Create an Extraction Schema
Define what data to extract from your documents.
Create an Extraction Schema
A schema defines what information to extract. Create a schema once and reuse it across multiple extraction runs.
Steps
- Navigate to Extract in the sidebar
- Click the Schemas tab
- Click New Schema in the top-right corner
- Fill in schema details (see below)
- Add fields using the Visual Builder
- Click Create Schema
Schema details
Name (required)
A short, descriptive name. Examples: "Key Facts", "Author Information", "Meeting Notes".
Description (optional)
Explain what the schema is for. Helpful when sharing or for future reference.
Extraction instructions (optional)
Natural language guidance for the AI. Use this to clarify what you want:
- "Only extract information about events after 1950"
- "Include middle initials for names when available"
- "Skip any data in footnotes"
Visibility
| Option | Who can use it |
|---|---|
| Private (default) | Only you |
| Shared | Your team |
| Public | All users |
Adding fields
Click Add Field to add a new field. Each field has:
Field name
Enter a descriptive name. This becomes the column header in results.
| Good | Avoid |
|---|---|
| "Primary Author" | "field1" |
| "Publication Year" | "data" |
| "Key Finding" | "x" |
Field type
| Type | When to use |
|---|---|
| Text | Names, titles, descriptions, quotes |
| Number | Years, counts, measurements, percentages |
| Boolean | Yes/no questions like "Is peer reviewed?" |
| Enum | One choice from a specific list (click Edit Values to define options) |
| Object | Grouped sub-fields like addresses |
| Array | Multiple values like multiple authors |
Enum values
When you select Enum:
- Click Edit Values
- Type a value and press Enter
- Repeat for all options (minimum 2)
- Click Save
The extraction will only return values matching your defined options.
Object fields (nested)
When you select Object:
- Click Add Field in the nested section
- Add sub-fields (name, type)
- Repeat for all sub-fields
Example: "Address" object with City (Text), State (Text), Country (Text).
Limit: Visual Builder supports 2 levels of nesting. Use the JSON Editor for deeper nesting.
Field options
Click More options on any field to access:
- Required field — Check if the field must always have a value
- Description — Field-specific guidance for the AI
Reordering fields
Drag the handle (⋮⋮) on the left of each field to reorder. Field order determines column order in results.
Saving
Click Create Schema. If there are validation errors, they appear below the form:
- "Field name cannot be empty"
- "Enum must have at least 2 values"
- "Field name cannot start with '_'"
Cloning a schema
To copy any schema (including templates):
- Go to Schemas tab
- Click the actions menu (⋮) on a schema
- Select Clone
- Enter a name for your copy
- Click Create Clone
Your clone starts as Private and can be edited freely.