Decisions

Beta FeatureThe Decisions API is currently in beta.

Overview

POST /v1/decisions serves TypeSafe’s Jev model, a System One model. Unlike the Responses or Chat Completions APIs, it doesn’t generate text. You send it your application’s state and a map of typed questions, and it returns a calibrated probability for each one. Use this when your code needs a number it can branch on, such as “how likely is this support ticket to be a bug report” or “which of these three categories does this transaction belong to,” rather than a paragraph it has to parse.

The request and response shapes are identical to TypeSafe’s System One API and OpenRouter’s Decisions API, so a client written against either works here by pointing it at /v1/decisions. Fields this endpoint doesn’t implement, including OpenRouter’s routing, are ignored rather than rejected: Jev is the only model behind the endpoint, so there’s nothing to route between.

Request

string

required

The model to use. See Model and aliases.

string | object | array

required

The application state the questions are judged against. Accepts a plain string, a JSON object, or a JSON array.

object

required

A map of caller-chosen question ids to question objects. Must contain at least one question. Each question has a type of noul, choice, or score, plus instructions and (depending on type) criteria. Both instructions and criteria accept structured JSON, not only strings: a string, an object, or an array, with any JSON values inside that object or array.

Question types

`noul`

A yes/no judgment. Returns a probability between 0 and 1 that the answer is yes. criteria is optional, and may carry true and false descriptions to sharpen the judgment.

curl https://api.concentrate.ai/v1/decisions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "jev-1.13.0",
    "state": "The export button does nothing when I click it, and no file downloads.",
    "questions": {
      "is_bug_report": {
        "type": "noul",
        "instructions": "Is this message reporting a bug?",
        "criteria": {
          "true": "The customer describes the product behaving incorrectly",
          "false": "The customer asks a question or makes a request without reporting a defect"
        }
      }
    }
  }'

import requests

response = requests.post(
    "https://api.concentrate.ai/v1/decisions",
    headers={
        "Authorization": "Bearer YOUR_API_KEY",
        "Content-Type": "application/json"
    },
    json={
        "model": "jev-1.13.0",
        "state": "The export button does nothing when I click it, and no file downloads.",
        "questions": {
            "is_bug_report": {
                "type": "noul",
                "instructions": "Is this message reporting a bug?",
                "criteria": {
                    "true": "The customer describes the product behaving incorrectly",
                    "false": "The customer asks a question or makes a request without reporting a defect"
                }
            }
        }
    }
)

print(response.json()["answers"]["is_bug_report"])

const response = await fetch("https://api.concentrate.ai/v1/decisions", {
  method: "POST",
  headers: {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json"
  },
  body: JSON.stringify({
    model: "jev-1.13.0",
    state: "The export button does nothing when I click it, and no file downloads.",
    questions: {
      is_bug_report: {
        type: "noul",
        instructions: "Is this message reporting a bug?",
        criteria: {
          true: "The customer describes the product behaving incorrectly",
          false: "The customer asks a question or makes a request without reporting a defect"
        }
      }
    }
  })
});

const data = await response.json();
console.log(data.answers.is_bug_report);

`choice`

Selects one option from a required map of option name to description. Returns the chosen option, a probability distribution over all options, and a confidence value. A description can be null when the option name is self-explanatory and needs no further explanation, for example { "shipping": null, "technical": null, "other": null }.

curl https://api.concentrate.ai/v1/decisions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "jev-1.13.0",
    "state": {
      "subject": "Re: order #4471",
      "body": "My package was marked as delivered but it never arrived."
    },
    "questions": {
      "category": {
        "type": "choice",
        "instructions": "Categorize this support ticket.",
        "criteria": {
          "shipping": "Questions about deliveries, tracking, or returns",
          "technical": "Questions about product bugs or errors",
          "account": "Questions about login, access, or account settings"
        }
      }
    }
  }'

