Super Decider quickstart

This guide takes you from an API key to acting on your first decision.

Before you start

You need:

• An ASI:Cloud account. Sign in at asicloud.cudos.org.
• An LLM API key.
• Balance on your account. See Payments to top up by card or crypto.

Save your key in an environment variable:

export ASI_API_KEY="your-api-key"

1. Send a request

A request has three parts: the model, the state the model reads, and the questions you want answered. This example is a support ticket about an order that has already shipped.

cURL

curl https://inference.asicloud.cudos.org/v1/systemone \
  -H "Authorization: Bearer $ASI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "super-decider",
    "state": "Order 8812 shipped on Oct 1. The customer writes: \"I want to cancel and get my money back.\" Policy: shipped orders cannot be cancelled; they can be returned within 30 days of delivery for a full refund.",
    "questions": {
      "next_step": {
        "type": "choice",
        "instructions": "What should support do next?",
        "criteria": {
          "refund_now": "Refund without a return",
          "start_return": "Start a return",
          "escalate": "Hand over to a human"
        }
      },
      "allowed": {
        "type": "noul",
        "instructions": "Is cancelling this order allowed by the policy?"
      },
      "urgency": {
        "type": "score",
        "instructions": "How urgent is this ticket?",
        "criteria": ["low", "medium", "high"]
      }
    }
  }'

Python

import os
import requests

resp = requests.post(
    "https://inference.asicloud.cudos.org/v1/systemone",
    headers={"Authorization": f"Bearer {os.environ['ASI_API_KEY']}"},
    json={
        "model": "super-decider",
        "state": (
            "Order 8812 shipped on Oct 1. The customer writes: "
            '"I want to cancel and get my money back." '
            "Policy: shipped orders cannot be cancelled; they can be returned "
            "within 30 days of delivery for a full refund."
        ),
        "questions": {
            "next_step": {
                "type": "choice",
                "instructions": "What should support do next?",
                "criteria": {
                    "refund_now": "Refund without a return",
                    "start_return": "Start a return",
                    "escalate": "Hand over to a human",
                },
            },
            "allowed": {
                "type": "noul",
                "instructions": "Is cancelling this order allowed by the policy?",
            },
            "urgency": {
                "type": "score",
                "instructions": "How urgent is this ticket?",
                "criteria": ["low", "medium", "high"],
            },
        },
    },
    timeout=30,
)
resp.raise_for_status()
answers = resp.json()["answers"]

JavaScript

const res = await fetch("https://inference.asicloud.cudos.org/v1/systemone", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.ASI_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "super-decider",
    state:
      'Order 8812 shipped on Oct 1. The customer writes: "I want to cancel and get my money back." ' +
      "Policy: shipped orders cannot be cancelled; they can be returned within 30 days of delivery for a full refund.",
    questions: {
      next_step: {
        type: "choice",
        instructions: "What should support do next?",
        criteria: {
          refund_now: "Refund without a return",
          start_return: "Start a return",
          escalate: "Hand over to a human",
        },
      },
      allowed: {
        type: "noul",
        instructions: "Is cancelling this order allowed by the policy?",
      },
      urgency: {
        type: "score",
        instructions: "How urgent is this ticket?",
        criteria: ["low", "medium", "high"],
      },
    },
  }),
});
const { answers } = await res.json();

Already using an OpenAI client? Super Decider also accepts chat completions.

2. Read the answers

The response has one answer per question, keyed by the question's id:

{
  "answers": {
    "next_step": {
      "type": "choice",
      "choice": "start_return",
      "probabilities": {"refund_now": 0.02, "start_return": 0.96, "escalate": 0.02},
      "confidence": 0.96
    },
    "allowed": {"type": "noul", "noul": 0.03},
    "urgency": {
      "type": "score",
      "score": 0.59,
      "legend": {"0": "low", "1": "medium", "2": "high"},
      "probabilities": {"0": 0.43, "1": 0.54, "2": 0.02},
      "confidence": 0.54
    }
  },
  "model": "super-decider"
}

The response also has usage with the input and output token counts.

• next_step: the model is 96% sure support should start a return.
• allowed: the probability that cancelling is allowed is 0.03, so the answer is almost certainly no.
• urgency: the expected level is 0.59 on a 0 (low) to 2 (high) scale, between low and medium.

3. Act on the answers

The probabilities let your code set the rules. Act on its own when the model is sure, and send everything else to a person:

next_step = answers["next_step"]

if next_step["choice"] == "start_return" and next_step["confidence"] >= 0.9:
    start_return(order_id=8812)
elif answers["urgency"]["score"] >= 1.5:
    page_on_call_agent(order_id=8812)
else:
    send_to_support_queue(order_id=8812)

Pick your thresholds by running the model on a set of past cases where you know the right answer. See Confidence.

Next steps

• State: what to put in the state and how to structure it.
• Questions: write clear questions and choose the right type.
• Thinking: turn on a short reasoning step for calculations and multi-step questions.
• API reference: every request and response field.