Skip to content

Laya is now available on Vercel AI Gateway, free through Boundless

Last updated: September 30, 2026

An open-weight System 1 model for fast, structured decisions in agents and automated workflows.

Built by ConvAI Innovations, Laya evaluates text and returns choices, scores, and probabilities your code can act on.

For developers building agents and workflow applications, Laya offers a dedicated model for recurring decisions: selecting a workflow path, classifying inputs, or scoring them against defined criteria. Free hosted access lets you test Laya in your application without running the model yourself.

Start building with Laya →

How Laya works

Most language models generate text one token at a time. Laya answers defined questions in a single forward pass, without generating a written response. You supply the input, the questions, and any available options.

This makes Laya useful for recurring decisions inside agents and workflows: classifying inputs, selecting among predefined tools, or deciding which path to take next. Your application uses its output alongside your own rules to control actions, permissions, and escalation. A generative model can handle writing or further reasoning when needed.

Laya’s open weights also let developers fine-tune and self-host their own version for decisions specific to their product.

Use cases where Laya shines

Laya is designed for repeated decisions with defined outcomes. These are useful starting points for evaluating it in your application:

Route and prioritize customer support tickets

Classify incoming tickets by issue, assign an urgency level, and select the appropriate queue. Your application can combine those outputs with account information and escalation rules.

Qualify inbound sales leads

Evaluate inquiries against your qualification criteria, such as use-case fit, purchase intent, and timeline. Use the results to route leads to sales, nurture, or further review.

Sort invoices for review

Classify invoice descriptions and accompanying messages by expense category or review reason. Combine those decisions with deterministic checks for totals, duplicates, and approval limits.

Organize and prioritize customer feedback

Group feedback by product area, distinguish bug reports from feature requests, and score reported impact against a defined rubric. Route the results to the relevant team or backlog.

For each workflow, start with labeled examples from your application. Measure decision quality and use the results to determine which actions can be automated and which need review.

Watch Laya and Opus play Tetris

See Laya and Opus tackle the same game of Tetris, making decisions as the board changes. The game illustrates a repeated decision loop: evaluate the current state, choose an action, and execute it. Developers can apply that pattern to routing requests, selecting tools, and choosing workflow paths.

The departments and urgency levels are defined by the developer. Laya supplies the decision signals; application code controls routing, permissions, and escalation.

Get started with Laya for free on Vercel AI Gateway

Laya is available free through Boundless for developers building agents and automated workflows. Explore how structured decisions fit into your application, from classification and scoring to routing and action selection.

Start building with Laya →

FAQ

What can I use Laya for?

Laya supports structured decisions such as classification, routing, scoring, and selecting among predefined actions. You provide the input and possible outcomes; your application uses the results to determine what happens next.

How is Laya different from a chat model?

Laya evaluates inputs and returns choices, scores, and probabilities without generating a written response token by token. A chat model can handle tasks such as writing replies or explaining results alongside it.

Is Laya free to use?

Laya is served by Boundless on Vercel AI Gateway and is free to use for a limited time here.

Can I fine-tune Laya on my own data?

Laya’s open weights allow developers to fine-tune it independently. This launch provides hosted inference through Vercel AI Gateway, fine-tuning and serving customized weights are separate steps.