import requests

response = requests.post(
    "https://api.concentrate.ai/v1/decisions",
    headers={
        "Authorization": "Bearer YOUR_API_KEY",
        "Content-Type": "application/json"
    },
    json={
        "model": "jev-1.13.0",
        "state": {
            "subject": "Re: order #4471",
            "body": "My package was marked as delivered but it never arrived."
        },
        "questions": {
            "category": {
                "type": "choice",
                "instructions": "Categorize this support ticket.",
                "criteria": {
                    "shipping": "Questions about deliveries, tracking, or returns",
                    "technical": "Questions about product bugs or errors",
                    "account": "Questions about login, access, or account settings"
                }
            }
        }
    }
)

print(response.json()["answers"]["category"])

const response = await fetch("https://api.concentrate.ai/v1/decisions", {
  method: "POST",
  headers: {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json"
  },
  body: JSON.stringify({
    model: "jev-1.13.0",
    state: {
      subject: "Re: order #4471",
      body: "My package was marked as delivered but it never arrived."
    },
    questions: {
      category: {
        type: "choice",
        instructions: "Categorize this support ticket.",
        criteria: {
          shipping: "Questions about deliveries, tracking, or returns",
          technical: "Questions about product bugs or errors",
          account: "Questions about login, access, or account settings"
        }
      }
    }
  })
});

const data = await response.json();
console.log(data.answers.category);

`score`

A position on ordered levels, described by a required array of one to ten level descriptions, ordered from lowest to highest. Two or more is what makes the position meaningful. Returns a probability-weighted value, the levels as a legend, a distribution, and a confidence value.

curl https://api.concentrate.ai/v1/decisions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "jev-1.13.0",
    "state": "I tried to cancel three times and support never replied. Extremely frustrating.",
    "questions": {
      "urgency": {
        "type": "score",
        "instructions": "How urgent is this message?",
        "criteria": [
          "Low: no action needed soon",
          "Medium: should be addressed within a few days",
          "High: needs a response today"
        ]
      }
    }
  }'

import requests

response = requests.post(
    "https://api.concentrate.ai/v1/decisions",
    headers={
        "Authorization": "Bearer YOUR_API_KEY",
        "Content-Type": "application/json"
    },
    json={
        "model": "jev-1.13.0",
        "state": "I tried to cancel three times and support never replied. Extremely frustrating.",
        "questions": {
            "urgency": {
                "type": "score",
                "instructions": "How urgent is this message?",
                "criteria": [
                    "Low: no action needed soon",
                    "Medium: should be addressed within a few days",
                    "High: needs a response today"
                ]
            }
        }
    }
)

print(response.json()["answers"]["urgency"])

const response = await fetch("https://api.concentrate.ai/v1/decisions", {
  method: "POST",
  headers: {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json"
  },
  body: JSON.stringify({
    model: "jev-1.13.0",
    state: "I tried to cancel three times and support never replied. Extremely frustrating.",
    questions: {
      urgency: {
        type: "score",
        instructions: "How urgent is this message?",
        criteria: [
          "Low: no action needed soon",
          "Medium: should be addressed within a few days",
          "High: needs a response today"
        ]
      }
    }
  })
});

const data = await response.json();
console.log(data.answers.urgency);

Response

{
  "id": "req_01a0cb13360d7a7d8af625155408796f",
  "model": "typesafe/jev-1.13.0",
  "answers": {
    "is_bug_report": {
      "type": "noul",
      "noul": 0.94
    }
  },
  "usage": {
    "input_tokens": 283,
    "output_tokens": 23
  }
}

id identifies the request; it’s the Response ID the logs page shows, so quote it when you contact support. answers is keyed by the same question ids you sent in the request. model is returned in canonical provider/model form, for example typesafe/jev-1.13.0. usage.output_tokens is the count TypeSafe reports; output tokens aren’t charged. Each answer’s shape depends on its type, which matches the question that produced it:

noul — { "type": "noul", "noul": 0.87 }. The single noul field is the probability of yes. There’s no confidence field, because the probability is the answer.
choice — { "type": "choice", "choice": "shipping", "confidence": 0.91, "probabilities": { "shipping": 0.91, "technical": 0.06, "account": 0.03 } }. confidence and probabilities are both optional.
score — { "type": "score", "score": 2.4, "confidence": 0.78, "probabilities": { "Low": 0.05, "Medium": 0.2, "High": 0.75 }, "legend": { "Low": "Low: no action needed soon", "Medium": "Medium: should be addressed within a few days", "High": "High: needs a response today" } }. confidence, probabilities and legend are all optional. legend maps each level back to the description you supplied, so you can render the answer without holding onto the original request.

