playground

Try the models

Typed, calibrated decisions over any state, in one call.

10 free runs a day, no account needed. Sign in to pay per token instead.

state and questions1,079 / 4,000

{
  "state": {
    "channel": "email",
    "customer": {
      "plan": "Pro",
      "months_subscribed": 14
    },
    "messages": [
      {
        "from": "customer",
        "text": "I was charged twice for September. I want the duplicate charge refunded, today please."
      },
      {
        "from": "agent",
        "text": "Sorry about that. I can see two charges of $49 on 2 September."
      }
    ]
  },
  "questions": {
    "refund_requested": {
      "type": "noul",
      "instructions": "Is the customer asking for money back?"
    },
    "team": {
      "type": "choice",
      "instructions": "Which team should handle this?",
      "criteria": {
        "billing": "Charges, invoices, payments and refunds.",
        "technical": "The product is not working as expected.",
        "account": "Login, plan changes and access."
      }
    },
    "urgency": {
      "type": "score",
      "instructions": "How urgent is this?",
      "criteria": [
        "Can wait a week",
        "Within a few days",
        "Today",
        "Immediately"
      ]
    }
  }
}

The same body /v1/systemone takes. Questions are noul (yes or no), choice (up to ten labels) or score (an ordered rubric).

decisions

Each answer appears here as a distribution.

what it will not do

It only picks between the options you give it — at most ten per question — and decides from the state alone, with nothing looked up. Treat its confidence as a probability, and check the decisions that matter. Full numbers.

same model, over the api

This page calls the same model the API does. The docs show the request that produces exactly this.