> For the complete documentation index, see [llms.txt](https://mariadb.com/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://mariadb.com/docs/tools/enterprise/mariadb-ai-rag/getting-started/overview.md).

# Overview

MariaDB AI RAG is an enterprise-grade RAG solution built on MariaDB, with modular components for document ingestion, semantic chunking, vector storage, retrieval, and LLM generation.

## Introduction to MariaDB AI RAG

MariaDB AI RAG is an enterprise-grade Retrieval-Augmented Generation (RAG) solution that integrates with MariaDB to provide AI-powered document processing, semantic search, and natural language generation capabilities.

The system enables organizations to leverage their document repositories and databases for AI-powered search and generation. By combining the reliability of MariaDB with modern AI capabilities, AI RAG provides accurate, context-aware responses based on your organization's proprietary data.

## System Architecture

MariaDB AI RAG follows a modular architecture with the following key components:

### Core Components

1. **Document Ingestion Module**
   * Handles document uploads and text extraction
   * Supports multiple file formats (PDF, DOCX, TXT, etc.)
   * Processes metadata extraction
   * Manages document storage and versioning
2. **Chunking Module**
   * Divides documents into semantically meaningful segments
   * Supports multiple chunking strategies (fixed size, recursive, semantic)
   * Optimizes chunk size and overlap for retrieval quality
   * Preserves document structure and metadata
3. **Vector Database Integration**
   * Creates and stores vector embeddings for document chunks
   * Integrates with MariaDB's vector search capabilities
   * Enables efficient similarity search
   * Manages vector indexes and optimizations
4. **Retrieval Module**
   * Converts queries into vector embeddings
   * Performs semantic similarity search
   * Ranks and filters results based on relevance
   * Supports metadata-based filtering
5. **Generation Module**
   * Integrates with multiple LLM providers (OpenAI, Google, Anthropic)
   * Constructs prompts with retrieved context
   * Generates natural language responses
   * Supports streaming output for real-time display
6. **Orchestration Module**
   * Coordinates the end-to-end RAG pipeline
   * Manages asynchronous processing
   * Handles error recovery and retries
   * Provides status monitoring and reporting
7. **API Interface**
   * Exposes RESTful endpoints for all functionality
   * Implements JWT-based authentication
   * Manages user permissions and access control
   * Provides comprehensive error handling and logging

### Architecture Diagram

```mermaid
---
config:
  theme: neutral
  layout: dagre
---
flowchart TB
    accTitle: RAG pipeline from document ingestion to response
    accDescr {
        A top-to-bottom chain of seven boxes joined by arrows. Document Ingestion
        leads to Chunking Module, which leads to Vector Database. Vector Database
        leads to Query Processing, which leads to Retrieval Module. Retrieval
        Module leads to Generation Module, which leads to Response.
    }
    n1["Document Ingestion"] --> n2["Chunking Module"]
    n2 --> n3["Vector Database"]
    n3 --> n4["Query Processing"]
    n4 --> n5["Retrieval Module"]
    n5 --> n6["Generation Module"]
    n6 --> n7["Response"]
    n1@{ shape: manual-input}
    n3@{ shape: cyl}
    n7@{ shape: terminal}
    style n1 fill:#FFFFFF
    style n2 fill:#FFFFFF
    style n3 fill:#FFFFFF
    style n4 fill:#FFFFFF
    style n5 fill:#FFFFFF
    style n6 fill:#FFFFFF
    style n7 fill:#FFFFFF
```

## Data Flow

1. **Document Processing Flow**
   * Documents are uploaded via the API
   * Text is extracted based on document format
   * Documents are divided into chunks
   * Chunks are converted to vector embeddings
   * Embeddings are stored in the vector database
2. **Query Processing Flow**
   * User submits a query via the API
   * Query is converted to a vector embedding
   * Similar document chunks are retrieved
   * Retrieved chunks are ranked by relevance
   * Top chunks are used as context for generation
   * LLM generates a response based on the context
3. **Database Integration Flow**
   * Database table/view is selected for ingestion
   * Rows are processed as individual documents
   * Column mapping defines content and metadata
   * Text is chunked and embedded
   * Embeddings are stored with source metadata

## System Requirements

* **Database**: MariaDB 10.6+ or MySQL 8.0+
* **Operating System**: Linux (Debian/Ubuntu, RHEL/CentOS), Windows Server
* **Hardware**: Minimum 4 CPU cores, 8GB RAM, 100GB storage
* **Network**: Internet connectivity for external AI services
* **Dependencies**: Python 3.9+, FastAPI, SQLAlchemy, PyTorch (optional for local embeddings)

For detailed installation instructions, see the [Installation Guide](/docs/tools/enterprise/mariadb-ai-rag/getting-started/installation.md).

<sub>*This page is: Copyright © 2026 MariaDB. All rights reserved.*</sub>

{% @marketo/form formId="4316" %}