 31 Mar 2026

Release 1.1.0 is a Beta release.

MariaDB AI RAG 1.1.0 introduces major improvements across the ingestion and retrieval pipeline—especially for complex document layouts and large document libraries.

MariaDB AI RAG is currently in beta and can be evaluated with a trial license. This release adds layout-aware text extraction, incremental S3 ingestion, cross-encoder reranking, Docker-based deployment, and answer citations.

New Features and Improvements

Layout-based text extraction (Built-in Docling + LlamaParse integration)

Improve ingestion quality for complex PDFs and other layout-heavy documents (tables, multi-column pages, headers/footers) using layout-aware extraction via Docling and LlamaParse. Cleaner extraction leads to better chunk boundaries and embeddings, and ultimately more accurate retrieval—especially when plain text extraction loses reading order or structure.

Bulk documents Ingestion from S3 storage with incremental sync

Ingest large document collections directly from S3-compatible object storage (bucket/prefix) and keep them up to date with incremental sync—processing only new or changed documents instead of re-ingesting everything. This turns S3 Storage into a scalable source of truth for your knowledge base while keeping indexing cost under control.

Cross-encoder reranker (Built-in Flashrank + Cohere integration)

Improve answer quality by reranking retrieved passages with a cross-encoder, so the system selects the most relevant context before generating a response. MariaDB AI RAG 1.1.0 supports Cohere reranking for managed, high-quality ranking and includes Flashrank as a built-in option for lightweight, local reranking—helping reduce hallucinations and boosting precision on ambiguous queries.

Docker deployment

Deploy MariaDB AI RAG as a containerized stack using Docker, so you can spin up all RAG pipeline components—including Docling for document parsing and the MCP Server—with a single, repeatable step. For quick evaluation, you can also use a MariaDB vector store deployed in a container alongside the AI RAG stack to get an end-to-end setup in minutes.

Citation of generated answer

Make responses more trustworthy by having citations alongside the generated answer—so users and applications can see which documents/chunks were used to generate each paragraph in the generated output.

Packaging

• MariaDB AI RAG 1.1.0 is delivered as a suite of Docker containers managed by Docker Compose.

Deprecation Note

Packager-based deployments (RPM, DEB, and MSI) provided in version 1.0.0 are no longer available.

Installation

• Follow the Deployment Guide in the MariaDB AI RAG documentation for installation and initial setup.