Amazon Textract Tools

Amazon Textract uses machine learning to read text, tables, and forms from scanned documents (PDFs and images) stored in Amazon S3. Use these tools in a workflow when you need to OCR a document, pull structured data out of forms and tables, or feed extracted text into a downstream Agent block.

Overview

Operations

Configuration

textract_detect_document_text

textract_analyze_document

Outputs

textract_detect_document_text

• data (json) — Textract DetectDocumentText result (the raw API response, containing the Blocks array, DocumentMetadata, etc.)

textract_analyze_document

• data (json) — Textract AnalyzeDocument result (the raw API response, containing the Blocks array with table/form relationships, DocumentMetadata, etc.)

YAML Example

textract_1:
  type: textract
  name: "Amazon Textract"
  inputs:
    operation: "textract_detect_document_text"
    s3Bucket: "my-bucket"
    s3Name: "path/to/document.pdf"
    awsRegion: "us-east-1"
    awsAccessKeyId: "{{AWS_ACCESS_KEY_ID}}"
    awsSecretAccessKey: "{{AWS_SECRET_ACCESS_KEY}}"
  connections:
    outgoing:
      - target: next-block-id

Tips

• Auth is AWS SigV4, not an API key. Provide awsRegion, awsAccessKeyId, and awsSecretAccessKey; the request is signed and sent to https://textract.{awsRegion}.amazonaws.com/. The IAM user/role needs textract:DetectDocumentText / textract:AnalyzeDocument plus s3:GetObject on the source bucket.
• The document must already live in S3. Both operations read from s3Bucket + s3Name (the object key) — there is no inline byte upload. Make sure the bucket is in the same region as awsRegion.
• Choose the right operation. Use textract_detect_document_text for plain OCR (lines and words). Use textract_analyze_document with featureTypes set to ["TABLES","FORMS"] (or add QUERIES/SIGNATURES/LAYOUT) when you need structured tables and key-value form data.
• The data output is the raw Textract response. Parse the Blocks array (e.g. {{textract_1.data}}) in a Function or Agent block to assemble lines, tables, or form fields downstream.