@codemation/core-nodes-ocr

Azure AI Content Understanding OCR integration for Codemation. Exposes three prebuilt analyzer nodes for document, invoice, and image analysis — designed to make it trivial to wire up OCR-powered workflows.

Install

pnpm add @codemation/core-nodes-ocr

Nodes

• analyzeInvoiceNode — runs the prebuilt-invoice analyzer; returns markdown + structured fields. The prebuilt-invoice ID is verified against the Azure SDK.
• analyzeDocumentNode — runs the prebuilt-document analyzer by default; accepts a custom analyzerId. The default ID follows Azure's published naming conventions but is not verified against a live resource — pass your own analyzerId if Azure returns "analyzer not found."
• analyzeImageNode — runs the prebuilt-imageAnalyzer by default; accepts a custom analyzerId. Same caveat as analyzeDocumentNode.

All three nodes read their input from item.binary (default key: "data") and emit { content: string; fields: Record<string, unknown> } as the item payload. Binary bytes are never put in item.json.

Credential

Register an azureContentUnderstandingCredentialType credential with:

• Endpoint (public) — your Azure resource endpoint, e.g. https://your-resource.cognitiveservices.azure.com/
• API key (secret)

Usage

import { analyzeInvoiceNode, azureContentUnderstandingCredentialType } from "@codemation/core-nodes-ocr";

// Wire up in your workflow:
workflow
  // assume a binary PDF is attached as item.binary["data"]
  .then(analyzeInvoiceNode.create({ binaryField: "data" }, "Extract invoice fields"));

The output item's json will contain:

{
  content: "# Invoice\n\n...",   // markdown text from the analyzer
  fields: {                      // structured fields (dates, amounts, line items, etc.)
    VendorName: "Acme Corp",
    InvoiceTotal: 1234.56,
    // ...
  }
}

For custom analyzer IDs (e.g. a fine-tuned model), pass analyzerId:

analyzeDocumentNode.create({ analyzerId: "my-custom-model", binaryField: "doc" }, "Analyze contract");

Output shape

All three nodes return the same OcrAnalysisOutput type:

type OcrAnalysisOutput = {
  content: string; // markdown representation of the document
  fields: Record<string, unknown>; // structured fields from the prebuilt model
};

Multi-page results are merged into a single content string; the fields property reflects the primary or first content segment unless the analyzer returns multiple segments (in which case a segments array is included).