Save pipeline state to file.

Pure text query - directly calls LightRAG's query functionality Args: query: Query text mode: Query mode ("l

Abstract method to check if the parser is properly installed. Must be implemented by subclasses. Returns: bool:

Mark multimodal content processing as complete in the document status.

Truncate context to maximum token limit Args: context: Context text to truncate Returns: Truncated context t

Get base directory for project.

Extract project name from input path.

Get path to summary checkpoint file.

Load data from JSON file.

Create entity and text chunk

Extract all possible JSON candidates from response

Get context for current processing item Args: item_info: Information about current item (page_idx, index, etc.) Returns:

Robust JSON parsing with multiple fallback strategies

Get mode-specific directory (fast or normal).

Get path to RAG checkpoint file.

Initialize the RAG system. Called automatically by context manager.

Parse PDF document using MinerU 2.0 Args: pdf_path: Path to the PDF file output_dir: Output directory path

Save data to JSON file.

Initialize the base parser.

Apply the appropriate chunk template based on content type Args: content_type: Type of content (image, table, equation,

Call image generation provider based on configuration.

Create context configuration from RAGAnything config

Filter images used in given sections.

Format a single section as markdown.

Get API kwargs, only including base_url if it's set.

Read the output files generated by docling and convert to MinerU format Args: output_dir: Output directory f

Run docling command line tool Args: input_path: Path to input file or directory output_dir: Output directory

Try to parse JSON string, return None if failed

Finalize all storages including parse cache and LightRAG storages This method should be called when shutting down to properly clean up resour

Get config-specific directory for plan and output.

Get figure info with full paths for JSON serialization.

Get path to plan checkpoint file.

Get appropriate processor based on content type Args: modal_processors: Dictionary of available processors content_type: Con

Get table info for JSON serialization.

Parse and index a document. Args: file_path: Path to document (PDF, DOC, etc.) output_dir: Directory

Check if content is from a paper document.

Load pipeline state from file.

Process multiple files in parallel Args: file_paths: List of file paths or directories to process output_dir

Query the paper content. Args: question: Question about the paper. mode: Query mode - "local", "glob

Separate text content and multimodal content Args: content_list: Content list from MinerU parsing Returns: (text_conten

Set content source for context extraction in all modal processors Args: content_source: Source content for context extraction (e.

Convert to dictionary for JSON serialization.

Build the tables/figures section based on available assets.

Build prompt for slide with layout rules and consistency.

Call multimodal LLM with text and images inline.

Create context extractor with tokenizer from LightRAG

Encode image to base64

Extract context from MinerU-style content list Args: content_list: List of content items with page_idx and type info

Extract context from plain text source Args: text_source: Plain text content current_item_info: Current item informat

Extract context based on page boundaries Args: content_list: List of content items current_item_info: Current item wi

Extract structured content for a single section using LLM. Args: content: Merged RAG content for the section section: Section

Extract text content from a content item Args: item: Content item dictionary Returns: Extracted text content

Return minimal fallback sections if parsing fails.

Generate doc_id based on document content Args: content_list: Parsed content list Returns: str: Con

Get recommended backend based on availability

Initialize multimodal processors with appropriate model functions

Parse LLM response into Section objects. Args: llm_response: The LLM response containing JSON is_slides: If T

Process multimodal content (using specialized processors) Args: multimodal_items: List of multimodal items f

Progressive fixing of quote and escape issues

Read the output files generated by mineru Args: output_dir: Output directory file_stem: File name without ex

Run mineru command line tool Args: input_path: Path to input file or directory output_dir: Output directory

Truncate text to max length.

Cleanup resources when object is destroyed

Convert using Pandoc (best for complex documents)

Create initial pipeline state.

Detect which stage to start from based on existing checkpoints.

Extract tables and figures from original markdown file. This function searches for HTML tables and images, then looks for their capt

Filter file paths to only include supported file types Args: file_paths: List of file paths or directories r

Generate queries for general document analysis using LLM. Args: rag_client: RAG client for API config overview: Document

Get page count range for slides.

Get the currently running session ID

Get path to state file.

Get the full summary text.

Get list of supported file extensions

Get all tables as markdown, sorted by original position.

Insert pure text content into LightRAG Args: lightrag: LightRAG instance input: Single document string or list of document s

Check if a session has been cancelled

Log a section separator for better readability.

Main entry point for Paper2Slides CLI.

Merge all RAG answers for a section. Args: rag_results: RAG query results section: Section name to merge clean_refs:

Abstract method to parse a document. Must be implemented by subclasses. Args: file_path: Path to the file to be