Schemas & Templates

A schema defines what fields the AI should look for in your documents. Templates are saved schemas you can reuse across many documents of the same type.

Choosing a schema during parsing

When you start the Document Parser, Step 2 asks you to choose a schema:

Auto-detect

The AI analyzes your document and extracts every field it can find — names, dates, amounts, tables, entities, and more. Best for:

• First-time exploration of a document type
• Documents where you don't know what fields to expect
• Quick one-off extractions

Use a saved template

Apply a previously saved template with specific field definitions. Best for:

• Processing many documents of the same type (invoices, contracts, forms)
• Consistent, repeatable extraction across your team
• When you only need specific fields, not everything

Skip

Extract raw text only, with no structured field extraction. Useful when you just need the OCR text output.

Creating a template

1. Go to Templates in the left sidebar
2. Click New Template
3. Give it a name and optional description
4. Use the Schema Builder to add fields

The Schema Builder

The Schema Builder is a table where each row defines a field to extract:

Column	Description
Name	The field name (e.g., "Invoice Number", "Total Amount")
Type	Data type: Text, Number, Array, Object, Boolean, Date
Description	Optional hint for the AI about what this field contains

Better field descriptions = better extraction

Adding a description like "The total amount due including taxes, in USD" helps the AI find the right value. Without it, the AI might grab the subtotal instead.

Field types

Type	Use for	Example
Text	Names, addresses, IDs	"John Smith", "INV-2024-001"
Number	Amounts, quantities, percentages	1500.00, 42, 15.5
Date	Dates in any format	"2024-03-15", "March 15, 2024"
Boolean	Yes/no, true/false fields	true, false
Array	Lists of items	Line items, multiple addresses
Object	Nested structured data	Address with street, city, zip

Reordering fields

Drag the handle on each row to reorder fields. The order affects how results are displayed, not extraction accuracy.

Managing templates

Templates are accessible from the Templates section in the sidebar:

• Edit — modify fields, name, or description
• Copy — duplicate a template to create a variation
• Import — load a template from a file
• Delete — remove a template you no longer need

Cloud sync

• Authenticated users: Templates sync to the cloud and are available on any device
• Guest users: Templates are stored locally in your browser only

Template best practices

1. Be specific with field names — "Invoice Date" is better than "Date"
2. Add descriptions — help the AI understand ambiguous fields
3. Use the right data type — Number fields will return numeric values, not strings
4. Start with auto-detect — run it once to see what the AI finds, then create a template from those results
5. Test with sample documents — try your template on 2-3 documents to verify it works well before batch processing