It is a multilingual, non-autoregressive System One decision engine that scores typed questions over text, email, tickets, or JSON in one forward pass.

Does Laya generate text?

No. It returns structured fields such as choice, score, noul, confidence, and routing metadata rather than free text.

Can Laya run locally?

Yes. The site describes Apache 2.0 weights and a Python package for local use. It lists Python 3.10 or newer plus dependencies including torch and transformers.

What does the hosted API cost?

The hosted API uses prepaid packs. The site says a typical /v1/decide call costs approximately $0.0004, with packs ranging from $5 to $500.

What does the API accept?

Core decide accepts string or structured JSON state and a map of questions typed as noul, choice, or score.

Is there a batch endpoint?

Yes. /v1/batch/decide applies the same questions to many states, with up to 20 states per request.

What is the confidence gate?

The /v1/gate endpoint applies a policy and returns allow, review, or block without an upstream call or token burn.

Is the hosted API official?

The API documentation calls it an unofficial community gateway and says it is not affiliated with Convai Innovations, TypeSafe, or Impossibl.

Laya AI Model (unofficial)

Unofficial docs and hosted API for the Laya AI decision model

AI Tools Developer Tools

Visit website

Launched: Nov 8, 2026
Website: layaaimodel.com
Pricing: Free
Platforms: Web, API
Based in: United States

Community documentation and a hosted prepaid API for Laya AI, a "System One" decision model that returns typed choice, score or yes/no outputs instead of chat text. Includes a live playground and comparison with Jev.

What Laya AI Model (unofficial) does

Typed decision outputs: Returns choice, score, and yes/no results with confidence and routing metadata instead of generated prose.
Structured state input: Accepts a string, object, or array as state, with a map of typed questions and criteria.
Batch decisions: Runs the same questions over many states with the batch endpoint, supporting up to 20 states.
Confidence gate: Applies a policy to return allow, review, or block without an upstream call or token burn.
Ready-made application APIs: Provides templates for email and support triage, moderation, lead qualification, prompt guarding, scam spotting, and more.
Local or hosted operation: Use Apache 2.0 weights with the Python package locally, or use prepaid hosted API keys.

Who it's for

Developers building agents and software workflows that need typed classification, scoring, routing, moderation, or confidence gates.

Why people use Laya AI Model (unofficial)

Chat-oriented language models return free text that must be parsed for routing, labels, scores, or yes/no decisions.

Laya evaluates a supplied state against typed questions and returns structured decision fields in one forward pass, locally or through a prepaid API.

The project combines open Apache 2.0 local weights with a hosted API, and focuses on typed decisions rather than text generation.

Pricing

Laya AI Model (unofficial) has paid plans from $5.

Starter

Prepaid pack for approximately 12,500 typical /decide calls.

11,900,000 input tokens
60 RPM

$10

Prepaid pack for approximately 25,000 typical /decide calls.

23,800,000 input tokens
60 RPM

$20

Prepaid pack for approximately 50,000 typical /decide calls.

47,600,000 input tokens
90 RPM

Staging

$50

Prepaid pack for approximately 150,105 typical /decide calls.

142,900,000 input tokens
120 RPM

Platforms

Laya AI Model (unofficial) runs on Web, API.

Laya AI Model (unofficial) FAQ

What is Laya Model?: It is a multilingual, non-autoregressive System One decision engine that scores typed questions over text, email, tickets, or JSON in one forward pass.
Does Laya generate text?: No. It returns structured fields such as choice, score, noul, confidence, and routing metadata rather than free text.
Can Laya run locally?: Yes. The site describes Apache 2.0 weights and a Python package for local use. It lists Python 3.10 or newer plus dependencies including torch and transformers.
What does the hosted API cost?: The hosted API uses prepaid packs. The site says a typical /v1/decide call costs approximately $0.0004, with packs ranging from $5 to $500.
What does the API accept?: Core decide accepts string or structured JSON state and a map of questions typed as noul, choice, or score.
Is there a batch endpoint?: Yes. /v1/batch/decide applies the same questions to many states, with up to 20 states per request.
What is the confidence gate?: The /v1/gate endpoint applies a policy and returns allow, review, or block without an upstream call or token burn.
Is the hosted API official?: The API documentation calls it an unofficial community gateway and says it is not affiliated with Convai Innovations, TypeSafe, or Impossibl.