Decisions API (OpenAI-compatible)

The Decisions API is the System One decisions API in the shape of OpenAI’s Decisions API, so code written against OpenAI’s SDK works against Fireworks without a translation layer. For the native request shape, see POST /inference/v1/systemone.

Endpoint

POST https://api.fireworks.ai/inference/v1/decisions

Authenticate with your normal Fireworks API key. Set model to the System One model path, for example accounts/<your-account>/models/<your-model>. Models that don’t support this API return a 400.

cURL

curl https://api.fireworks.ai/inference/v1/decisions \
  -H "Authorization: Bearer $FIREWORKS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "accounts/<your-account>/models/<your-model>",
  "input": "Customer wrote: the package arrived late and damaged, I want my money back.",
  "questions": [
    {
      "type": "predicate",
      "name": "refund",
      "instructions": "Does the customer ask for a refund?"
    },
    {
      "type": "choice",
      "name": "tone",
      "instructions": "What is the customer'\''s tone?",
      "choices": [
        { "value": "angry", "description": "Angry or hostile" },
        { "value": "neutral", "description": "Neutral" },
        { "value": "positive", "description": "Positive" }
      ]
    },
    {
      "type": "score",
      "name": "urgency",
      "instructions": "How urgent is this ticket?",
      "levels": [
        { "label": "not urgent" },
        { "label": "somewhat urgent" },
        { "label": "urgent" },
        { "label": "critical" }
      ]
    }
  ]
}'

Request

Field	Type	Description
`model`	string	Required. The System One model path, for example `accounts/<your-account>/models/<your-model>`.
`input`	string or array	Required. The situation to decide about. A string, or an array of `{"role": "user", "content": ...}` messages whose text is joined with newlines. Content can be a string or an array of `{"type": "input_text", "text": "..."}` parts.
`questions`	array	Required. Ordered list of questions. Answers come back in the same order.
`safety_identifier`	string	Optional. Accepted and ignored.

Each question:

Field	Type	Description
`type`	string	Required. `predicate`, `choice`, or `score`.
`name`	string	Optional label for the question, echoed back in its answer.
`instructions`	string	Required. The question itself, in plain language.
`choices`	array	Required for `choice`. Two or more options, each `{"value": ..., "description": "..."}`. Values are strings or booleans. Up to 208 options.
`levels`	array	Required for `score`. Two or more levels in order, lowest first, each `{"label": "...", "description": "..."}`.

predicate, choice, and score map to the noul, choice, and score question types of the System One decisions API.

Response

{
  "model": "accounts/<your-account>/models/<your-model>",
  "answers": [
    { "type": "predicate", "name": "refund", "probability": 0.98, "confidence": 0.96 },
    {
      "type": "choice",
      "name": "tone",
      "choice": "angry",
      "probabilities": [
        { "value": "angry", "probability": 0.91 },
        { "value": "neutral", "probability": 0.08 },
        { "value": "positive", "probability": 0.01 }
      ],
      "confidence": 0.87
    },
    {
      "type": "score",
      "name": "urgency",
      "score": 1.7,
      "probabilities": [
        { "value": 0, "label": "not urgent", "probability": 0.05 },
        { "value": 1, "label": "somewhat urgent", "probability": 0.3 },
        { "value": 2, "label": "urgent", "probability": 0.55 },
        { "value": 3, "label": "critical", "probability": 0.1 }
      ],
      "confidence": 0.4
    }
  ],
  "usage": {
    "input_tokens": 212,
    "input_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0 },
    "output_tokens": 0,
    "output_tokens_details": { "reasoning_tokens": 0 },
    "total_tokens": 212
  }
}

Values above are illustrative. name is null for questions that didn’t set one. For score answers, each probability’s value is the level index and label is the level’s label.

Differences from OpenAI’s Decisions API

Text only. input_image parts return a 400.
208 options per choice question, where OpenAI allows 255.
safety_identifier is accepted but ignored.

Everything else — model, limits, caching, and errors — is identical to /inference/v1/systemone.

API Reference

Inference

Deployments

Training

Training SDK

Evals

Admin

Decisions API (OpenAI-compatible)

Endpoint

Request

Response

Differences from OpenAI’s Decisions API

​Endpoint

​Request

​Response

​Differences from OpenAI’s Decisions API

Endpoint

Request

Response

Differences from OpenAI’s Decisions API