System One

POST /v1/systemone

The same decisions as /v1/decisions, in the System One request shape used by TypeSafe Jev and Cloudflare Clef. Use it if your code already speaks that shape; every decision model works on both endpoints.

Endpoint

POST https://api.crossmodel.ai/v1/systemone
Authorization: Bearer cm-YOUR_KEY
Content-Type: application/json

Request

{
  "model": "typesafe/jev",
  "state": "Help! My payouts have been failing for 3 days.",
  "questions": {
    "urgent": { "type": "noul", "instructions": "Does this message convey urgency?" },
    "team": {
      "type": "choice",
      "instructions": "Which team should handle this?",
      "criteria": { "billing": "Payments and payouts", "technical": null, "sales": null }
    },
    "frustration": {
      "type": "score",
      "instructions": "How frustrated is the customer?",
      "criteria": ["No sign of frustration", "Clearly unhappy", "Threatening to leave or complain"]
    }
  }
}

Parameter	Type	Required	Notes
`model`	string	Yes	A decision model ID, e.g. `typesafe/jev`.
`state`	string, object or array	Yes	The content to decide about. An array may mix `{ "type": "text", "text" }` and `{ "type": "image_url", "image_url": { "url" } }` parts.
`images`	string[]	No	Image data URLs, appended after `state`.
`questions`	object	Yes	Questions keyed by name. The key order is the question order.

Any other top-level field is rejected with 400 unknown_parameter.

Question types

`type`	`criteria`	Answer
`noul`	Optional object with `true` and / or `false`: what each answer means.	`noul`: the probability that the answer is yes.
`choice`	Object whose keys are the options and whose values are descriptions (or `null`).	The chosen `choice`, a `confidence`, and a probability per option.
`score`	Array of level descriptions, lowest first.	A `score` (the expected level index, 0 to the number of levels minus 1), a `confidence`, a `legend`, and a probability per level.

Response

{
  "model": "typesafe/jev",
  "answers": {
    "urgent": { "type": "noul", "noul": 0.97 },
    "team": {
      "type": "choice",
      "choice": "billing",
      "confidence": 0.91,
      "probabilities": { "billing": 0.93, "technical": 0.06, "sales": 0.01 }
    },
    "frustration": {
      "type": "score",
      "score": 1.42,
      "confidence": 0.64,
      "legend": {
        "0": "No sign of frustration",
        "1": "Clearly unhappy",
        "2": "Threatening to leave or complain"
      },
      "probabilities": { "0": 0.03, "1": 0.52, "2": 0.45 }
    }
  },
  "usage": { "input_tokens": 148, "output_tokens": 0 }
}

Answers are keyed by the same names as the questions. If a model declines one question, that answer is { "type": "refusal" }, optionally with a reason.

How the two shapes map

`/v1/decisions`	`/v1/systemone`
`input`	`state` (plus `images`)
`questions` array with `name`	`questions` object keyed by name
`predicate`	`noul`
`choices[].value` / `description`	`criteria` keys / values
`levels[].label` / `description`	`criteria` array entries
`answers` array	`answers` object
`usage.total_tokens`	not present

Billing and errors are the same as on /v1/decisions.