Decision model · TypeSafe AI

Jev API

Jev is TypeSafe AI's decision model. It does not write text: you send a state (the text or JSON to judge) plus named questions, and every answer comes back typed, with probabilities and a confidence value.

Model ID: jev-latest
Input: Text or JSON state
Output: Typed answers
Latency: Under 1 s

Capabilities

Capabilities & endpoints

One synchronous endpoint. Jev does not use the chat format, so the OpenAI and Anthropic SDKs do not apply; send plain HTTPS requests. Calling jev-latest on /v1/chat/completions returns an error.

POSThttps://tokenra.io/v1/decisionsTested

Send a state and named questions; every answer comes back typed. Auth: Authorization: Bearer.

Questions

Question types

Each question has a name you choose, a type, instructions and, depending on the type, criteria. All three types were answered on 2026-10-23.

type	criteria	Answer fields
`noul`	Optional. Object with exactly the keys `"true"` and `"false"`, each describing when it applies.	`noul`: probability (0–1) that the answer is true.
`choice`	Required. Object that maps each option key to its description, at least one option.	`choice` (the most likely key), `probabilities` per option, `confidence`.
`score`	Required. Array of at least two level descriptions, from the lowest to the highest.	`score` (0 to number of levels − 1, can be fractional), `legend`, `probabilities` per level, `confidence`.

Request

Request parameters

Parameter	Type	Required	Description
`model`	string	Yes	`jev-latest`. The response reports the version that answered, for example `typesafe/jev-1.14-20261014` or `jev-1.14.0`.
`state`	string / object / array	Yes	What to judge, shared by every question: text, or JSON such as a ticket or a record. Chinese text was tested.
`questions`	object	Yes	Your question names mapped to question objects; at least one. Each answer comes back under the same name.
`questions.*.type`	string	Yes	`noul`, `choice` or `score`.
`questions.*.instructions`	string / object / array	Yes	The question itself.
`questions.*.criteria`	object / array	No	Depends on the type, see the table above. Required for `choice` and `score`.
`stream`	boolean	No	Not supported. Leave it out; `true` returns an error.

Response

Response format

{
  "model": "typesafe/jev-1.14-20261014",
  "answers": {
    "department": {
      "type": "choice",
      "choice": "billing",
      "probabilities": {
        "billing": 0.88,
        "technical": 0.12,
        "sales": 0
      },
      "confidence": 0.82
    },
    "frustration": {
      "type": "score",
      "score": 1.02,
      "legend": {
        "0": "Calm",
        "1": "Frustrated",
        "2": "Very angry"
      },
      "probabilities": {
        "0": 0,
        "1": 0.98,
        "2": 0.02
      },
      "confidence": 0.97
    },
    "is_urgent": {"type": "noul", "noul": 0.95}
  },
  "usage": {"input_tokens": 427, "output_tokens": 73},
  "id": "gen-dec-1792772733-dG0LU2uG8E83JEMaXCmh",
  "provider": "TypeSafe"
}

Field	Type	Description
`model`	string	The Jev version that answered.
`answers.<name>.type`	string	The question type.
`answers.<name>.noul`	number	noul only: probability that the answer is true.
`answers.<name>.choice`	string	choice only: the most likely option key.
`answers.<name>.score`	number	score only: expected level, from 0 to the number of levels − 1.
`answers.<name>.probabilities`	object	choice and score: probability of each option or level; they add up to 1.
`answers.<name>.legend`	object	score only: level index to your level description.
`answers.<name>.confidence`	number	choice and score: confidence from 0 to 1. Use it to route unsure cases to a person.
`usage`	object	`input_tokens` and `output_tokens` used for billing.
`id / provider`	string	Request ID and serving provider, when the upstream includes them.

The answers are keyed by your question names; do not rely on their order. In testing they came back sorted by name, not in request order.

During testing the upstream answered in two shapes: with model like typesafe/jev-1.14-20261014 plus id and provider, or with model like jev-1.14.0 and no extra fields. answers and usage were the same in both, so read only the fields you need.

Errors

Errors & limits

Request validation

Invalid requests return HTTP 400 with error.code invalid_request and one of these messages. Do not retry them; fix the request.

Message	Cause
`questions must not be empty`	`questions` is missing or empty.
`state must be a string, object or array`	`state` is missing, for example when a chat `messages` body is sent instead.
`unsupported question type; expected noul, choice or score`	Unknown `type`.
`choice criteria must be a nonempty object of choices`	A choice question without options.
`score criteria must contain at least two string levels`	A score question with fewer than two levels.
`noul criteria must contain only true and false string descriptions`	Other keys in noul criteria.
`decisions do not support streaming`	`stream: true` was sent.

Common errors

HTTP	error.code	Meaning	What to do
`401`	—	Missing, wrong or deleted API key. The message is `Invalid token`.	Check the `Authorization: Bearer` header and the key in the console.
`400`	—	The body is not valid JSON.	Fix the JSON and send `Content-Type: application/json`.
`403`	`insufficient_user_quota`	Your balance or plan quota is used up.	Top up in the console, then retry.
`503`	`model_not_found`	Unknown model ID, or no channel serves it for your group.	Check the ID in the model list or with `GET /v1/models`.
`429`	—	Too many requests.	Back off exponentially and retry.
`5xx`	—	The upstream provider failed or timed out.	Retry a few times; keep the request id if it persists.

Limits and notes

Requests took 0.35 to 1 second in testing.
Billing: input and output tokens are both charged at $0.042 per 1M tokens (checked 2026-10-24). A test request with 358 input and 60 output tokens was charged 9 quota units, that is $0.000018. Question definitions count as input tokens. Rates can change; the live rate is on the Model Square.
A one-sentence state with the three example questions used 427 input and 73 output tokens; question definitions count as input.