Acurast, the decentralized computing network built from upcycled smartphones, announced that it can now run Laya, an open-source, decision-focused AI model, securely and at scale across its distributed infrastructure. The deployment marks a significant step toward making autonomous, real-time decision-making AI accessible outside proprietary ecosystems, offering an alternative to closed commercial models.

Developed by Convai Innovations and released under the Apache-2.0 license, a permissive open-source license that permits commercial use, modification, and redistribution, Laya operates as a "System 1" decision model — a name that echoes the dual-process framework from cognitive science, which contrasts fast, automatic responses with slow, deliberate reasoning. Instead of generating text responses, the model receives a problem space, evaluates the current state, and selects the next action from the available options in real time. This approach moves AI applications beyond conversational output toward continuous, machine-speed decision execution.

"Today's AI conversation is constantly dominated by ever-larger models generating ever-more text, but not end results. Proprietary models like Jev show the immense demand for rapid decision-making, but they force developers into closed ecosystems," said Alessandro De Carli, Founder of Acurast, in a written statement. "By running Laya on Acurast, we've changed that. We are proving that System 1 AI can run securely and at scale on hardware everyone already owns, providing a truly open, decentralized alternative to centralized cloud lock-in," he added.

Live demonstrations of the deployment show Laya instances handling rapid decision tasks directly on decentralized mobile edge nodes. Examples include real-time gameplay in Snake and Tetris, autonomous navigation and combat decisions in the original Doom engine, classification of incoming mail as inbox, spam, or phishing, and evaluation of headlines from real RSS feeds as news, satire, clickbait, or manipulation. Security functions are on display as well, including detecting prompt-injection attempts and flagging toxic messages in live chat. External users can also submit their own inputs, including strategies, headlines, or adversarial prompts, and observe the model's decisions in real time.

Where traditional AI workloads typically depend on large, costly centralized server infrastructure, Acurast takes a different approach by leveraging the processing capacity of everyday mobile devices. For decision-class workloads, that distinction is what makes the setup viable: because Laya's output is a single selected action rather than long-form generated text, the computation involved is light enough to run on standard mobile processors rather than specialized accelerator hardware.

Decentralized, Confidential Compute at the Network Edge

Under this model, every Laya decision is executed on an Android smartphone functioning as a secure node within the Acurast network. The platform uses the Trusted Execution Environments built into modern devices — hardware-isolated secure areas of the processor designed to protect code and data from the rest of the system — to keep workloads confidential and verifiable. Decisions are generally delivered in approximately 0.2 to 1 seconds using CPU only. Workloads are distributed dynamically across the global network, results include cryptographic proof, and participating processors are compensated in ACU, the network's native token.

"Our goal is to open developers' eyes to the quite incredible opportunities that come from using Acurast compute. This deployment proves that decision-oriented AI can run cheaply, verifiably, and without a data center in sight on a decentralized smartphone network today," De Carli said. "As autonomous agents evolve, we want developers to realize that the infrastructure they run on can—and should—be as open and distributed as the software itself," he added.

The infrastructure is currently live and does not require developers to obtain API access or server allocation from a centralized provider. It is designed to highlight the economic advantages of running small, continuously operating System 1 workloads on a permissionless network. Live instances of Laya can be tested at laya.acurast.com, where users can challenge the model with custom inputs and deploy their own instances directly on Acurast edge devices. Because the live instances accept outside submissions, including adversarial prompts, the model's security and decision behavior can be examined by any visitor in real time rather than judged solely from the team's own examples.