ABOUT THE MODEL

What is Laya AI?

Laya AI is a workspace for short-text decisions. Provide a compact state and a typed question. The model returns a yes/no probability, a choice distribution, or a score instead of a paragraph.

What the model returns

A noul question estimates the probability of a yes answer. A choice question selects from labels you provide and returns their probability distribution. A score question works with ordered levels. These are structured answers for routing, review, or scoring rules; they are not generated chat messages.

For example, paste a short support request as the state and ask which team should handle it. The answer is a label and its distribution, so your application can decide whether to route the request automatically or send it for review.

English or multilingual

Select laya-english for English text or laya-multilingual for multilingual input. Each question has a total token budget: 512 for English and 1024 for multilingual, including text, instructions, and labels. The gateway rejects inputs that would be truncated. Test results on your own text before automating a consequential decision.

Where Laya comes from

Convai Innovations publishes the Laya code and model weights under Apache 2.0. This site runs the English and multilingual checkpoints through its own hosted API. The upstream project also describes other checkpoints and local deployment options; those are separate from the two model choices offered here.

Read the upstream Laya repository ↗

Use the same shape in code

The playground and API accept the same state, model, and questions fields. You can test a rule here, inspect the answer, and move the request into your application.