Laya ecosystem
Laya Projects & Ecosystem
Track useful Laya runtimes, local servers, playgrounds, benchmarks, and real integrations. GitHub project data refreshes daily.
Submit a projectData updated: Nov 6, 2026
Laya ecosystem directory
NandhaKishorM/laya
Reference Laya implementation, Router, checkpoints, benchmarks, fine-tuning and docs.
mizorewww/laya-mlx
Native MLX runtime for local Laya inference on Apple Silicon.
mizorewww/laya-coreml
Core ML port that runs Laya on the Apple Neural Engine without PyTorch or MLX; about 5 ms per short decision on an M3 Max.
ollaya-dev/ollaya
Ollama-style CLI, daemon and desktop app that pulls and serves open decision models (Laya, Kev, Von, Decider, NLI, GLiClass and more) behind a TypeSafe-compatible API.
receptron/laya
Run Laya from Node.js and TypeScript through ONNX Runtime.
FluidInference/FluidUse
Local computer use on Apple silicon; its Swift LayaManager runs Laya natively on Core ML, alongside other on-device decision models.
codejunkie99/keel
Local-first macOS coding workspace that uses local Laya to pick task routes.
ipenywis/laya-ultrafast
Browser-agent experiment using local Laya as the fast decision layer.
wdobry/laya-playground
Local Playground, model-driven games, benchmark and agent integration skill.
ThinkFlowLab/system1-agents
Prebuilt browser-use, computer-use, robotics and game agents driven by Laya or Jev.
lkarlslund/laya.cpp
Native C++ inference on ggml with CUDA, Vulkan and Core ML backends, all three checkpoints, a Jev-compatible HTTP server with batching and prebuilt binaries.
virajbhartiya/laya-vs-jev
Side-by-side T-Rex experiment comparing local Laya and hosted Jev.
marcreichel/laya-php
PHP SDK for laya-serve with typed enum, int and bool answers, a Laravel service provider, a test fake and support for per-request token limits.
1Panel-dev/laya-server
Self-hosted Jev-compatible Laya API with a web console, API keys and Docker setup.
r33drichards/laya-vision
Adds image inputs to Laya: typed, calibrated choice, noul and score answers about a picture plus optional text, in one forward pass. It can also play simple games from pixels.
alvarobartt/sys1
System One compatible API server in Rust with candle, token-based batching and Laya heads.
uu889/laya-opencv
One-click installer and local web workbench for Laya (formerly laya-workbench), in Chinese and English: fine-tune Laya or build a Laya-format model from any encoder, then pair it with OpenCV and hard rules for visual inspection such as fruit picking, pest monitoring and steel-pipe QC.
0xBakeer/arbiter
Local serving layer with Jev-compatible API, routing, batching and a Playground.
PromptEngineer48/laya-vs-jev-arena
Laya and Jev drive live game decisions with latency and run metrics.
afshinm/laya-mps
Memory-conscious local Laya runtime and server for Apple GPU / MPS.
tc3oliver/laya-apple
Runs Laya on the MLX GPU and the Apple Neural Engine at the same time: short single-question decisions go to the Neural Engine, longer work stays on the GPU.
ChenneyZhuang/laya-browser-agent
Local, open-weight alternative to Jev for browser-agent decisions: no cloud, no API key and no screenshots; works with browser-use jev-ultrafast.
rupeshs/laya-openvino
Unofficial OpenVINO backend for fast CPU inference, with a pre-converted INT8 English checkpoint (405 MB, 98.9% top-answer agreement with PyTorch).
sfmqrb/gutcheck
grep for meaning: ask a plain-English question about every line, log record, JSON field or diff hunk and explore the answers in a terminal UI. One binary, runs the Laya checkpoints on CPU.
F0Rextasy/omp-laya-judge
Local System-1 judge MCP server and skill for the oh-my-pi coding agent: Laya gates notes, file edits, shell commands and error recovery without spending LLM tokens.
DaragonTech/LibLayaX
Shared library with a plain C API, built on laya.cpp, that runs Laya inside any application with no server or Python; prebuilt for six platforms, with Delphi (DLaya) and Lua (LLaya) bindings.
killkli/open-jev-laya
Browser-only Laya multilingual demos with ONNX and WebGPU, including decision games.
aryanbhujade/laya-mlx-voice-browser
LayaBrowse: control Safari or Chromium browsers by voice on an Apple-silicon Mac, with a local Laya model choosing each browser action. No API key or generative LLM.
antonellof/laya-trader
Paper-trading demo: Laya MLX reads crypto and S&P 500 signals and a strategy turns its probability into trades, with a live dashboard and backtests. Paper trading only.
wexare-ai/browser-laya
npm library (@wexare/laya-web) and hosted playground that run Laya entirely in a browser tab with no server or API key; a 900 MB first download, with an 8-bit build for phones that matches the full model on 99.3% of a 720-question test.
instax-dutta/sysone-bench
Independent byte-identical System One benchmark for Laya, Jev and open baselines.
