Decisions

Server

Decision models (e.g. jev) that answer typed questions with probabilities instead of text.

Ask a decision model (OpenAI Decisions format)

POST

https://ai.brandkamu.id

/v1/decisions

Ask a decision model (e.g. jev-1.13.0) typed questions about an input and get probabilities back: yes/no (predicate), a pick from a list (choice) or a level (score). Accepts every model except embedding models. input and a questions array are required; streaming is not supported. A question the model won't answer comes back as {"type": "refusal"}. The response gets usage.cost (USD) added.

Ask a decision model (OpenAI Decisions format) › Request Body

DecisionsRequest

model

string · required

required

The text to judge, or messages with input_text parts.

object[] · required

Questions of type predicate, choice (with choices) or score (with levels). Answers come back in the same order.

Ask a decision model (OpenAI Decisions format) › Responses

Answers in the same order as the questions.

DecisionsResponse

model

string

object[]

object

POST/v1/decisions

curl https://ai.brandkamu.id/v1/decisions \
  --request POST \
  --header 'Content-Type: application/json' \
  --header 'Authorization: Bearer <token>' \
  --data '{
  "model": "jev-1.13.0",
  "input": "Customer: I was charged twice for order #4411, please refund one today.",
  "questions": [
    {
      "type": "predicate",
      "name": "is_refund",
      "instructions": "Is the customer asking for a refund?"
    },
    {
      "type": "choice",
      "name": "urgency",
      "instructions": "How urgent?",
      "choices": [
        {
          "value": "low"
        },
        {
          "value": "high"
        }
      ]
    },
    {
      "type": "score",
      "name": "frustration",
      "instructions": "How frustrated?",
      "levels": [
        {
          "label": "calm"
        },
        {
          "label": "annoyed"
        },
        {
          "label": "angry"
        }
      ]
    }
  ]
}'

Example Request Body

{
  "model": "jev-1.13.0",
  "input": "Customer: I was charged twice for order #4411, please refund one today.",
  "questions": [
    {
      "type": "predicate",
      "name": "is_refund",
      "instructions": "Is the customer asking for a refund?"
    },
    {
      "type": "choice",
      "name": "urgency",
      "instructions": "How urgent?",
      "choices": [
        {
          "value": "low"
        },
        {
          "value": "high"
        }
      ]
    },
    {
      "type": "score",
      "name": "frustration",
      "instructions": "How frustrated?",
      "levels": [
        {
          "label": "calm"
        },
        {
          "label": "annoyed"
        },
        {
          "label": "angry"
        }
      ]
    }
  ]
}

json

application/json

Example Responses

{
  "model": "jev-1.13.0",
  "answers": [
    {
      "type": "predicate",
      "name": "is_refund",
      "probability": 0.99
    },
    {
      "type": "choice",
      "name": "urgency",
      "choice": "high",
      "probabilities": [
        {
          "value": "low",
          "probability": 0.07
        },
        {
          "value": "high",
          "probability": 0.93
        }
      ],
      "confidence": 0.86
    }
  ],
  "usage": {
    "input_tokens": 358,
    "output_tokens": 66,
    "cost": 0.00002
  }
}

json

application/json

Ask a decision model (TypeSafe System One format)

POST

https://ai.brandkamu.id

/v1/systemone

The same decision models in TypeSafe's System One format: a state and an object of named questions (noul, choice, score). Accepts every model except embedding models; no streaming. The response's usage.cost is SumoRouter's price in USD.

Ask a decision model (TypeSafe System One format) › Request Body

SystemOneRequest

model

string · required

state

required

The text or state to judge.

object · required

Named questions, each {"type": "noul" | "choice" | "score", "instructions": …}.

Ask a decision model (TypeSafe System One format) › Responses

Answers keyed by question name.

Provider-compatible JSON object passed through from the upstream (with `model` set to the catalog slug and `usage.cost` added in USD).

PassthroughObject

POST/v1/systemone

curl https://ai.brandkamu.id/v1/systemone \
  --request POST \
  --header 'Content-Type: application/json' \
  --header 'Authorization: Bearer <token>' \
  --data '{
  "model": "jev-1.13.0",
  "state": "I was charged twice, refund please",
  "questions": {
    "is_refund": {
      "type": "noul",
      "instructions": "Is this a refund request?"
    }
  }
}'

Example Request Body

{
  "model": "jev-1.13.0",
  "state": "I was charged twice, refund please",
  "questions": {
    "is_refund": {
      "type": "noul",
      "instructions": "Is this a refund request?"
    }
  }
}

json

application/json

Example Responses

{
  "model": "jev-1.13.0",
  "answers": {
    "is_refund": {
      "type": "noul",
      "noul": 0.98
    }
  },
  "usage": {
    "input_tokens": 279,
    "output_tokens": 22,
    "cost": 0.000012
  }
}

json

application/json

Embeddings Models