Define a model by hand
Write the native Decision JSON yourself and execute it. This is the only way models are created.
A model is readable JSON. Writing it yourself — by hand or with your own development tools — is how every Decision is created: aityx has no model-generation endpoint and the console has no AI assistant. Native execution needs only the model and matching state.
This small example calculates an early-payment amount and deadline. Save it as payment-model.json:
{
"name": "Early payment",
"inputs": {
"amount": { "type": "number > 0", "desc": "Invoice amount in USD" },
"invoice_date": { "type": "date", "desc": "Invoice date" },
"payment_on": { "type": "date", "desc": "Planned payment date" }
},
"decisions": {
"discount_by": { "type": "date", "expr": "invoice_date + duration(\"P10D\")" },
"payable_amount": {
"type": "number",
"expr": "if payment_on >= invoice_date and payment_on <= discount_by then decimal(amount * 0.98, 2) else amount"
}
}
}
Submit the bare definition in decision; do not add a nested wrapper or repeat questions:
jq -n --slurpfile model payment-model.json \
'{decision: $model[0], state: {amount: 25000, invoice_date: "2026-10-01", payment_on: "2026-10-11"}}' \
> payment-request.json
curl --fail-with-body -sS https://api.aityx.ai/v1/systemone \
-H "Authorization: Bearer $AITYX_API_KEY" \
-H "Content-Type: application/json" \
-d @payment-request.json
Under these example terms, payment on October 11 qualifies for $24,500, with an October 11 discount deadline. These are worked expectations for the small model, not a separate recorded timing benchmark. The invoice guide has the complete recorded approval example.
Check the definition
Native execution validates the model before evaluating the state. An invalid definition returns 422 invalid_model; mismatching facts return 422 invalid_state.
The console’s free Check reports Ready to run or Cannot run, with blocking details. It is a console operation, not a public validation route; do not call console/internal endpoints from your application.
See Model format for tables, types, functions and expressions. To change a model, edit the JSON and compare the revised rules against known cases before switching.

