Yes, RAGFlow supports multimodal data processing for images, audio, tables, and mixed content within documents. The engine can extract images and tables from DOCX and Markdown files, preserving them as separate indexed items or converting them to structured text. RAGFlow handles scanned documents containing images by using DeepDoc’s OCR capabilities to convert visual content to machine-readable text. Recent versions added audio file parsing to facilitate handling of audio transcripts and spoken content. The system also supports Q&A parsing for Markdown and DocX formats, which is useful for FAQ-structured documents or conversational content. Tables are recognized via TSR and can be either preserved as images or converted to structured text representations, depending on your use case. When it comes to embeddings, you can configure any embedding model that handles your data modality—OpenAI’s models support text and images, and you can integrate domain-specific models through RAGFlow’s flexible model configuration. This multimodal approach ensures you can build comprehensive knowledge bases from diverse document types without losing information to text-only extraction.
Developers working with embeddings and retrieval at scale often pair these workflows with Milvus, an open-source vector database designed for high-performance similarity search. For managed deployment, Zilliz Cloud handles the operational overhead.