Image Credit: Shutterstock
- Acurast deployed the open-source AI decision model Laya across its decentralized smartphone compute network for real-time Android processing.
- The model runs on smartphone CPUs without dedicated GPUs and typically returns decisions in about 0.2 to 1 second.
- Developers can now deploy Laya through Acurast's production compute network for low-latency decision workloads across distributed hardware.
Acurast has deployed open-source AI decision model Laya across its decentralized smartphone compute network, allowing applications to process real-time decisions directly on Android devices.
The model runs on smartphone CPUs and typically returns decisions in about 0.2 to 1 second, according to Acurast. Its network includes more than 280,000 smartphones across over 175 countries.
Laya Returns Decisions in 0.2 to 1 Second
Laya was developed by Convai Innovations as a non-autoregressive “System 1” model. Unlike large language models that generate responses one token at a time, Laya evaluates a defined state and produces a structured decision in a single forward pass.
The model can select between predefined options, assign scores or return probabilities for yes-or-no decisions. Its code and model weights are available under an Apache 2.0 license.
Acurast said potential uses include content moderation, security screening, email classification, navigation and automated routing, where applications require fast structured outputs rather than conversational responses.
Android Phones Run Laya Without Dedicated GPUs
Acurast runs Laya on Android smartphones connected to its distributed compute network without requiring dedicated GPUs. Decisions are processed through the devices’ CPUs.
The company said workloads run inside hardware-backed Trusted Execution Environments designed to isolate computation from the device owner and operating system while supporting verification of completed workloads.
The deployment extends Acurast’s use of consumer smartphones for distributed AI processing rather than centralized data-center infrastructure.
Snake, Tetris, and Doom Demonstrate Low-Latency Decisions
Acurast has also deployed demonstrations showing Laya making repeated decisions in games including Snake, Tetris, and Doom.
In Tetris, the model selects where pieces should be placed, while the Doom demonstration involves repeated decisions around movement, aiming and navigation.
Developers can now deploy Laya through Acurast’s production compute network, making the model available for low-latency decision workloads across distributed smartphone hardware.
