Decisions
POST /v1/decisions classifies, routes or scores text using predefined questions.
Use Microsoft-Decision-1. Check current availability and prices in the
live catalogue.
curl https://gateway.primexos.com/v1/decisions \
-H "Authorization: Bearer $PRIMEX_API_KEY" \
-H 'Content-Type: application/json' \
-d '{
"model": "Microsoft-Decision-1",
"state": "The integration fails with an error.",
"questions": {
"team": {
"type": "choice",
"instructions": "Which team should handle this ticket?",
"criteria": {"support": "Account questions", "engineering": "Software errors"}
}
}
}'
Use choice for named options, noul for a yes/no probability, and score for
an ordered scale. Send text state and 1–32 named questions. Choice and score
questions accept 2–32 criteria. Streaming, JSON state, custom generation parameters
and written explanations are unavailable. Evaluate the model on representative
examples before choosing automated decision thresholds.
The response contains model, named answers, and reported usage.input_tokens
and usage.output_tokens. Only input tokens are charged; output tokens are free.
Missing, invalid or incomplete native usage is not estimated. Rates and discounts
come from the current catalogue. This is Primex's native decision contract; it
does not implement the OpenAI Decisions request or answer-array format.
An optional X-Primex-Operation-ID identifies an original operation. Reusing it
returns a conflict and does not repeat execution. After an uncertain
transport outcome, check Usage before submitting new work.
POST /v1/decisions · Primex-native
| Parameter | Type | Required | Description |
|---|---|---|---|
model | string | Yes | Offering ID to route to (discover IDs via GET /v1/models for this service_type). |
state | string | Yes | Text evaluated by every question; initially text only, not JSON state. |
questions | object | Yes | One to 32 named noul, choice or score questions. Choice and score questions require 2–32 criteria. |
Microsoft Decision 1 bills native input tokens at the selected offering rate. Output is uncharged. No streaming or estimated tokens.