LLMDocumentContentExtractor

Extract textual content from image-based documents using a vision-enabled LLM.

Key Features

• Converts image-based documents (such as scanned PDFs or images) to text using a vision LLM.
• Works with any vision-capable ChatGenerator.
• Processes documents in parallel using a configurable thread pool.
• Returns failed documents separately with content_extraction_error metadata for debugging or reprocessing.
• Supports optional image resizing to reduce memory usage and processing time.
• Configurable detail level for image processing (for OpenAI models).

Configuration

1. Drag the LLMDocumentContentExtractor component onto the canvas from the Component Library.
2. Click on the component to open the configuration panel.
3. On the General tab:
   • Select or configure the vision-capable chat generator (LLM) to use for extraction.
4. Go to the Advanced tab to configure additional settings:
   • Enter the prompt with instructions for the LLM on how to extract content from the document image. The prompt must not contain Jinja variables.
   • Set the file_path_meta_field to specify which metadata field contains the file path.
   • Optionally set the image detail level (auto, high, or low) if using an OpenAI model.
   • Optionally configure size to resize images before processing.
   • Set raise_on_failure and max_workers as needed.

Connections

LLMDocumentContentExtractor receives a list of image-based documents as input, typically from a converter such as PDFMinerToDocument. Each document must have a valid file path in its metadata.

It outputs two lists: documents contains successfully processed documents updated with extracted text content, and failed_documents contains documents that could not be processed. Connect documents to DocumentSplitter or DocumentWriter for further processing.

Source Code

To check this component's source code, open llm_document_content_extractor.py in the Haystack repository.

Usage Examples

Basic Configuration

  LLMDocumentContentExtractor:
    type: haystack.components.extractors.image.LLMDocumentContentExtractor
    init_parameters:
      chat_generator:
        type: haystack.components.generators.chat.openai.OpenAIChatGenerator
        init_parameters:
          api_key:
            type: env_var
            env_vars:
            - OPENAI_API_KEY
            strict: true
          model: gpt-4o
      prompt: Extract all text content from this document image. Preserve the original structure including headings, 
        paragraphs, lists, and tables. Return only the extracted text without any additional commentary.
      file_path_meta_field: file_path
      detail: auto
      raise_on_failure: false
      max_workers: 3

LLMDocumentContentExtractor accepts a list of image-based documents as input. Each document must have a valid file path in its metadata. It outputs successfully processed documents (with extracted text content) and a separate list of failed_documents for documents the LLM could not process. It typically receives documents from converters such as PDFMinerToDocument and sends processed documents to DocumentSplitter for further processing.

Using the component in a pipeline

This index uses LLMDocumentContentExtractor to extract text from image-based documents (such as scanned PDFs or images) using a vision-enabled LLM:

# haystack-pipeline
components:
  PDFMinerToDocument:
    type: haystack.components.converters.pdfminer.PDFMinerToDocument
    init_parameters:

  LLMDocumentContentExtractor:
    type: haystack.components.extractors.image.LLMDocumentContentExtractor
    init_parameters:
      chat_generator:
        type: haystack.components.generators.chat.openai.OpenAIChatGenerator
        init_parameters:
          api_key:
            type: env_var
            env_vars:
              - OPENAI_API_KEY
            strict: true
          model: gpt-4o
          generation_kwargs:
      prompt: "Extract all text content from this document image. Preserve the original structure including headings, paragraphs, lists, and tables. Return only the extracted text without any additional commentary."
      file_path_meta_field: file_path
      detail: auto
      size:
      raise_on_failure: false
      max_workers: 3

  DocumentSplitter:
    type: haystack.components.preprocessors.document_splitter.DocumentSplitter
    init_parameters:
      split_by: sentence
      split_length: 5
      split_overlap: 1

  document_embedder:
    type: haystack.components.embedders.sentence_transformers_document_embedder.SentenceTransformersDocumentEmbedder
    init_parameters:
      model: sentence-transformers/all-mpnet-base-v2

  DocumentWriter:
    type: haystack.components.writers.document_writer.DocumentWriter
    init_parameters:
      document_store:
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
        init_parameters:
          hosts:
            - ${OPENSEARCH_HOST}
          http_auth:
            - ${OPENSEARCH_USER}
            - ${OPENSEARCH_PASSWORD}
          use_ssl: true
          verify_certs: false
      policy: WRITE

connections:
  - sender: PDFMinerToDocument.documents
    receiver: LLMDocumentContentExtractor.documents
  - sender: LLMDocumentContentExtractor.documents
    receiver: DocumentSplitter.documents
  - sender: DocumentSplitter.documents
    receiver: document_embedder.documents
  - sender: document_embedder.documents
    receiver: DocumentWriter.documents

inputs:
  files:
    - PDFMinerToDocument.sources

Parameters

Inputs

Parameter	Type	Description
documents	List[Document]	List of image-based documents to process. Each document must have a valid file path in its metadata.

Outputs

Parameter	Type	Description
documents	List[Document]	Successfully processed documents, updated with extracted content.
failed_documents	List[Document]	Documents that failed processing, annotated with failure metadata.

Init Parameters

These are the parameters you can configure in Pipeline Builder:

Parameter	Type	Default	Description
chat_generator	ChatGenerator		A ChatGenerator instance representing the LLM used to extract text. This generator must support vision-based input and return a plain text response.
prompt	str	DEFAULT_PROMPT_TEMPLATE	Instructional text provided to the LLM. It must not contain Jinja variables. The prompt should only contain instructions on how to extract the content of the image-based document.
file_path_meta_field	str	file_path	The metadata field in the Document that contains the file path to the image or PDF.
root_path	Optional[str]	None	The root directory path where document files are located. If provided, file paths in document metadata will be resolved relative to this path. If None, file paths are treated as absolute paths.
detail	Optional[Literal]	None	Optional detail level of the image (only supported by OpenAI). Can be "auto", "high", or "low". This will be passed to chat_generator when processing the images.
size	Optional[Tuple[int, int]]	None	If provided, resizes the image to fit within the specified dimensions (width, height) while maintaining aspect ratio. This reduces file size, memory usage, and processing time.
raise_on_failure	bool	False	If True, exceptions from the LLM are raised. If False, failed documents are logged and returned.
max_workers	int	3	Maximum number of threads used to parallelize LLM calls across documents using a ThreadPoolExecutor.

Run Method Parameters

These are the parameters you can configure for the component's run() method. This means you can pass these parameters at query time through the API, in Playground, or when running a job. For details, see Modify Pipeline Parameters at Query Time.

Parameter	Type	Description
documents	List[Document]	List of image-based documents to process. Each must have a valid file path in its metadata.