AI that returns data,
not paragraphs
Define the shape you want, and every AI step returns predictable, typed data that validates against it. Sentiment as an enum, a total as a number, a flag as a boolean, ready to drive the next step without parsing prose.
sentimentenumpriorityenumsummarystringneeds_humanboolean{
"sentiment": "negative",
"priority": "high",
"summary": "Late delivery",
"needs_human": true
}Define the shape, get it back
Define the schema
Describe the fields and types you expect back, the same way you would design any data structure.
The AI fills it in
The step returns data that matches your schema, validated so the fields and types are what you asked for.
The next step reads it
Typed values flow straight into the following step, ready to map, branch on, or act upon.
Predictable AI output
You define the schema
Specify exactly the fields and types you want back, so AI output fits the shape your workflow needs.
Validated every run
The result is checked against your schema on every execution, so downstream steps can rely on it.
Real typed fields
Numbers, dates, and booleans come back as their actual types, not as strings you have to convert.
Enums and booleans
Constrain a field to a fixed set of values, so the answer is always one you can switch on.
Flows into the next step
Structured output drops straight into mapping, conditions, and actions with no glue code.
No flaky parsing
Skip the fragile string handling that breaks the moment the model phrases things differently.
When the next step needs clean data
Classify and route
Return a category the workflow can branch on, instead of a sentence someone has to interpret.
Extract fields
Pull specific values out of text or a document as typed fields ready to use.
Score and decide
Get a number or a verdict back in a known shape, so the next step can act on it directly.
Drive a branch
Use a boolean or enum to choose the path, with no parsing between the AI and the decision.
Structured output vs parsing text
Get typed data you can use directly, instead of pulling answers out of a paragraph.
Structured output when
- The next step needs specific fields
- You want enums and types, not prose
- The result drives a branch or an action
Parsing free text means
- Brittle string handling that breaks on rewording
- No guarantee the fields are even present
- Silent failures that surface downstream
Get data, not paragraphs
Start free and add an AI step that returns clean, typed data to your workflow. No card needed.