Laya: FREE Alternative to Jev AI

Julian Goldie (0:00)

Laya is a free Jev AI alternative. It does the same job as Jev. Does it last seven times faster, according to some of the tests. And on most tests, it gets most answers right. And you can run it yourself for free. So let me explain what that means, because most people have never heard of Jev, and even fewer have heard of Laya. So you want to think about your AI agent for a second. Maybe it's Claude Code, maybe it's Hermes, maybe it's OpenClaude, whatever you use day to day. It's making little choices all day long. So email comes in, for example, is it a customer or is it junk? A message comes in. Is this person happy? Or do we need to do something about it? A task comes in. Is this urgent or can it wait till tomorrow? Those are tiny yes or no questions. So you can pick one from a list and rate it one to three. It's pretty simple stuff. But most people send those questions to a giant AI model. And the same model that writes their blog posts and builds their websites is usually the same model that's making decisions, which is not that efficient. So it's kind of like asking a surgeon to sort your mail. It works, but it's slow. It costs tokens every single time. And the big model writes your whole sentence back. So your agent then has to read that sentence and figure out what it meant. And that's the problem Jev was built to fix. So Jev is a tool from a company called TypeSafe. You can give it an email message or a ticket. You can ask it a simple question. It gives you back a clean answer.

It's very fast.

Coming back with answers, making decisions, and it can automate things very quickly. So for example, if you're sorting through loads of leads, you can give it to Jev AI before it goes to AI agents. And every question takes about a core of second, which basically makes everything way faster. Now, Laya actually does the same job as Jev. So this is Laya. It's a free open source project, as you can see right here. Came out a few days ago from a small team called Convye Innovations, and it's open source, which means anyone could use it, change it, and run it on their own machine.

Three days already has like 5,000 stars on GitHub, which is pretty cool.

I would check it yourself before you install it, but we're going to cover how it works today and how it works, right? So basically, you can give it text, an email, message, a support ticket, anything. You can ask it three kinds of questions. Pick one from a list, give it a score on a scale, or answer yes or no of a probability. Then read the text once and answers all your questions at the same time. Now, one question, according to the tests, takes about 33 milliseconds. That's 300th of a second. If you send it 10 questions at once, it drops to about 7 milliseconds per question. Now, Jev in independent tests takes 236 to 276 milliseconds per question. So Laya is roughly seven times faster. Speed is nice, but does it get the answers right? Well, let me run you through the tests. So there's a test called Typed Decisions.

It's 2,000 choices across four kinds of business work, sorting invoices, handling customer service, security staff, and watching what agents are doing. Jev actually scored 0.727 on that test. Laya scored 0.766. So Laya got about four more out of every 100 right. There's a news sorting test with four categories. Jev got 0.910, Laya got 0.95.

And there's an emotional test with six feelings. Jev got 0.48, and Laya got 0.595. It's quite a big gap when you look at them like that so far. And on the sureness number, which is the researchers' core calibration, Laya is about three times more honest than Jev after tuning. I mean, when Laya says 90%, he's right about 90% of the time. That's the number your agent depends on. So this one matters a lot. Now, one thing to be fair about, the Laya team measured their own numbers here on their own hardware. So the Jev numbers come from other people who publish them because the Laya team didn't have access to Jev. So it's not like a perfect side by side. The team says this in their own write up as well. And this one test where Jev wins clearly. This is actually a test, as you can see right here, with 77 different categories. Jev actually scored 0.87, Laya scored 0.425, less than half. Why? Because Laya has a limited amount of room for the list of choices you give it. So when you hand it 77 options, each one only gets three or four words of room. They start to like kind of blur together, and Jev can handle up to 255 options with no problem. So if your question has more than 20 possible answers, Jev is still the better pick today. You can adjust Laya's settings or split your big list into a small list first and a small list second, but that's extra work. Jev doesn't need. So it's a pretty good option, pretty easy to set up. And basically, if you want to use this, you can just get the GitHub light you see right here, and then you can give this to Claude or to any other AI agent like you can see. Now, if you want to indicate like what the use case is, or how does Jev work, or what does Jev actually do? So Jev is just kind of like a quick way of speeding up decisions. It can't write. It just makes decisions very quickly. So you can give it details on the decisions you're going to make and the possible answers, and then it can sort through things and filter things very, very quickly. Let me show you an example in practice.