Choice - Krun

A choice question picks one option from 2 to 64 alternatives that have no natural order: departments, intents, tools, categories.

Complete example

curl https://api.krun.ai/v1/decide \
  -H "Authorization: Bearer $KRUN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "context": "I was charged twice for the same order and need the money back.",
    "questions": {
      "department": {
        "type": "choice",
        "options": {
          "billing": "Billing and payment issues",
          "returns": "Returns and refunds",
          "shipping": "Shipping and delivery issues"
        }
      }
    }
  }'

from krun import Krun

client = Krun()  # reads KRUN_API_KEY

result = client.decide(
    context="I was charged twice for the same order and need the money back.",
    questions={
        "department": {
            "type": "choice",
            "options": {
                "billing": "Billing and payment issues",
                "returns": "Returns and refunds",
                "shipping": "Shipping and delivery issues",
            },
        },
    },
)

answer = result.choice("department")
print(answer.choice, answer.confidence)  # "returns" 0.8855, or None if the model abstains

Response:

{
  "model": "krun-one-v0.3",
  "answers": {
    "department": {
      "type": "choice",
      "choice": "returns",
      "confidence": 0.885515,
      "probabilities": {"billing": 0.049104, "returns": 0.934619, "shipping": 0.016277},
      "abstain": false,
      "abstention_status": "advisory"
    }
  },
  "usage": {
    "input_tokens": 56
  }
}

The body always has a context and a questions map keyed by ids you choose (department here); the answer comes back under the same id. A request can mix choice questions with noul and score questions, up to 16.

Question fields

A choice question is one entry of questions:

Field	Required	Description
`type`	Yes	`"choice"`
`options`	Yes	Option id → description (2–64). Use `""` or `null` for label-only options.
`task_type`	No	`"intent"` (default) or `"tool"` for tool routing.

Reading the answer

choice is the selected option id, or null when the model abstains.
confidence is the margin between the two most likely options, not the probability of choice. See Confidence and probabilities.
probabilities has one calibrated probability per option, in request order.
abstention_status is calibrated for intent questions with label-only options and advisory otherwise.

Choice behaves exactly as it did before the other primitives existed: existing requests need no change.

⌘I