On choice and score answers, confidence summarizes how concentrated the probability distribution is. It isn’t a statement that the chosen answer is correct: several genuinely acceptable options will spread probability across them and lower confidence even when the model is judging soundly.

Fields that TypeSafe returns beyond what’s documented here are forwarded to you rather than stripped. If you’re writing a strict client-side schema, allow for unknown fields on the response and on individual answers.

Redaction, ZDR and HIPAA

Keys with redaction enabled can’t use this endpoint: the request is refused with a 400 before anything is sent upstream. Redaction runs on chat bodies, and a Decisions request has no chat body, so accepting it would send your state to TypeSafe unredacted. Use a key without redaction enabled for Decisions traffic. TypeSafe offers neither zero data retention nor a Business Associate Agreement, so keys with ZDR or HIPAA mode enabled are refused with a 422, exactly as the chat endpoints refuse a provider that can’t meet the policy.

Model and aliases

The current model is jev-1.13.0. The following aliases resolve to it:

jev-latest
jev-preview
jev
jev-1.13 (the slug OpenRouter publishes as typesafe/jev-1.13)

A provider/ prefix is accepted: typesafe/jev-1.13.0 and typesafe/jev-1.13 both work. A prefix that isn’t a provider of the model, such as openai/jev-1.13.0, is a 400.

Aliases move when TypeSafe ships a new release, so the answers behind jev-latest can change without a change on your side. The response’s model field reports the versioned ID that answered. If you’ve tuned confidence thresholds against a version, pin its ID (jev-1.13.0) and move on your own schedule.

Limits

Limit	Value
Tokens per request	64k
`state` plus the longest question	32k
Choice options	up to 255
Score levels	1 to 10
Input token price	$0.042 per million
Output tokens	not charged

Decisions requests share your key’s token rate-limit windows with every other endpoint. During the beta there’s no per-request limit beyond that, but all Concentrate traffic to TypeSafe shares one account limit of 1,200 requests per minute; if you plan a sustained high-volume workload, contact us first.

Errors

Status	Meaning
`400`	The request is malformed, the model is unknown or isn’t a Decisions model, the `provider/` prefix doesn’t match, the key has redaction enabled, or the request exceeds a token budget (`max_tokens_exceeded`)
`401`	Missing or invalid API key
`402`	Insufficient credit, or the key’s spend cap is reached
`403`	The key’s settings don’t allow this model or provider
`404`	The path doesn’t exist
`422`	The key has ZDR or HIPAA mode enabled, or TypeSafe rejected the request’s shape (the message is theirs)
`424`	The upstream TypeSafe request failed
`429`	Your token rate limit, or TypeSafe’s, was hit; a `retry-after` header is forwarded when TypeSafe sends one
`499`	You closed the connection before we answered; recorded for your usage, never delivered
`504`	TypeSafe didn’t respond in time

See Error Handling for the general error response format.

Writing questions

TypeSafe’s guide to state, instructions and criteria

Jev 1.13 jaggedness

Known failure modes and what to do instead

Create Response

Generate text with the Responses API

Error Handling

Understand error responses across the API

Responses

Chat Completions (Beta)

Messages (Beta)

Decisions (Beta)

Models API

Features

Utilities

Reference

Catalog

Overview

Request

Question types

`noul`

`choice`

`score`

Response

Redaction, ZDR and HIPAA

Model and aliases

Limits

Errors

Writing questions

Jev 1.13 jaggedness

Create Response

Error Handling

​Overview

​Request

​Question types

​noul

​choice

​score

​Response

​Redaction, ZDR and HIPAA

​Model and aliases

​Limits

​Errors

​Related Documentation

Writing questions

Jev 1.13 jaggedness

Create Response

Error Handling

Overview

Request

Question types

`noul`

`choice`

`score`

Response

Redaction, ZDR and HIPAA

Model and aliases

Limits

Errors

Related Documentation