Convai's Laya targets fast AI decisions and guardrails

Convai Innovations has released Laya, a text model it frames as a "System-One" component: a fast, narrow decision-maker rather than a general-purpose chat assistant. According to the model's Hugging Face page, Laya is designed for routing, scoring and guardrail tasks — the kinds of quick classification calls that sit around larger language models in production pipelines.

The naming is a deliberate nod to the psychology of fast versus slow thinking. Where a "System-Two" model deliberates and generates long answers, Laya is pitched at the reflexive layer that decides where a request should go, whether it's safe to process, and how to rank or filter candidates.

Why it matters

As teams stitch together multi-model systems, the unglamorous plumbing increasingly determines cost and reliability. A dedicated decision model can offload work that would otherwise burn tokens on a large generator.

• Routing: directing queries to the right model or tool
• Scoring: ranking or filtering outputs and candidates
• Guardrails: flagging unsafe or out-of-scope inputs before they reach a bigger model

Convai has not published parameter counts, context length or benchmark figures alongside the initial release, so buyers will want to test Laya against their own routing and safety workloads. The model is distributed under a custom "other" license, which developers should review before deploying it in commercial settings.