The Laya project directory

Everything the world is building on Laya.

Laya turns any text into typed decisions in milliseconds, on your own machine. Here is what it does, and 2447 projects people built with it.

Input text

“Hi, we were billed twice for March. Please refund the duplicate today or we will cancel our plan.”

laya-mlx · 146 input tokens · 0 output tokens

Which team should handle this ticket?

choice
✓ billing98%
technical1%
sales1%

How urgent is this?

score
not urgent4%
soon15%
✓ blocking81%

Does the customer threaten to cancel or leave?

yes / no
✓ yes84%
no16%

Decided in 30.7 ms · 3 typed answers · one forward pass

Real outputs from Laya, not a mock-up. Run it yourself →

Projects
2447
New this week
45
Platforms
18

Seen on X this month

All demos →

Acurast’s Laya Tetris demo lets an Android phone choose each move

Where would you drop the next Tetris piece? An Android phone gets to make that call in @Acurast Laya demo. The game works out where the piece could land. Then an AI model running on the phone picks one of those spots. When the piece drops, you can see exactly what the developer needed compute for. One piece, one decision. Here’s how I’d try it. Open Tetris, pick a landing spot in your head and watch where the model puts the piece. Then look at the device info beside the game. That’s the phone that handled the decision. I like this example because “contribute compute” sounds vague when you’re the one being asked to plug in a phone. Watching a model bury the gap you wanted to keep open? Now you know exactly which move to blame it for 😄 Try Laya here 👇 If you’d like to put a phone on the other side of a request like that, here’s my Cloud Rebellion invite 👇

♥ 556↻ 134.4k viewsView on X

Laya navigates Snake one model decision at a time

Snake became surprisingly interesting to watch once every move depended on a model response. In @Acurast’s Laya demo, step mode advances the snake one cell after each decision. The interface shows its position, the food location and the available direction choices alongside the model’s response. Watching the snake collect food and continue navigating made the sequence easy to follow. Each choice changes the situation the model has to deal with next. That’s the aspect I find useful: a decision that looks reasonable now can affect the options available several moves later. It also makes response time visible in the experience. When the game waits for an answer, you can directly see how the inference loop affects its pace. I wouldn’t judge the whole model from one short game. But this is an accessible way to explore repeated decisions on phone hardware, with enough feedback to understand what is happening. Choose Snake: Join through my invite:

♥ 468↻ 112.1k viewsView on X

Laya Chat Moderator classifies scams, spam and toxic messages

The simulated giveaway message in Laya Chat Moderator caught my attention: send 50 ACU to receive 1,000 back. The model flagged it as a scam. In the same short run, repetitive chat was labelled spam, ordinary game comments were allowed and an insulting message received a toxic label. What I liked about @Acurast’s demo was watching several kinds of decisions arrive through one queue. Each message has a result, making it possible to inspect the classification while the conversation continues. The run matched the script’s labels on eight messages. That’s a small synthetic example, so I’d be careful about extending the result to a busy real community. Still, it makes the workload concrete: messages arrive, a model running on phone hardware classifies them, and the interface displays the decisions. For me, it’s a useful starting point for thinking about response time, ambiguous language and where human review would still be valuable. Choose Live Chat Moderator: My referral:

♥ 463↻ 52.3k viewsView on X

Laya Mail Sorter classifies sample emails on Android

The mistakes in @Acurast’s Laya Mail Sorter were the most useful part of the demo for me. The model sorted ten example emails into inbox, spam or phishing. In this run, it matched eight of the ten expected labels. A fake tax-refund message and a parcel customs-fee message slipped into the inbox. Seeing those errors alongside the successful classifications made the experience more informative. There is a specific task to evaluate, and the interface lets you inspect where the model’s judgment differs from the examples. Laya runs on an Android phone in the Acurast network, so this is also a practical illustration of a small classification workload on distributed hardware. My takeaway is that deployment and task quality deserve separate attention. A model can run and return decisions while still making consequential mistakes. I’d use this demo to explore those boundaries, especially with differently worded examples, before treating any classifier as a security filter. Choose Mail Sorter: My referral:

♥ 462↻ 62.4k viewsView on X

Acurast uses Laya Tetris to demonstrate smartphone-powered decentralized compute

The future of AI may need more computing resources than traditional infrastructure alone can provide. This is where the combination of AI and DePIN becomes interesting. @Acurast is building a decentralized cloud powered by smartphones, allowing everyday devices to contribute computing resources to applications running across its network. There are two ways to participate. Processor Lite supports compatible Android phones and iPhones. It lets users contribute during idle periods, such as overnight while charging, without factory resetting their devices. Processor Core turns a compatible Android phone into a dedicated Processor. It requires a factory reset, making it more suitable for spare devices. What interests me is the bigger picture. AI applications need computing power, while DePIN offers a different way to distribute infrastructure resources. Acurast connects these two ideas by turning smartphones into potential compute providers. Its Laya Tetris demo, where an AI model running on an Android phone chooses game moves, is a practical example of this concept. I believe @Acurast is worth watching as AI and decentralized infrastructure continue to develop.

