Extract Fields

The Extract Fields node uses AI vision to pull specific named values from images and documents. Unlike OCR (which extracts all text) or Table Extract (which finds tabular data), Extract Fields targets the exact fields you define: invoice number, due date, total amount, customer name.

When to use Extract Fields vs. other Vision nodes

Node	Use when
OCR	You need all text from an image
Read PDF	You need all text from a PDF
Table Extract	The data is in a table/grid format
Extract Fields	You need specific named values from a document
Ask Document	You want to ask a natural language question about a document

Configuration

Field	Type	Required	Description
Source	Select	Yes	Where the image/document comes from: URL, File upload output, or previous node
Source URL/path	FX formula	Yes	The image URL or file path. Supports variables.
Fields to extract	Field list	Yes	Define each field you want to pull (see below)
Document type	Select	No	Hint: Invoice, Receipt, ID card, Contract, Custom. Helps AI accuracy.

Defining fields

Each field has:

Property	Description
Field name	The key in the output (e.g., `invoice_number`)
Field label	Human-readable description (e.g., "Invoice Number" or "The invoice/PO number at the top of the document")
Field type	Expected type: Text, Number, Date, Currency, Boolean
Required	Whether extraction should fail if this field isn't found

Example field configuration for invoices

Field name	Label	Type	Required
`invoice_number`	Invoice or PO number	Text	Yes
`date`	Invoice date	Date	Yes
`due_date`	Payment due date	Date	No
`vendor_name`	Vendor / company name	Text	Yes
`total_amount`	Total amount due	Currency	Yes
`tax_amount`	Tax / VAT amount	Currency	No
`line_items_count`	Number of line items	Number	No
`payment_terms`	Payment terms (Net 30, etc.)	Text	No

Output variables

{{extract_fields.result.invoice_number}}   → "INV-2024-0847"
{{extract_fields.result.date}}             → "2024-03-15"
{{extract_fields.result.total_amount}}     → 4250.00
{{extract_fields.result.vendor_name}}      → "Acme Supplies Inc."
{{extract_fields.confidence}}              → 0.94

Each field is available at {{extract_fields.result.<field_name>}}. The confidence score (0-1) indicates how confident the AI is about the overall extraction.

Common patterns

Invoice processing

Email Trigger (new email with attachment) → Extract Fields (invoice fields) →
  If-Else (total > 5000?) →
    Yes: HITL (manager approval) → Create Record (accounting DB)
    No: Create Record (accounting DB) → Send Email (auto-approved notification)

Receipt expense tracking

Form Submission (receipt upload) → Extract Fields (vendor, amount, date, category) →
  Google Sheets (append to expense report) → If-Else (amount > policy limit?) →
    Yes: Slack (notify manager for approval)

ID verification

Form Submission (ID upload) → Extract Fields (name, DOB, ID number, expiry) →
  TinyGPT (verify name matches form submission) → If-Else (match?) →
    Yes: Update Record (verified = true)
    No: Send Email (verification failed, please re-upload)

Tips for better extraction

Tip	Why
Set the document type	Gives the AI context about where to look for fields
Write descriptive labels	"Total amount including tax at the bottom" is better than "Total"
Use specific field types	Currency type handles $ symbols and decimal parsing automatically
Mark truly required fields	Don't mark everything required; some documents may lack optional fields
Provide multiple name variants	"Invoice Number / PO Number / Reference Number" in the label handles different document formats

Warning

Extraction accuracy depends on document quality. Blurry photos, handwritten text, or unusual layouts may produce lower confidence scores. Always validate extracted data for high-stakes workflows (financial, legal) with a Human-in-the-Loop node.

Note

Extract Fields costs 5 credits per execution. The cost is the same regardless of the number of fields defined.