- Schema turns document content into named JSON fields.
- Tables reconstructs table structure as table output.
- Split assigns pages to semantic topics so later steps can run only where they belong.
Comparison
Use Schema When
You can name the fields you want before processing the document:- Invoice number, vendor, total, and line items
- Parties, effective date, renewal terms, and termination clauses
- Policyholder, coverage limit, deductible, and claim number
- Bank account holder, balances, and transaction summary
Use Tables When
The row and column structure matters:- 10-K and 10-Q financial statements
- Capitalization tables
- Loss runs and bordereaux
- Pricing schedules
- Tables that continue across pages
Use Split When
The document has distinct sections that should not all receive the same instructions:- Annual reports with business overview, risk factors, financials, and leadership sections
- Loan files with application, underwriting, appraisal, and closing documents
- Insurance packets with policy terms, schedules, claims, and endorsements
- Research or diligence packs with mixed document types
split_id to Schema or Tables for topic-specific output.
Patterns
Extract -> Schema
Best for single-structure documents.Extract -> Tables
Best for table-first output.Extract -> Split -> Schema
Best for long mixed documents.Practical Rule
If you want values, use Schema. If you want tables, use Tables. If you need different logic for different pages, use Split first.Schema Guidelines
Write schemas that produce reliable JSON.
Tables API
Extract and merge structured tables.
Split API
Assign pages to topics.