♥ 447↻ 26.3k viewsView on X

Laya Tetris displays the Android phone making each move

A Tetris piece was falling on my screen, and I had no idea which device was making the decisions behind the game. I tried the Laya demo to get a better look at what Acurast compute actually does. The game creates possible placements for each piece, then an AI model running on an Android phone chooses where it should go. With @Acurast , the interesting part is that the decision was being processed inside the Trusted Execution Environment (TEE) of a regular phone that had already been onboarded as a Processor. The move was selected in real time, and the phone’s details appeared right beside it. No traditional server farm or data centre was needed for that particular workload. An ordinary Android device was handling the computation. This also gives a clearer picture of what happens after someone connects their phone to the network. Who is actually using the computing power being provided? In this demo, it was a simple game, but the same network can handle other types of work. Developers can deploy serverless workloads, LLM inference and agent tasks, which are then matched with available phones. The device processes the assigned job inside its TEE and reports the result. The processor can earn for the time it stays available and contributes computing power. What I liked about the demo is that you don’t need to understand model weights or dig through the game’s code to see the process. A move gets made, and the device responsible for the computation is displayed right there. It makes the idea of putting an idle phone to work much easier to understand. That spare Android can become part of a network providing confidential compute to developers who don’t want to build and maintain their own infrastructure. Sometimes seeing a real workload run on a phone explains the idea better than reading another technical breakdown.

♥ 445↻ 55.8k viewsView on X

colibri

JustVugg

Runs large open models locally in pure C, including Laya for typed decisions

40.1k·GitHub repo

laya

NandhaKishorM

Runs multilingual typed-decision models and provides routing, serving, and integration options

23.9k·GitHub repo

laya-mlx

mizorewww

Runs Laya typed-decision models locally on Apple Silicon using MLX

6.3k·GitHub repo
C

laya

convaiinnovations

A multilingual-ready System 1 model that returns calibrated typed decisions over text in one pass

16.3k·HF model

laya-coreml

mizorewww

Runs Laya typed-decision models locally on Apple Silicon using Core ML and the Neural Engine

1.4k·GitHub repo

laya

receptron

Runs Laya typed-decision models from Node.js and TypeScript through ONNX Runtime

450·GitHub repo
C

laya-multilingual

convaiinnovations

A multilingual System 1 model that returns calibrated typed decisions over text in one pass

8.5k·HF model

system-one-connector

itsmostafa

Connects AI agents to hosted and locally run System One decision models through an MCP server

365·GitHub repo

typesafe-mcp

itsmostafa

Connects AI agents to typed-decision models through an MCP server

303·GitHub repo
C

laya-demo

convaiinnovations

A Gradio demo for Laya typed decisions, multilingual routing, moderation, and RAG filtering

288·HF Space

openJev-verdict-2.0

Heman10x-NGU

Develops and benchmarks a non-autoregressive typed-decision model against Jev and Laya

286·GitHub repo

quackd

rokbenko

A Python CLI coordinates robots using LLMs, vision-language-action models, and decision models

244·GitHub repo

laya-ultrafast

ipenywis

Runs a browser automation agent using local Laya decisions through MLX

181·GitHub repo

mu

qybaihe

Provides a coding agent with a judgment kernel that can use Laya as its local judge

179·GitHub repo
C

laya-typed-decisions

convaiinnovations

A Laya model fine-tuned for agent traces, customer service, invoices, and security incidents

3k·HF model

FluidUse

FluidInference

Automates form filling on Apple devices with on-device models, including Laya through Core ML

160·GitHub repo

jev-cookbook

datawhalechina

Teaches Jev and demonstrates fine-tuning and serving the open-source Laya model

158·GitHub repo

laya-playground

wdobry

A local website combines Laya model demos, games, a benchmark, and an agent skill

157·GitHub repo
A

laya-mlx

aac6fef

An MLX FP16 conversion of Laya for native inference on Apple silicon

122·HF model

keel

codejunkie99

A macOS coding workspace uses local Laya or hosted Jev to select task routes

122·GitHub repo

laya-vs-jev

virajbhartiya

A side-by-side T-Rex game compares local Laya and hosted Jev decision models

100·GitHub repo

laya.cpp

lkarlslund

A native C++ runtime runs Laya typed-decision models across GPU and Apple backends

89·GitHub repo

jevals

openlayer-ai

Agent evaluation and guardrail tools run with local Laya, Kev, or hosted Jev

86·GitHub repo

system-one-foundation-models

peterfriese

Bridges Laya and Jev System One decision models into Apple’s Foundation Models framework

72·GitHub repo