Laya

Laya is an open decision model stack for production routing and scoring. You pass any state (email, ticket, JSON doc) plus typed questions — choice, score, or noul — and get structured answers with confidences in one forward pass. Three checkpoints cover English, 100+ languages, and typed-decision workflows; a Router picks the right one per request after a sub-millisecond language/script check. Trained with RL against proper scoring rules (RLCD). Install with pip install laya.

Most agent stacks still burn a full LLM call to classify a ticket, then hope the JSON comes back clean. Founders shipping support, ops, and risk pipelines need fast, calibrated labels without hallucinated free text. Laya is that layer: milliseconds, typed outputs, multilingual out of the box — the cheap System 1 beside your slower planner.

pip install laya, then Router(preload=True). Define questions with type, instructions, and criteria. Call router.predict(state, questions). English traffic hits the ModernBERT English checkpoint; non-Latin scripts route to the multilingual model automatically. Batching drops cost further (~7 ms per question on a T4 in their numbers). Override model when you want a fixed checkpoint.

Not another chat model and not a coding agent harness. No token generation means no parse step and no invented labels. Complementary to memory hubs and browser agents — this is the decision kernel you put under them when the job is classify, score, and gate.

• Public demo available
• API / SDK surface

• RTK
• Paritok