Three ways in

Call the Decision API

Myles-G 12B and Myles-G 31B, hosted by us: £0.0015 per million input tokens, output free. Paid for with prepaid credit.

Decision API

Run it yourself

The same API on your own servers or GPU, including disconnected networks, under a licence per deployment.

Licensing

Quick start

One request, one routing decision. Every option comes back with a probability, not just the winner.

Request

curl https://api.launchables.ai/v1/decide \
  -H "Authorization: Bearer $LAUNCHABLES_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "state": "Charged twice for order 4471, need the extra payment back",
    "questions": {
      "route": {
        "type": "choice",
        "instructions": "Which team should handle this?",
        "criteria": ["billing", "returns", "technical", "other"]
      }
    }
  }'

Response

{
  "model": "myles-g-12b",
  "answers": {
    "route": {
      "type": "choice",
      "choice": "billing",
      "probabilities": {
        "billing": 0.93, "returns": 0.04,
        "technical": 0.01, "other": 0.02
      }
    }
  }
}

Illustrative probabilities. Several questions about one situation go in one request and are answered in parallel. The full API reference

Three kinds of question

noul

Yes or no.

The probability that the answer is yes. Is this a scam? Is the refund within policy? Does the answer cite its source?

choice

One of several labels.

A probability for every label, up to 26. Which team? Which category? Which tool should the agent call?

score

A level on a scale.

The most likely level, a probability-weighted score and a probability for every level. How urgent? How risky?

Decide from an image

Myles-G 12B and Myles-G 31B read PNG and JPEG images in the same request as the text. Send each image once in assets, write its name where it belongs in the state or the question, and the answers come back exactly as they do for text: a calibrated probability for every option. Approve this receipt? Is it safe for the agent to tap this? Fit to ship? Which item is the target?

Request body

{
  "state": {
    "receipt": "receipt_photo",
    "note": "Customer asks for a refund"
  },
  "questions": {
    "total_matches": {
      "type": "noul",
      "instructions":
        "Does the total on receipt_photo equal 23.40?"
    }
  },
  "assets": {
    "receipt_photo": {
      "format": "jpeg",
      "data_base64": "/9j/4AAQSkZJRgABAQ..."
    }
  },
  "image_budget": 280
}

Response

{
  "model": "myles-g-12b",
  "answers": {
    "total_matches": {
      "type": "noul", "noul": 0.94
    }
  },
  "usage": {
    "input_tokens": 331, "output_tokens": 1
  }
}

About a fifth of a second. The median time to answer with one image at the default size, on our current GPU.
Billed as input tokens. An image counts as ordinary input tokens, charged once per request when it is in the state.
Choose the detail. image_budget is 70, 140 or 280 tokens per image, and 280 is the default. Use 280 for receipts and screenshots with small text.
Same endpoint. POST /v1/decide, with the image named in the state, the instructions or the criteria.

Illustrative probability and token counts. Speed measured 16 Oct 2026.

Images in the docs Quick start

Decision API

Myles-G 12B and Myles-G 31B, hosted by Launchables

£0.0015 per million input tokens, output free, excluding VAT

Get an API key

Prepaid credit. Buy credit in the console from £5 by card, with optional automatic top-up. Credit never expires, and keys pause when it runs out.
Limits per key. 1,200 requests a minute by default, with no daily cap. Ask for more.
One endpoint. POST /v1/decide, with GET /v1/models and GET /health.
Need it in house? The same API runs on your own hardware under a licence, with nothing sent to us.

From your code

Python

import os, requests

def decide(state, questions):
    r = requests.post(
        "https://api.launchables.ai/v1/decide",
        headers={"Authorization": f"Bearer {os.environ['LAUNCHABLES_API_TOKEN']}"},
        json={"state": state, "questions": questions},
        timeout=30,
    )
    r.raise_for_status()
    return r.json()["answers"]

refund = decide(
    "Bought 12 days ago, no receipt. Policy: 30 days with proof of purchase.",
    {"ok": {"type": "noul", "instructions": "Is a refund permitted?"}},
)["ok"]["noul"]

# keep what clears your threshold, hand the rest to a person
print("approve" if refund >= 0.9 else "decline" if refund <= 0.1 else "to a person")

JavaScript

const response = await fetch('https://api.launchables.ai/v1/decide', {
  method: 'POST',
  headers: {
    Authorization: `Bearer ${process.env.LAUNCHABLES_API_TOKEN}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    state: { from: 'J. Smith', message: 'Is my parcel lost? It is a week late.' },
    questions: {
      priority: {
        type: 'score',
        instructions: 'How urgent is this?',
        criteria: ['Routine', 'This week', 'Today'],
      },
    },
  }),
});
const { answers } = await response.json();
console.log(answers.priority.level, answers.priority.probabilities);

Build on decisions you can trust.

Get an API key Request a self-hosted evaluation

A question in, a probability for every option out.

Three ways in

Try it in your browser

Call the Decision API

Run it yourself

Quick start

Three kinds of question

Yes or no.

One of several labels.

A level on a scale.

Decide from an image

Decision API

From your code

Resources

API reference

Model card

Evaluation weights

Evidence

Build on decisions you can trust.