Output schema

Where: workflow creation wizard, step 4 — JSON Output Schema The structure of extractedData in an analysis result follows exactly the schema you define here. Omni imposes no common fields of its own.

Two ways to enter it

There are two tabs, and they convert to each other automatically. Editing one is reflected in the other.

Tab	When to use it
Field Builder	Building fields up one at a time. Types and formats are picked from a form
JSON Schema Input	When you already have a schema, or want to paste one wholesale

Field Builder

Selecting Add Field adds a row to the table. No separate modal opens.

Column	Contents
Field Name	Required. Names cannot repeat. Reorder rows with the drag handle on the left
Type	The six types below, distinguished by icon
Format/Constraint	Format for `string`, maximum size for `array`. Empty for the other types
Description	What the AI uses to understand this field. It can be blank, but filling it in raises accuracy
Action	A + on `object` and `array` rows to add child properties, and delete on every row

object and array rows carry an expand arrow and a badge with the child property count. Selecting the arrow expands the child fields one indent level below. With no fields defined, the screen prompts you to select Add Field to start. JSON Schema preview shows how the fields you have built so far assemble into a JSON Schema.

JSON Schema Input

Paste JSON Schema draft-07. When it is valid, the screen confirms the schema is valid.

Output schema JSON input — The JSON input tab — validation results are shown alongside.

Field types

Type	Meaning
`string`	Text data
`number`	A number, decimals included
`integer`	A whole number
`boolean`	`true` / `false`
`object`	A nested object
`array`	An array of values

String formats

A string type can carry a format.

Format	Example
None	No restriction
Date	`2026-01-15`
Date-Time	`2026-01-15T14:30:00Z`
Time	`14:30:00`
Email	`user@example.com`
URL	`https://www.example.com`
IPv4	`192.168.1.1`
Enum	Enter each allowed value and press Enter to add it

Arrays

An array can hold values or objects.

["value1", "value2", "value3"]
[{ "key": "value" }, { "key": "value" }]

Choosing the array type puts a maximum array size in the Format/Constraint cell. The default is Max: 10.

Constraints

Constraint	Detail
Root type	Must be `"type": "object"`, with the fields inside `properties`
Nesting depth	2 levels maximum (an object inside an object)
Field names	Names cannot repeat
Allowed types	`string` · `number` · `integer` · `boolean` · `object` · `array`

Validation messages

JSON Schema Input shows one of these when the schema breaks a rule.

Message	Cause
JSON syntax error	The JSON itself does not parse
The JSON Schema must be an object ({})	The top level is not an object
It must include `"type": "object"`	The root has no type declaration
The root type must be `"object"`	The root type is not an object
Maximum nesting depth (2 levels) exceeded	Nested three or more levels deep
Unsupported type	A value outside the six allowed types
Duplicate key	The same field name was defined twice

When you switch tabs with an error in the JSON, the previous Field Builder state is restored; when a field is invalid, only that field is dropped.

Fill in the descriptions

Each field’s description is what the AI agent uses to decide how to fill that field. Write both the data format and which entry of which document it comes from.

{
  "state_file_number": {
    "type": "string",
    "description": "State file number. Extracted from the state file number entry on the certificate of good standing."
  }
}

Using the same terms across the policy text, the item names, and the schema descriptions makes a large difference to accuracy — creating a workflow.

Results come back wrapped in an envelope

Every leaf field you define comes back in the analysis result not as a bare value but as an object with three keys: value, reasoning, and sourceStep. Intermediate objects and array structures follow your schema exactly.

{
  "company": {
    "legal_name": {
      "value": "ACME TRADING LLC",
      "reasoning": "Read from the entity name field on the certificate of good standing.",
      "sourceStep": 1
    }
  }
}

How to read it is covered in reading analysis results.

You can define a verdict field such as verificationStatus in your output schema, but it is a value of your own and is separate from the system verdict at the top level of the response. Matching its value set to the system’s (verified · pending_review · rejected) keeps your downstream branching simple.

Overview

Project

Workflow Setup

Profiles & Analysis

Two ways to enter it

Field Builder

JSON Schema Input

Field types

String formats

Arrays

Constraints

Validation messages

Fill in the descriptions

Results come back wrapped in an envelope

​Two ways to enter it

​Field Builder

​JSON Schema Input

​Field types

​String formats

​Arrays

​Constraints

​Validation messages

​Fill in the descriptions

​Results come back wrapped in an envelope

Two ways to enter it

Field Builder

JSON Schema Input

Field types

String formats

Arrays

Constraints

Validation messages

Fill in the descriptions

Results come back wrapped in an envelope