Classification

POST /v1/evaluate answers typed questions about one piece of state: pick an option, place it on a scale, or give the probability of a yes. It runs on Jev, TypeSafe's classifier model. The request and response follow the AI Gateway /v1/evaluate wire, whose fields are the AI SDK's evaluation-model fields.

The call runs on your project's TypeSafe key and is metered like any model op. A smith's own TypeSafe key is not used. Nothing is stored: the state, the questions, and the answers are not persisted.

Classify text

# Authorization: tenant-admin token (server-side only)
curl https://api.cloud.ingram.tech/v1/evaluate \
  -H "Authorization: Bearer $IC_TOKEN" \
  -H "IC-Api-Version: 2026-05-01" \
  -H "Content-Type: application/json" \
  -d '{
    "state": "My card was charged twice.",
    "questions": {
      "route": {
        "type": "choice",
        "instructions": "Route this ticket.",
        "criteria": { "billing": "payment problems", "shipping": "delivery problems" }
      },
      "urgency": {
        "type": "score",
        "instructions": "How urgent is this ticket?",
        "criteria": ["can wait", "today", "now"]
      },
      "refund": { "type": "boolean", "instructions": "Does the customer ask for a refund?" }
    }
  }'

state is a string, a JSON object, or a JSON array. Each key of questions is an id you choose; its answer comes back under the same id.

`type`	`criteria`	Answer
`choice`	option name → description (`null` for none); at least one option	`choice`, the most probable option, and `probabilities` over every option
`score`	ordered levels, lowest first; at least two	`score`, a fractional level counted from `0`, and `probabilities` keyed by level index
`boolean`	optional `{ "true": …, "false": … }` descriptions	`probability`, the model's P(true)

instructions and each description take a string, a JSON object, or a JSON array.

model is optional and defaults to typesafe-ai.jev-latest. jev-latest moves when TypeSafe ships a new version; pin typesafe-ai.jev-1.13.0 for output that stays stable.

Response

{
  "model": "typesafe-ai.jev-latest",
  "answers": {
    "route": { "type": "choice", "choice": "billing", "probabilities": { "billing": 0.99, "shipping": 0.01 } },
    "urgency": { "type": "score", "score": 1.2, "probabilities": { "0": 0.1, "1": 0.6, "2": 0.3 } },
    "refund": { "type": "boolean", "probability": 0.8 }
  },
  "usage": { "inputTokens": 58, "outputTokens": 0 },
  "rounding": { "probabilityDecimals": 2, "scoreDecimals": 2 },
  "providerMetadata": { "typesafe": { "confidence": { "route": 0.97, "urgency": 0.9 } } }
}

Probabilities and scores are rounded to the places in rounding. providerMetadata.typesafe.confidence is TypeSafe's confidence in each choice and score answer. It is not a probability, and boolean answers carry none. warnings appears when the provider reports one.

Limits & errors

TypeSafe accepts at most 255 options on a choice question, 10 levels on a score question, 64k tokens per request, and 32k tokens for the state plus the longest question.

Status	Code	When
`422`	`invalid_request`	The body breaks the shape above.
`422`	`model_kind_mismatch`	`model` is not a classifier model.
`422`	`model_key_missing`	Your project has no TypeSafe key.
`400`	`model_request_failed`	TypeSafe refused the request: a bad key, or a limit above. The message carries TypeSafe's reason.
`402`	`budget_exceeded`, `card_required`, `insufficient_credits`	As for runs; only a project-wide budget applies. See Usage and Billing.
`429`	`rate_limit_exceeded`	TypeSafe is rate-limiting your key. Retry after the `Retry-After` seconds.
`500`	`evaluation_failed`	Any other failure.

Keys

Add your TypeSafe key in Settings → Models, or over the API:

# Authorization: tenant-admin token (server-side only)
curl -X PUT https://api.cloud.ingram.tech/v1/tenant/model_keys/typesafe-ai \
  -H "Authorization: Bearer $IC_ADMIN_TOKEN" \
  -H "IC-Api-Version: 2026-05-01" \
  -H "Content-Type: application/json" \
  -d '{ "api_key": "…" }'

TypeSafe bills your account for the tokens; your wallet is charged the platform fee alone. See Your own keys.

The call needs a tenant token with the runs:write scope. A smith token is refused with 403 tenant_token_required. See Auth & tokens.

Embeddings Models