OVERVIEW

WHAT IS A DECISION MODEL

A decision model answers narrow, multiple-choice questions instead of generating open-ended text. You give it a state that describes the current situation and a set of questions about that state. It returns a probability for each answer option, scoring all questions in parallel. A generative LLM produces its answer one token at a time. That makes it flexible, but slow and expensive when all you need is a choice between a few options. A decision model skips text generation and only scores the options, so each decision is faster and cheaper. A decision model can’t write code, compose messages, or explain its reasoning. Use it as a fast decision layer next to your LLM, not as a replacement. The LLM sets up the state and acts on the result; the decision model makes the call.

WHEN TO USE A DECISION MODEL

Decision models fit high-volume decisions where speed and cost matter:

Agent guardrails: check each action an agent proposes, such as exposing secrets, deleting data, or sending data outside your network, before it runs.
Content moderation: classify large volumes of user content against your policies.
Routing: pick which tool, workflow, or model handles a request.
Automated testing and navigation: choose the next action in a game, UI, or browser session.

API INTERFACE

You send each decision as one request. The request holds the state to evaluate and the questions to answer about it. The response holds one answer per question.

	Value
Endpoint	`POST https://inference.oumi.ai/v1/systemone`
Auth header	`X-API-Key: $OUMI_API_KEY`

You can start right away with our pretrained Iris model. Set model to projects/{project_id}/models/iris-latest, using the ID of any project you own. No training or deployment needed.

REQUEST

Field	Type	Required	Description
`model`	string	Yes	The decision model to call, as a path that includes your project: `projects/{project_id}/models/{model_name}`. For example, `projects/FOOBAR-123/models/iris-latest`.
`state`	string, object, or array	Yes	The content to evaluate: a plain string, or a JSON object or array of related context.
`questions`	object	Yes	The questions to answer, keyed by a name you choose. That name is the key for the answer in the response.

Each question has a type, instructions that state the question, and criteria that describe the possible answers. There are three question types:

Type	Use it for	`criteria` format
`noul`	Yes or no questions	An object with a `true` and a `false` description. Optional.
`choice`	Picking one option from a set	An object that maps each option name to its description.
`score`	Placing the state on an ordered scale	An array of descriptions, ordered from lowest to highest.

instructions and each criteria description can be a plain string, or a JSON object or array of structured guidance.

curl -X POST https://inference.oumi.ai/v1/systemone \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $OUMI_API_KEY" \
  -d '{
    "model": "projects/{project_id}/models/iris-latest",
    "state": {
      "customer_tier": "enterprise",
      "ticket": "My checkout page shows a blank screen after I click Pay. I have tried two browsers."
    },
    "questions": {
      "is_bug": {
        "type": "noul",
        "instructions": "Is the customer reporting a software defect?",
        "criteria": {
          "true": "The customer describes broken or unexpected product behavior.",
          "false": "The customer is asking a question or requesting a feature."
        }
      },
      "team": {
        "type": "choice",
        "instructions": "Which team should own this ticket?",
        "criteria": {
          "account": "Login, permissions, or profile issues.",
          "frontend": "Rendering, layout, or browser compatibility issues.",
          "payments": "Checkout, billing, or payment processing issues."
        }
      },
      "urgency": {
        "type": "score",
        "instructions": "How urgent is this ticket?",
        "criteria": [
          "Can wait for the next release",
          "Should be fixed this week",
          "Blocking revenue right now"
        ]
      }
    }
  }'

RESPONSE

Field	Type	Description
`model`	string	The model path from your request.
`answers`	object	One answer per question, keyed by the question names from your request.
`usage`	object	`inputTokens` and `outputTokens` for the request.

The fields in each answer depend on the question type:

Type	Fields
`noul`	`noul`: the probability that the answer is `true`, from 0 to 1.
`choice`	`choice`: the most likely option. `probabilities`: the probability of each option. `confidence`: how sure the model is of its pick.
`score`	`score`: the expected position on the scale, from 0 to the last index. `probabilities`: the probability of each position. `legend`: maps each position to its criteria description. `confidence`: how sure the model is of its pick.

Response

{
  "model": "projects/{project_id}/models/iris-latest",
  "answers": {
    "is_bug": {
      "type": "noul",
      "noul": 0.9252490387264224
    },
    "team": {
      "type": "choice",
      "choice": "payments",
      "confidence": 0.9057838757165146,
      "probabilities": {
        "account": 0.011615685025438532,
        "frontend": 0.05119506449688511,
        "payments": 0.9371892504776764
      }
    },
    "urgency": {
      "type": "score",
      "score": 1.8715208986248113,
      "confidence": 0.8411583967409364,
      "probabilities": {
        "0": 0.022584699202479703,
        "1": 0.08330970297022941,
        "2": 0.8941055978272909
      },
      "legend": {
        "0": "Can wait for the next release",
        "1": "Should be fixed this week",
        "2": "Blocking revenue right now"
      }
    }
  },
  "usage": {
    "inputTokens": 334,
    "outputTokens": 0
  }
}

In this example, the model is 93% sure the ticket is a bug, routes it to the payments team, and rates it as blocking revenue.

AVAILABLE MODELS

Iris

Call and fine-tune the Iris decision model.

Getting started

Oumi workflow

WHAT IS A DECISION MODEL

WHEN TO USE A DECISION MODEL

API INTERFACE

REQUEST

RESPONSE

AVAILABLE MODELS

Iris