Skip to main content
The LlamaParse integration parses documents such as PDFs, Word, and PowerPoint files, and returns the content as text, markdown, structured JSON, or images. Use it to pull real content out of uploaded or generated files and feed it into the rest of your workflow.

​
Before you begin

  • In your Next Matter portal, go to Automations library, click Connect in the LlamaParse app tile, and enter your LlamaParse API key to connect.

​
Extract data from a document

  1. Click Workflows > + Create new… > Create a new workflow and give your workflow a name. If you have a workflow ready, click … > Edit workflow.
  2. Click + Add step and select Integrations > LlamaParse.
  3. Click the step and click Configure integration.
  4. In File URL, add the data reference to the document you want to parse, for example the output of a Download file or Export step step. Supported formats include PDF, DOCX, PPTX, and other formats supported by LlamaParse.
  5. Select a Parsing mode:
    • Fast — text only.
    • Cost Effective, Agentic, or Agentic Plus — also support markdown, JSON, and images. Higher modes handle more complex layouts, at a higher cost.
  6. Select Content to return: Text, Markdown, JSON, or Images.
  7. (Optional) Turn on Cost Optimizer to let LlamaParse pick the most cost-effective parsing strategy per page. This can slow down parsing.
  8. (Optional) Turn on Disable cache to skip LlamaParse’s default 48-hour result cache and always reparse the document.
  9. (Optional) Expand Page Ranges to parse only part of the document:
    • Target pages — a comma-separated list, for example 1,3,5-10. Pages are 1-based. By default, all pages are extracted.
    • Max pages — the maximum number of pages to parse, starting from page 1.
  10. (Optional) Expand Agentic Options to add a Custom prompt with extra extraction rules or formatting instructions. Available with Agentic and Agentic Plus parsing modes.
  11. (Optional) If Content to return is set to Images, expand Output Options and toggle which image types to save: embedded images (logos, figures, photos), page screenshots, and layout images (detected regions such as tables and figures).
  12. Save your changes.
Parsing runs asynchronously. The step pauses until LlamaParse finishes processing the document, then resumes automatically with the extracted content.

​
Use the extracted data in later steps

Once the step completes, select the output that matches the Content to return you chose (Text, Markdown, JSON, or Images) as a variable. Use it in downstream steps, for example to populate a form field, send an email, or run further automations on the extracted content.
Parsing a document larger than a few pages, or in Agentic / Agentic Plus mode, can take a while. Design downstream steps to run after the LlamaParse step completes, not in parallel with it.