PDF Vector icon

PDF Vector

Convert PDFs, Word, Excel documents, and images to clean markdown, extract structured data with AI, process invoices with specialized parsing, and search millions of academic papers across PubMed, ArXiv, Google Scholar, and more.

Actions18

Document → Ask

AI-generated

Overview

The node allows users to ask questions about documents such as PDFs, Word, Excel, CSV, and images by leveraging AI models. It supports providing the document either via a public URL or as binary data from a previous node. The node processes the document using different AI model tiers to extract relevant information and answer the user's question. This is useful for extracting insights, summaries, or specific data points from various document types automatically.

Use Case Examples

  1. A user provides a URL to a PDF report and asks for the main findings.
  2. A user uploads an image containing text and asks a question about its content.
  3. A user inputs a Word document as binary data and queries specific structured data within it.

Properties

Name Meaning
Input Type Specifies how the document is provided, either by a public URL or binary data from a previous node.
Document URL The public URL of the document file, required if Input Type is URL.
Input Binary Field The name of the binary property from a previous node containing the file, required if Input Type is Binary Data.
Model The AI model tier to use for processing the document, affecting capabilities and cost.
Question The question to ask about the document, minimum 4 characters.
Document ID Optional identifier for usage tracking, returned in the response.

Output

JSON

  • answer - The AI-generated answer to the question about the document.
  • documentId - The optional identifier provided for usage tracking, returned in the response.
  • metadata - Additional metadata about the processed document or response.

Dependencies

  • An API key credential for the PDF Vector service

Troubleshooting

  • Ensure the document URL is publicly accessible if using URL input type.
  • Verify the binary data field name matches the output of the previous node if using binary input.
  • Check that the question meets the minimum character requirement (4 characters).
  • Be aware of model limitations regarding file size and page count to avoid processing errors.
  • Handle API errors gracefully, especially network or authentication issues.

Links

Discussion