Decision Models (System One)

Decision models evaluate a set of questions against a shared state, returning boolean probabilities, choices, or scale scores. Use them for classification, assessment, and scoring. Responses are non-streaming JSON.

Endpoints and authentication

Operation	Method and path
Decision request	POST /v1/systemone
Available decision models	GET /v1/models?output_modalities=decision
TypeSafe SDK compatible request	POST /typesafe/v1/systemone
TypeSafe format model list	GET /typesafe/v1/models

BASE_URL="https://api.ope.ai"
API_KEY="YOUR_API_KEY"

curl "$BASE_URL/v1/models?output_modalities=decision" \
  -H "Authorization: Bearer $API_KEY"

Examples use jev. Replace it with an actual model ID from the current available list. Call decision models through the systemone endpoint.

Request fields

Field	Type	Required	Description
model	string	Yes	Non-empty decision model ID
state	string / object / array	Yes	Text or JSON content evaluated by all questions
questions	object	Yes	At least one entry; keys are custom question names and values are question definitions
questions.<name>.type	string	Yes	noul, choice, or score
questions.<name>.instructions	string / object / array / null	No	Instructions for the question
questions.<name>.criteria	object / array	Depends on type	See below

Question type	criteria	Answer fields
noul	Optional object, such as {"true":"Condition met","false":"Condition not met"}	noul: probability of true, from 0 to 1
choice	Required object: option name → description	choice: selected option; confidence, probabilities
score	Required non-empty ordered array; indices are scale values	score: weighted value; confidence, legend, probabilities

curl: three question types in one request

curl "$BASE_URL/v1/systemone" \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
    "model": "jev",
    "state": "My payments have been failing for three days. Please resolve this immediately!",
    "questions": {
      "is_urgent": {"type": "noul", "instructions": "Does the customer convey urgency?"},
      "department": {
        "type": "choice",
        "instructions": "Which team should handle this?",
        "criteria": {"billing": "Payments, invoicing, refunds", "technical": "Outages, APIs, system problems"}
      },
      "frustration": {
        "type": "score",
        "instructions": "How frustrated is the customer?",
        "criteria": ["Calm", "Frustrated", "Very angry"]
      }
    }
  }'

Example HTTP 200 response (probabilities and scores are illustrative):

{
  "model": "jev-1.13.0",
  "answers": {
    "is_urgent": {"type": "noul", "noul": 0.95},
    "department": {
      "type": "choice",
      "choice": "billing",
      "confidence": 0.8,
      "probabilities": {"billing": 0.87, "technical": 0.13}
    },
    "frustration": {
      "type": "score",
      "score": 1.04,
      "confidence": 0.94,
      "legend": {"0": "Calm", "1": "Frustrated", "2": "Very angry"},
      "probabilities": {"0": 0, "1": 0.96, "2": 0.04}
    }
  },
  "usage": {"input_tokens": 426, "output_tokens": 73}
}

Keys in answers match questions, and each answer's type matches its question. model is the actual version that answered and may differ from the requested ID. usage.input_tokens and usage.output_tokens report input and output token counts.

score can be fractional. legend and probabilities use string indices for scale levels. Interpret probabilities and scores according to your use case; example values are not fixed results.

TypeSafe SDK connection

Set the TypeSafe SDK baseURL to https://api.ope.ai/typesafe and apiKey to your platform API Key. The SDK endpoint uses the same request and response structure.

export TYPESAFE_BASE_URL="https://api.ope.ai/typesafe"
export TYPESAFE_API_KEY="YOUR_API_KEY"

Errors

Errors use detail. Invalid request structure usually returns HTTP 422, with loc identifying the field:

{
  "detail": [{
    "loc": ["body", "questions", "frustration", "score", "criteria"],
    "msg": "Field required",
    "type": "missing"
  }]
}

Common statuses: 402 insufficient quota, 403 permission or content restriction, 404 unavailable model or incorrect model type, 422 invalid structure or parameters, 429 rate limit, and 503 temporarily unavailable service. Other errors may have a string detail with optional type and code fields.