Skip to main content
A schema tells the AI exactly what fields to produce. Instead of a free-text response, the AI returns structured data with the specific fields you defined.

​
Why use schemas?

Without a schema, an Ask AI step returns free text. You might ask “list the competitors” and get a paragraph, or bullet points, or a numbered list — the format is unpredictable. With a schema, the AI must return data in the exact structure you specified: 
competitors: ["Google", "Microsoft", "Amazon"]
market_cap: 2800000000000
industry: "Technology"
This makes outputs:
  • Consistent — Same fields every time
  • Exportable — Perfect for spreadsheets
  • Chainable — Downstream steps can reference specific fields like analysis.competitors

​
Creating a schema

​
In the Schema Editor

  1. Open the AI Chef step’s settings panel
  2. Click the Schema selector
  3. Choose New Schema (or select an existing one)
  4. The Schema Editor opens — a visual builder where you:
    • Add properties by clicking Add property
    • Set each property’s name, type (string, number, boolean, array, object), and description
    • Mark fields as required or optional
    • Nest objects inside objects for complex structures
Schema editor showing fields for company_name (string), industry (string), and competitors (array) with types and descriptions

​
From the Schemas page

  1. Go to Schemas in the sidebar
  2. Click Create Schema
  3. Build your schema using the same visual editor
  4. Save it — now it’s available in any recipe

​
AI-assisted schema generation

  1. In the schema editor, describe what you want in plain English
  2. Click Generate Schema
  3. Review the AI-generated schema and adjust as needed

​
Using a schema in a step

  1. Add an AI Chef step to your recipe
  2. In the settings panel, select your schema from the Schema dropdown
  3. Write a prompt that asks for the data your schema expects
Match your prompt to your schema. If your schema has a competitors field, make sure your prompt mentions competitors. The AI uses both the prompt and the schema to produce the result.

​
Viewing schema results

When a step with a schema completes, the side panel shows two tabs:
  • Graph — Expandable tree view of the data. Click fields to expand nested objects and arrays.
  • JSON — Raw data view, useful for copying or debugging.
Side panel showing structured output in Graph view with expandable field tree

​
Using schema fields as tokens

Schema outputs produce green token pills in the tokens panel. You can reference specific fields:
  • The full structured output (green pill labeled with step name)
  • Individual fields (green pills labeled step_name > field_name)
Drag these into downstream steps to use specific fields. For example, drag Analysis > competitors into an Export Table step to export just the competitors list.

​
Tips and limits

Add descriptions to fields. Click a property in the schema editor and add a description like “Annual revenue in US dollars.” This helps the AI produce better results.
The top-level type must be an object. You can’t create a schema that’s just an array at the root. Wrap it in an object: instead of a bare list, use items: [list of things].
Keep schemas under 10 levels of nesting. Very deeply nested schemas are harder for the AI to fill correctly and take longer to validate.
Simpler schemas work better. Start with a few required fields. You can always add optional fields later. Over-constrained schemas cause the AI to retry more often.

​
Next steps

  • Outputs — How structured outputs appear in the side panel
  • Step Types — Learn about AI Chef and other steps that use schemas
  • Exporting Results — Export structured data as spreadsheets