Preserve source references
Pass page numbers through parsing and extraction. Store evidence alongside fields so a reviewer can trace a value back to the document.
Explore model inputs, response structures and integration patterns. Use the Python and JavaScript examples to understand how to handle a result.
Public API access is not yet available. These examples use sample responses. For integration enquiries, contact Istren.
Select a supported destination for an incoming request.
decision.routetextstringThe source text to classify, route, or scan for entities.
routesarrayAllowed destinations. Use stable identifiers for application routing.
Resolve the returned route against your application’s allowed destinations. Your application controls the downstream action.
Inspect in examples{
"model": "decision.route",
"input": {
"text": "The same subscription charge appears twice on my statement. Can you check the duplicate?",
"routes": [
"billing",
"technical_support",
"account_access"
]
}
}Choose Python or JavaScript, download the file, then run it with python3 example.py or node example.mjs. The examples include sample data and do not send API requests.
Keep orchestration in your application. A document intake pipeline has model steps and application checks with different responsibilities.
PDF to page content
MODELExtractContent to typed record
Record to accepted values
Pass page numbers through parsing and extraction. Store evidence alongside fields so a reviewer can trace a value back to the document.
Check required fields, types, allowed values, and business rules. A well-formed response can still contain an incorrect value.
Separate missing content from an empty value. Route unsupported inputs and unresolved results to review before downstream actions.
Use representative examples and a held-out dataset. Agree on the measures that matter for your application before comparing models.
Download evaluation template