This Free AI Runs on Your Old Laptop and Fact-Checks in a Blink

A solo developer has released a free AI model called Laya, after an earlier post about his homemade "Jev" design went viral and dozens of people asked him to build a general version anyone could test. Laya is small — 421 million parameters, a fraction of what powers ChatGPT — trained on a single professional graphics card. That smallness is the point: it can run on a low-end PC, and there's a free public demo you can try in a browser right now.

So what does it actually do? It's a decision-maker, not a chatbot. You give it a question and a set of possible answers, and it scores each one. That makes it useful for everyday guardrail jobs: routing a customer request to the right department, checking whether a claim is true, flagging toxic comments, catching sneaky prompts designed to trick an AI, or grading written answers against a rubric. It does all of this in about 35 milliseconds — faster than you can blink.

Why should you care? Normally this kind of work needs big, expensive cloud models, which means monthly bills and your data leaving your machine. A model this small can sit on your own laptop, offline, for free. Notably, its training data is entirely human-annotated — no AI-generated filler — which its creator says makes it more trustworthy. And it's built to give honest probabilities: when it says 70% confident, it means 70%.

The catch is real, though. Laya doesn't write or explain — it only chooses among options you give it, so it can't replace a chatbot. Its 25,000 examples are modest by industry standards, its benchmarks were set by the creator rather than independent testers, and he openly says it "will need more refinement." This is an impressive hobbyist project, not a supported product.