LlamaParse Connector
This connector uses LlamaParse to extract text and structured data from various document formats including PDFs, images, and other document types.
Features
- Document Parsing: Extracts text content from PDFs, images, and other document formats
- Structured Output: Returns parsed content in markdown format
- Multiple Input Types: Supports both tag-based document streams and document objects
Input
- Data: Accepts a wide range of file or content types, including binary files, mixed media, or serialized formats
Outputs
- Text: Outputs extracted text as markdown‑formatted content, preserving headings, lists, and paragraphs where available
- Table: Outputs structured tables extracted from the document (markdown table format)
Configuration
The connector supports the following configuration options:
 Required Configuration
– `api_key` (string, optional): Your LlamaParse API key. If not provided, some features may be limited.
Simple Configuration
- parse_mode (string, default: “cost_effective”): The parsing mode to use for document processing:
- Â cost_effective (3 cred./page): Best for text-heavy documents without diagrams and images
- agentic (10 cred./page): Best for documents with diagrams and images, may struggle with complex layouts
- agentic_plus (90 cred./page): Highest parsing setting for complex layouts, multi-page tables, diagrams, and images
- parse_page_with_llm (Legacy): Legacy LLM-based parsing mode
- parse_page_with_lvm** (Legacy): Legacy LVM-based parsing mode with additional configuration options
- lvm_model (string, optional): The LVM model to use when parse_mode is set to “parse_page_with_lvm”, “agentic”, or “agentic_plus”. Options include “anthropic-sonnet-4.0”, “anthropic-sonnet-3.5”, “gpt-4o”, “gpt-4o-mini”.
- `use_system_prompt_append` (boolean, default: false): Whether to add custom instructions to the system prompt.
- `system_prompt_append` (string, optional): Additional instructions to append to the system prompt when use_system_prompt_append is enabled.
- `spreadsheet_extract_sub_tables` (boolean, default: false): Extract sub-tables from spreadsheets for better table parsing.
Example Configuration
json
{
“api_key”: “your-llamaparse-api-key-here”,
“parse_mode”: “agentic”,
“use_system_prompt_append”: false,
“spreadsheet_extract_sub_tables”: false
}
 Parse Mode Selection Guide
- Use cost_effective for:
       • Text-heavy documents (reports, articles, books)
       • Documents without diagrams, charts, or images
       • Budget-conscious processing
- Use agentic for:
       • Documents with diagrams, charts, or images
       • Mixed content documents
       • When you need better visual understanding
- Use agentic_plus for:
       • Complex layouts and multi-page tables
       • Documents with intricate diagrams
       • When maximum accuracy is required
       • Complex technical documents
 Error Handling
- If parsing fails, the connector returns an empty string and logs the error
- Temporary files are automatically cleaned up after processing
- Thread-safe operations prevent concurrent access issues
- Proper object failure handling with completion codes
