system-one: typed decisions in TypeScript, as Promises or Effects
One set of question definitions drives a zero-dependency Promise client and an Effect service across Jev, Cloudflare AI Gateway and Laya.
BUILDS
113 builds made with System One models.
Agents, routers, scorers and games — built on models that decide. Every entry links to the models it uses, so you can pull the same one and start from there.
113 builds · page 5 of 5
A day of RL training Laya on Doom
Shantanu Goel trained Laya for about 24 hours on FreeDoom's first level from text-only observations; zero-shot Jev still played far better.
zero-shot-ie-bench: seven extraction and classification families side by side
GLiNER, GLiClass, Laya, Von, Jev and more in one web UI, with demos, benchmarks, model sizes and licences.
Does Laya's advertised zero-shot win over Jev reproduce?
A frozen protocol on four datasets with raw predictions, calibration and latency: Laya ties Jev on one and trails by 10–33 points on the others.
Chakravyuha: a browser ring maze solved one Laya decision at a time
A Mahabharata-inspired ring formation walked step by step to its centre, with every move chosen by a local Laya model instead of the hosted Jev API.
chunklaya: where Laya stops reading, and a chunking fix
Shows across 400 documents that the decision head's attention falls off after roughly 200 tokens, then provides a chunking harness that restores accuracy on long inputs.
Hermes skill_router: pick the right skill before the agent runs
A Hermes plugin that classifies each task against an index of about 300 skills with a local Laya model and injects the best match as context for the turn.
jeffrey: a coding-agent CLI where a decision model steers
At every step Jev or Laya picks the next tool, scores progress and estimates risk; the language model only executes and never decides.
Laya-CoreML: ~5 ms decisions on the Neural Engine
Validated Core ML ports of every Laya checkpoint: 4.98 ms median on an M3 Max and 2.78× better energy use than MLX.
laya-mlx: Laya on Apple Silicon at 60 decisions a second
An MLX port tuned to run in under 1 GB of memory; its demo plays Snake at 60 decisions per second on an M3 Max.
laya-vision: typed decisions about images
An experimental vision variant that keeps Laya's question format but takes an image as its input.
@receptron/laya: Laya from Node.js and TypeScript
An npm package that runs the model with ONNX Runtime, without Python or PyTorch, using the same request shape as a hosted System One API.
Laya: the open-source decision engine
The model itself: typed choice and score questions answered in a single forward pass, Apache 2.0, about 33 ms per question on a T4.
Laya demo Space: eight workflows, nothing to install
A Gradio demo on Hugging Face covering email triage, phishing detection, guardrails, ticket routing, RAG filtering and multilingual routing.
A Laya browser agent trained on Mind2Web
Chooses the next web action with Laya, trained and evaluated on the Mind2Web benchmark; the weights are on Hugging Face.
Laya in ONNX
An ONNX export that runs wherever onnxruntime does, from browsers to edge boxes, later joined by an fp16 variant.
Why decision models don't need to be autoregressive
Laya's author on building decision models, training against proper scoring rules, and why the model is open.
Many of the first builds here were collected by madewithlaya.com. Every entry links to its creator’s own post, repository or site.