Multiple decision primitives - Krun

One call can combine all three decision primitives. A support ticket, for example, needs a department (choice), a hand-off decision (noul) and a severity (score):

curl https://api.krun.ai/v1/decide \
  -H "Authorization: Bearer $KRUN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "context": "Customer says this is the third time exports failed and wants a human immediately.",
    "questions": {
      "department": {"type": "choice", "options": {"billing": "", "support": "", "sales": ""}},
      "needs_human": {"type": "noul", "instructions": "Is the customer asking for human assistance?"},
      "severity": {
        "type": "score",
        "instructions": "How severe is the reported issue?",
        "levels": ["Minor issue", "Feature degraded", "Blocking issue"]
      }
    }
  }'

from krun import Krun, NoulQuestion, ScoreQuestion

client = Krun()

result = client.decide(
    context="Customer says this is the third time exports failed and wants a human immediately.",
    questions={
        "department": {"type": "choice", "options": {"billing": "", "support": "", "sales": ""}},
        "needs_human": NoulQuestion("Is the customer asking for human assistance?"),
        "severity": ScoreQuestion(
            "How severe is the reported issue?",
            ["Minor issue", "Feature degraded", "Blocking issue"],
        ),
    },
)

department = result.choice("department").choice   # "support"
needs_human = result.noul("needs_human").noul      # 0.94
severity = result.score("severity").score          # 1.62

import { Krun } from "@krun-ai/sdk";

const client = new Krun();

const result = await client.decide({
  context: "Customer says this is the third time exports failed and wants a human immediately.",
  questions: {
    department: { type: "choice", options: { billing: "", support: "", sales: "" } },
    needs_human: { type: "noul", instructions: "Is the customer asking for human assistance?" },
    severity: {
      type: "score",
      instructions: "How severe is the reported issue?",
      levels: ["Minor issue", "Feature degraded", "Blocking issue"],
    },
  },
});

result.answers.department.choice; // "support"   (ChoiceAnswer)
result.answers.needs_human.noul;  // 0.94        (NoulAnswer)
result.answers.severity.score;    // 1.62        (ScoreAnswer)

The three answers come back under the same ids, in request order, each with its own type:

{
  "department": {"type": "choice", "choice": "support", "confidence": 0.91,
    "probabilities": {"billing": 0.03, "support": 0.94, "sales": 0.03},
    "abstain": false, "abstention_status": "calibrated"},
  "needs_human": {"type": "noul", "noul": 0.94},
  "severity": {"type": "score", "score": 1.62, "confidence": 0.72,
    "legend": {"0": "Minor issue", "1": "Feature degraded", "2": "Blocking issue"},
    "probabilities": {"0": 0.02, "1": 0.34, "2": 0.64}}
}

Acting on the answers

dept = result.choice("department")
if result.noul("needs_human").noul >= 0.8:
    hand_off_to_agent(queue=dept.choice or "support")
elif result.score("severity").score >= 1.5:
    page_on_call()
else:
    auto_reply(dept.choice)

How it runs

One call, one model job. All questions are answered together; latency is close to a single question.
Independent answers. Each question is scored on its own, so the answer to one question does not depend on which other questions are in the request.
Billing. Three questions are three decisions, whatever their types. usage.input_tokens sums the real token counts of the three questions.
Limits. Up to 16 questions per request, any mix of types.

⌘I