Introduction

OneHot is an API from Sqwish for making decisions about text. You send the text and a question, with the answers you allow. OneHot’s models, called kettei, read the text and give each answer a probability. Every result has a probability for each outcome you allowed, and they add up to 1. This is a real result from kettei-core for a support ticket and the question “Which team should handle this ticket?” (quickstart):

{
  "probabilities": {"billing": 0.927, "technical": 0.054, "sales": 0.019},
  "top": "billing",
  "p_top": 0.927
}

top is the likeliest outcome and p_top its probability. The models only score the outcomes you listed. They never write text of their own, so your code always gets one of your outcomes, with numbers it can act on.

from sqwishai import OneHot

refund = {"id": "refund", "kind": "binary", "question": "Does the customer ask for money back?"}

with OneHot() as client:  # reads your key from SQWISH_API_KEY
    result = client.decide(context="Please refund the duplicate charge on order 4471.",
                           decisions=[refund])

print(result["decisions"]["refund"]["probabilities"])  # one for "yes" and one for "no"

What you can build

Route work: which team should take a ticket, and how urgent it is.
Guard actions: whether a tool call an agent wants to make is risky, or whether a message tries to override its instructions.
Check AI output: whether an answer is supported by its sources, or relevant to the question.

The examples show each of these with a request you can copy.

How it works

Describe a decision

Write a question and the answers you allow. A decision can be a yes or no question, a choice of one category, or a rating on a scale (core concepts).

Send it with your text

Call POST /v1/decide with the text and up to 128 decisions about it. Each decision gets its own answer.

Act on the answer

Read each answer’s probabilities, or add a policy so the answer tells you what to do, such as sending an unclear case to a person (read results).

A probability says how strongly the model favours an answer over the others you listed. It is not a guarantee that the answer is right. Test your decisions on examples whose answers you know before you let them act on their own.

Try it

The playground in the console runs decisions without an account, and has ready-made examples to start from.

Open the playground

Pick an example, change the question and see each answer’s probability.

To call the API from your code, sign up in the console and create an API key. New accounts may wait on a waitlist first (access and keys).

Next steps

Quickstart

Install a client and get your first answer in a few minutes.

Core concepts

Decisions, outcomes, probabilities, models and families.

Improve a decision

Use your own examples to make a decision more accurate.

Pricing

What decisions, fine-tuning and keeping a model ready cost.

Your data

Your data is not used to train the models anyone else uses, and a model fine-tuned on your data answers only you. For more, see data residency, the privacy policy and the terms of service.

Get started

Make decisions

Improve accuracy

Account and billing

Data and privacy

Migrate

What you can build

How it works

Try it

Open the playground

Next steps

Quickstart

Core concepts

Improve a decision

Pricing

Your data

​What you can build

​How it works

​Try it

Open the playground

​Next steps

Quickstart

Core concepts

Improve a decision

Pricing

​Your data

What you can build

How it works

Try it

Next steps

Your data