Skip to content

Supported File Formats

Flexible GraphRAG supports a wide range of document, image, and audio formats through its two parser options: Docling (local, default) and LlamaParse (cloud API).

Document Formats

Format Extensions Docling LlamaParse
PDF .pdf Advanced layout analysis, table extraction, formula recognition, configurable OCR (EasyOCR, Tesseract, RapidOCR) Automatic OCR, multimodal vision processing
Microsoft Office .docx, .xlsx, .pptx, .doc, .xls, .ppt DOCX, XLSX, PPTX structure preservation Full Office suite including legacy formats and hundreds of variants
Web .html, .htm, .xhtml HTML/XHTML markup structure analysis HTML/XHTML content extraction and formatting
Data .csv, .tsv, .json, .xml CSV structured data processing CSV, TSV, JSON, XML with enhanced table understanding
Documentation .md, .markdown, .asciidoc, .adoc, .rtf, .txt, .epub Markdown, AsciiDoc with markup preservation Extended format support including RTF, EPUB, and hundreds of text format variants

Image Formats

Format Extensions Docling LlamaParse
Standard Images .png, .jpg, .jpeg, .gif, .bmp, .webp, .tiff, .tif OCR text extraction with configurable backends (EasyOCR, Tesseract, RapidOCR) Automatic OCR with multimodal vision processing

Audio Formats

Format Extensions Docling LlamaParse
Audio .wav, .mp3, .mp4, .m4a Automatic speech recognition (ASR) Transcription for MP3, MP4, MPEG, MPGA, M4A, WAV, WEBM

Parser Comparison

Docling (default, free, local)

  • Local processing — no API costs, no data sent to third parties
  • Specialized CV models: DocLayNet layout analysis, TableFormer for tables
  • Configurable OCR backends: EasyOCR, Tesseract, RapidOCR
  • Optional local VLM support: Granite-Docling, SmolDocling, Qwen2.5-VL, Pixtral
  • GPU acceleration supported (CUDA / Apple Silicon) — see Docling GPU Configuration
  • Output formats: Markdown, JSON (lossless with bounding boxes), HTML, plain text, DocTags

LlamaParse (cloud API, 3 credits/page by default)

  • Cloud-based with advanced AI, multimodal parsing with Claude Sonnet 3.5
  • Supports hundreds of file format variants
  • Three modes: parse_page_without_llm (1 credit), parse_page_with_llm (3 credits, default), parse_page_with_agent (10–90 credits)
  • Multimodal mode: bring your own API keys (OpenAI GPT-4o, Anthropic Claude, Google Gemini, Azure OpenAI)
  • Output formats: Markdown, plain text, raw JSON, XLSX (extracted tables), PDF, images, structured output (beta)
  • Get your API key at LlamaCloud

Output Format Selection

Flexible GraphRAG saves both markdown and plaintext from the parser, then automatically selects which to use for knowledge graph extraction, vector embeddings, and search indexing:

  • Markdown — preferred for documents with tables
  • Plaintext — preferred for text-heavy documents

Override with PARSER_FORMAT_FOR_EXTRACTION=auto|markdown|plaintext in .env.

Save intermediate parsed output for inspection with SAVE_PARSING_OUTPUT=true.

See Parser Output Files for details on the saved files.