Torchcast.ai

Probabilities, not prose

Send a piece of text (an email, a ticket, a document, a JSON record) and the questions you need answered about it. Every answer comes back as a calibrated probability your code can branch on directly.

Built for developers who route, triage or screen text at scale: support tickets, emails, documents, records.

Ask the question the way your code needs it

noul

Yes or no (noul)

A single number: the probability that the answer is yes. Is this email a refund request? Was the parcel damaged in transit?

0.97

choice

One of your options

The most likely of the options you name, with the probability of every option. Which team should handle this ticket?

billing · 0.96

score

A level on your scale

A position on your own scale of up to 10 levels, with the probability of each level. How severe is this issue?

score 1.06 · Inconvenient 0.84

Two questions about a support email

Request
curl https://api.torchcast.ai/v1/decisions \
  -H "Authorization: Bearer $TORCHCAST_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "torchcast-decider-latest",
    "state": "Hi, I was charged twice for my October invoice. Please send the extra payment back to my card.",
    "questions": {
      "refund_request": {
        "type": "noul",
        "instructions": "Is the customer asking for money back?"
      },
      "team": {
        "type": "choice",
        "instructions": "Which team should handle this email?",
        "criteria": {
          "billing": "Payments, invoices and refunds.",
          "technical": "Bugs, errors and outages.",
          "sales": "Plans, pricing questions and upgrades."
        }
      }
    }
  }'
Response
{
  "model": "torchcast-decision-4b",
  "answers": {
    "refund_request": {
      "type": "noul",
      "noul": 0.97
    },
    "team": {
      "type": "choice",
      "choice": "billing",
      "confidence": 0.96,
      "probabilities": {
        "billing": 0.96,
        "technical": 0.01,
        "sales": 0.03
      }
    }
  },
  "usage": {
    "input_tokens": 61,
    "output_tokens": 2
  }
}

In this illustrative response, the refund question comes back at 0.97 and the email goes to billing at 0.96. Pick a threshold that fits the cost of a mistake in your application.

Torchcast does not store or log the text you send. It is used only to answer that request.

From sign-up to first answer

  1. 01

    Create an account

    Sign up in the developer console with your email address. There is no password: we send you a sign-in link.

  2. 02

    Add credit

    The API draws on a prepaid balance in US dollars. Buy credit by card in the console, under Billing > Add credits. Sales tax or VAT is added on top where it applies, and your balance is credited with the pre-tax USD amount. Credit expires 365 days after it is added. For invoicing or larger amounts, contact us at contact@torchcast.ai.

  3. 03

    Create a key

    Create an API key in the console. You can give each key its own spend limit.

  4. 04

    Send a request

    One endpoint, one JSON body: the text, and up to 128 named questions about it.

Prepaid, billed per token

  • Each successful request is charged from your prepaid USD balance by its billable input tokens: the text, counted once, plus your questions and their answer options. There are no subscriptions.
  • Prices are per million billable input tokens and can differ by model. The current price of each model is listed by GET /v1/models and in the console.
  • A request that returns an error from the API is never charged, including one the API ends with a timeout. A successful request sent again is charged again. Per-key spend limits cap what any one key can spend.
  • Credit expires 365 days after it is added. Prices exclude tax; any sales tax or VAT is added at checkout.

Start with a key