Decisions

What it is

POST /v1/decisions is the endpoint for decision models (System One). Unlike chat models they do not generate text: you send application state and one or more typed questions, and get back typed answers with probabilities your code can branch on directly — no prompt-and-parse step.

It replaces the pattern of asking an LLM a narrow question and extracting a label from its answer: faster, cheaper, more predictable.

Models: typesafe/jev-1.13 (TypeSafe Jev, 32K context) and jaredpalmer/kev-4b (Jared Palmer Kev 4B, a compact open Apache-2.0 alternative, 8K context). Both share the same /v1/systemone contract.

Three primitives

Primitive	Question	What comes back
`noul`	Does this condition hold?	Probability of yes (0…1)
`choice`	Which one of these options?	The selected option, a probability per option, confidence
`score`	Where on an ordered scale?	Probability-weighted position, per-level probabilities, confidence

A single request may carry any number of questions of mixed types — one answer per question.

Request

{
  "model": "typesafe/jev-1.13",
  "state": "…context: string, object or array…",
  "questions": {
    "is_bug":  { "type": "noul",   "instructions": "…", "criteria": { "true": "…", "false": "…" } },
    "team":    { "type": "choice", "instructions": "…", "criteria": { "billing": "…", "tech": "…" } },
    "urgency": { "type": "score",  "instructions": "…", "criteria": ["…", "…", "…"] }
  },
  "session_id": "optional — request grouping",
  "user": "optional — end-user identifier"
}

Fields

Field	Type	Required	Description
`model`	string	yes	`typesafe/jev-1.13` (aliases `jev`, `jev-1.13`) or `jaredpalmer/kev-4b` (aliases `kev`, `kev-4b`)
`state`	string \| object \| array	yes	Context to evaluate: ticket text, a state object, an array of related data
`questions`	object	yes	Map “name → question”. The name becomes the key in `answers`
`session_id`	string	no	Grouping of related requests (workflow, conversation) — up to 256 chars
`user`	string	no	End-user identifier — up to 256 chars

Questions

Common fields: type (required) and instructions (required) — the question in words.

noul — criteria with true and false keys (both required): the pair of descriptions defines the boundary of the condition.
choice — criteria as an “option → description” object (required). The answer carries the pick and the full distribution.
score — criteria as an array of ordered levels (required, ≥1). Level 0 is the low end of the scale.

Criteria and instructions accept structured guidance (objects/arrays), not only strings.

Response

{
  "id": "gen-dec-…",
  "model": "typesafe/jev-1.13-20260917",
  "provider": "TypeSafe",
  "answers": {
    "is_bug":  { "type": "noul", "noul": 0.96 },
    "team":    { "type": "choice", "choice": "payments", "confidence": 0.75,
                 "probabilities": { "account": 0, "frontend": 0.16, "payments": 0.84 } },
    "urgency": { "type": "score", "score": 1.99, "confidence": 0.99,
                 "probabilities": { "0": 0, "1": 0.01, "2": 0.99 },
                 "legend": { "0": "Can wait", "1": "This week", "2": "Blocking revenue" } }
  },
  "usage": { "cost": 0.00002, "cost_rub": "0.0023", "billing_source": "openrouter",
             "input_tokens": 476, "output_tokens": 70 }
}

noul → noul: probability of yes.
choice → choice (the pick), probabilities per option, confidence.
score → score (weighted position 0…N−1), probabilities per level, legend mapping indexes to your wording, confidence.
usage.cost — the request cost in USD (as on other endpoints), cost_rub — the charge in rubles.

Pricing and billing

Only input tokens are billed (state plus questions) at the model’s catalog price; output tokens are free. A typical request (~500 tokens) costs a fraction of a cent. Charging follows the actual usage.cost from the response.

Examples

curl https://api.mixen.ai/v1/decisions \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "typesafe/jev-1.13",
    "state": "My checkout page shows a blank screen after I click Pay. Tried two browsers.",
    "questions": {
      "is_bug": {
        "type": "noul",
        "instructions": "Is the customer reporting a software defect?",
        "criteria": {
          "true": "Describes broken or unexpected product behavior.",
          "false": "Asks a question or requests a feature."
        }
      },
      "team": {
        "type": "choice",
        "instructions": "Which team should own this ticket?",
        "criteria": {
          "payments": "Checkout, billing, payment processing.",
          "frontend": "Rendering, layout, browser compatibility."
        }
      },
      "urgency": {
        "type": "score",
        "instructions": "How urgent is this ticket?",
        "criteria": ["Can wait for the next release", "Should be fixed this week", "Blocking revenue right now"]
      }
    }
  }'

import requests

r = requests.post(
    "https://api.mixen.ai/v1/decisions",
    headers={"Authorization": f"Bearer {KEY}"},
    json={
        "model": "typesafe/jev-1.13",
        "state": "Task: clean up inactive accounts. Proposed tool call: delete_rows(...)",
        "questions": {
            "safe_to_run": {
                "type": "noul",
                "instructions": "Is this action safe to run without a human approving it first?",
                "criteria": {
                    "true": "Reversible or low-impact, clearly within the task.",
                    "false": "Destructive, irreversible, or broader than the task."
                },
            }
        },
    },
    timeout=60,
)
noul = r.json()["answers"]["safe_to_run"]["noul"]
if noul < 0.8:
    escalate_to_human()

const r = await fetch('https://api.mixen.ai/v1/decisions', {
  method: 'POST',
  headers: { Authorization: `Bearer ${KEY}`, 'Content-Type': 'application/json' },
  body: JSON.stringify({
    model: 'typesafe/jev-1.13',
    state: { channel: '#support', message: 'Help! Payouts failing for 3 days.' },
    questions: {
      escalate: {
        type: 'noul',
        instructions: 'Does this message convey urgency?',
        criteria: { true: 'Explicitly time-sensitive', false: 'No urgency expressed' },
      },
    },
  }),
});
const { answers } = await r.json();
if (answers.escalate.noul > 0.8) notifyOnCall();

Patterns

Routing — which team/queue owns an incoming item.
Agent gating — is a tool call safe (reversible? within scope?): run, refuse, or ask a human by a noul threshold.
Classification and tagging — one category plus any number of binary tags in a single request.
Cascades — draft with a cheap model, verify against context with a decision model, escalate to an expensive one only on failed checks.

Pick thresholds from confidence/probabilities, not just the top answer: low confidence is the signal to hand the case to a human or re-ask.

Limitations

The model does not generate text and does not explain its decisions — probabilities only. Need a rationale? Ask a chat model afterwards.
Context is per-model: 32K for Jev, 8K for Kev 4B (state plus questions).
Errors are standard platform codes: 400 (invalid parameters, including criteria errors), 402, 404 model_not_found, 429, 502 (upstream failure).