Makankn/laya-plays-google-snake
Laya plays the real Google Snake game from screen pixels and arrow keys: a visual tracker builds the board and Laya chooses among safe moves.
alipyth/traffic-solver-laya-
A 3D city whose traffic lights are controlled by Laya: every decision is shown on screen, from what Laya read to what the light did.
elisaterumi-ai/layapet
A virtual pet in Portuguese: free-text messages go to the multilingual Laya checkpoint on CPU, which picks feed, play, sleep or pet zero-shot, with no keyword lists.
azalio/doomLaya
Laya and Jev play FreeDoom: the model picks the action, target and weapon while a controller moves and aims, with training notes and a side-by-side comparison.
jinlio/mjbrain
Riichi mahjong advisor built on a fine-tuned Laya checkpoint (Mahjong Soul first). Recommend-only: it suggests moves and never plays them for you.
brennanMKE/laya-go
Pure-Go Laya inference with no Python or cgo: Metal GPU kernels on Apple Silicon, a native CPU engine elsewhere, parity with the Python package and a drop-in laya-server.
lcbkmm/laya-thalamus
Decision middleware for AI agents: routes tool calls with local Laya and falls back to an LLM when confidence is low, with a circuit breaker and evaluation suite.
sirgio03/laya-ignite
Bootstraps specialized Laya decision heads from plain-language criteria: an LLM generates training examples once, then heads are trained on the frozen encoder and calibrated.
DataZooDE/anofox-decide
DuckDB extension that answers natural-language yes/no and choice questions from SQL, running Laya or Julia-1 fully offline or calling TypeSafe Jev and other System One APIs.
hakan-gecili/pacman-arena
Laya, tev1 and Nimble play Pac-Man locally with the same information over 30 seeds: Laya decides in 32 ms, by far the fastest, but scores lowest.
BarnacleLabs/cyprid
Installs, starts and stops open decision models on a Mac, Laya among them, behind one Jev-style API and web page, with a side-by-side accuracy table on the Kev suite and JevBench.
apscot/laya-java
Unofficial pure-Java port that runs the three Laya checkpoints inside the JVM with no Python; published on Maven Central with a Spring Boot starter that serves /v1/systemone and /v1/decisions, plus a client and a Gradle plugin.
elepedus/laya-rk3588-turingpi-rk1
Rust backends that run Laya on the RK3588 NPU or Mali GPU of a Turing Pi RK1 behind the Jev-compatible API, with measured latency against the CPU server.
satiricalguru/Civic-Signal-Laya
Review workbench for public-service requests in many languages: Laya picks the service team, urgency and access barrier, shows the full probabilities and flags low-confidence cases for a human.
larchanka/snake-game-laya-ai-example
Terminal games suite powered by Laya for real-time decision-making, starting with an autonomous Snake fully controlled by the model.
MindrLabs/typed-decision-slay-the-spire
Three Laya checkpoints and Nimble-9B play the same 200 Slay the Spire seeds: Nimble climbs higher zero-shot, but after one short fine-tune Laya English gains 6.7 floors and answers in 24 ms with about 3 GB of memory.
tnaftali/s1-tui
Terminal UI for typed decisions: type a state and a question, get the answer with its confidence and full probabilities, switching live between local Laya on Apple Silicon and the hosted Jev API.
trinacria-tech/laya-pi
Docker stack that runs Laya on a Raspberry Pi 4 (8 GB), CPU only: an ONNX Runtime /v1/systemone API plus live demos.
msudars/laya-reader
Daily digest of new arXiv papers ranked by Laya against your reading profile; runs locally on CPU and tracks how good its picks are.
cloudwallker/chong-laya
Chinese shopping-decision lab: describe a purchase and inspect the buy / wait / skip choice Laya makes, with probabilities and raw responses. A learning project, not shopping advice.
StevenLi-phoenix/laya-skill
Claude Code skill and plugin for typed decisions: run Laya locally, in the browser on WebGPU or as a Jev-compatible server, switch to TypeSafe Jev by changing the base URL, and fine-tune, compare and export Laya to ONNX.
rupeshs/laya-ov-rs
Unofficial Rust port of the laya-openvino CPU backend: loads the INT8 English or typed-decisions checkpoint (about 400 MB each) and returns the same JSON as the Python package.
andrewgoodchild/doomANE
Laya plays Freedoom in real time with every forward pass on the Apple Neural Engine (about 110–140 ms per decision); it clears the maps it was developed on but none of six unseen maps, and the README reports both.
blacksoxx/system-one-aiops-demo
Reproducible alert-triage demo: a fake shop on a kind cluster sends every Alertmanager alert to laya-serve (typed-decisions), and the typed answers sort alerts into page on-call, Slack or closed as noise.
edwardmonteiro/laya-android
Asteroids-style dogfight for Android where multilingual Laya flies the enemy ship entirely on the phone, with live probabilities, latency on the HUD and CI evaluation tests.
suradisworkspace/laya-router
Per-turn model routing for Claude Code: a local Laya decides whether each new turn goes to the fast or the strong model, with the reasons shown in the status line.
Badge for maintainers
Is your project listed here? Add this badge to your README and readers land straight on your project card. It is entirely optional and does not affect inclusion.
[](https://laya-ai.com/projects#project-nandhakishorm-laya)Inclusion policy
Projects need to provide runnable inference, a meaningful runtime or port, a real integration, a reproducible benchmark, a working demo, or tooling that materially helps Laya developers. A repository is not listed merely because its name contains Laya.
Submit a project