The output object defines what data you want to extract: Keys are the names of attributes you want to extract Each attribute requires: prompt: Instructions for finding/extracting this data contexts: Array of input or other output attribute names this depends on. Optional Output Configuration Each output attribute can optionally include: format: Data type ('number', 'json', 'url', 'text', 'email', 'tag', 'date', 'boolean') format_details: Format-specific configuration (varies by format type). For json format, you can provide either a description (string) or a schema (JSON Schema object) or both. tools: Array of tools to use (['web_search', 'web_scrape', 'query_pdf', 'query_image']) max_tool_calls: Number of tool calls allowed (0-10) run_when: When to run this extraction ('always', 'any_filled', 'all_filled')

The input object contains your source data: Keys are column/attribute names Values are arrays of strings (all arrays must be the same length) Maximum 1000 rows per request