Schemas

The only document-specific layer in Docex is the optional built-in schema registry (apps/worker/src/registry.js). Users can bypass it entirely by supplying their own schema at request time.

Built-in schemas

The registry includes convenience defaults for common tasks:

Schema ID	Use case
`invoice/global-accounts-payable`	Extract vendor, total, line items, due date
`trade-license/uae`	Extract company name, license number, expiry
`identity/passport`	Extract full name, document number, date of birth
`receipt/retail`	Extract merchant, items, total, date

Custom schemas

The biggest unlock of the task-agnostic architecture is user-defined schemas. Pass any schema at request time without touching the registry:

const result = await docex.run({
  file: "suspicious-email.png",
  prompt: "Analyze this email attachment for phishing indicators",
  schema: {
    label: "Phishing scan",
    expectedFields: [
      "is_phishing",
      "confidence_score",
      "suspicious_elements",
      "recommended_action"
    ],
    promptScaffold: "Analyze this image for security threats and phishing indicators."
  }
});

This makes Docex genuinely task-agnostic. The built-in schemas are just convenience defaults.

Schema ID vs. custom schema

Use schemaId to reference a built-in registry schema
Use schema to provide an inline custom schema
If neither is provided, the prompt alone drives the analysis (generic vision summary)

Backward compatibility

The old docTypeHint parameter is mapped to schemaId internally. Existing code using docTypeHint will continue to work.