Laya
Multilingual, non-autoregressive System 1 decision engine: typed choice, score and noul
decisions over any state, in a single forward pass.
A choice question picks one of several options, a score question places the state on a
scale, and a noul question gives the probability that the answer is yes. Laya runs on your own
hardware, reads 100+ languages, and its Router picks the checkpoint for each request.
Start here
Install Laya and make a first decision with the README’s Installation and Quickstart. They stay in the README so there is one copy to keep current.
Find a guide
| To | Read |
|---|---|
| get typed values back from a JSON schema or a pydantic model | Schema-driven decisions |
| log, redact, cache or gate every decision without forking Laya | Prediction hooks |
| adopt Laya incrementally without granting execution permission | Staged adoption |
| route, screen or triage inside a LangChain or LangGraph app | LangChain & LangGraph |
| run Laya from a terminal or connect its built-in MCP stdio server | Command line and MCP server |
run the SDK or the laya-serve HTTP API in a container, on CPU or an NVIDIA GPU |
Docker quickstart |
| build for ARM64 hosts or DGX Spark | ARM64 and DGX Spark containers |
| specialise a checkpoint for your own decisions | Browser-agent fine-tuning example and the fine-tuning notebook |
| judge benchmark results and deployment limits | Benchmarks and known limits |
| look up a class, function or parameter | Python API reference |
| score a labelled dataset, compare to a baseline, or gate a build on it | Evaluation harness |
Routing, the HTTP API and confidence gating are in the README for now.