AI
Jev looks like fuzzy logic at first glance. It isn't. It's a calibrated model, now callable from Laravel's AI SDK, and here's how to use one.
On this page #
When I first read about Jev, on Freek’s blog and on X, I thought: fuzzy logic. I built an app with fuzzy logic more than ten years ago, for a client whose sites kept getting hacked. It fingerprinted each page, checked it regularly, and when the live page drifted from that fingerprint, analysed the new content and assigned a probability that it was malicious. Above a threshold, it fired an alert: email, SMS, and a dashboard that started blinking red for that site.
Jev isn’t fuzzy logic. It’s a different idea, and worth ten minutes before you wire it into a Laravel app.
What is Jev #
Fuzzy logic lets you reason in degrees instead of true or false. You write the rules and the weights by hand: “if traffic is unusual and content changed overnight, hacked is likely.” Every number in that rule is your own guess, dressed up as logic.
Jev is a trained model instead: you send it a state and a typed question, and it hands back a probability, skipping the rules and the weight-tuning entirely.
The word that matters is calibrated. Of everything Jev calls 80% likely, about 80 in 100 should actually be true, over enough cases. That’s a promise worth checking against your own data before you trust a threshold to it. Fuzzy logic never promised that. A rule scoring something “0.8 hacked” is still just a guess, dressed up as a number.
Why Jev #
A calibrated number is one you can act on. You pick the cutoff based on what a wrong answer costs you: 0.9 for something expensive to get wrong, 0.5 for something a human reviews anyway. The model hands you the number; where you draw the line is still up to you.
TypeSafe AI released Jev on 15 September. Two days later, Laravel’s AI SDK added support for it, merged into the 1.x branch as a Classification API.
Example #
Say you’re triaging support tickets: is this one urgent, and which team should get it?
use Laravel\Ai\Classification;
use Laravel\Ai\Classification\Boolean;
use Laravel\Ai\Classification\Choice;
$result = Classification::of($ticket->body)
->question('urgent', new Boolean('Does this need an immediate response?'))
->question('team', new Choice('Which team should handle it?', [
'billing' => 'Payments and refunds',
'technical' => 'Bugs and outages',
]))
->classify();
$result['urgent']->isTrue(threshold: 0.8); // true
$result['team']->choice; // 'technical'
isTrue() takes a threshold argument instead of assuming 0.5, which is where the decision above actually lands in code.
Testing doesn’t need a live call. Fake the gateway and assert on what you asked:
Classification::fake();
Classification::of($ticket->body)
->question('urgent', new Boolean('Does this need an immediate response?'))
->classify();
Classification::assertClassified(fn ($prompt) => $prompt->asks('urgent'));
What’s next #
Jev itself is still early: no tagged Laravel AI release yet, and the model is behind a waitlist. The Classification API works today against the fake gateway, so the integration can be built and tested before a key ever arrives.
Read more #
- Introducing System One Models & Jev , TypeSafe AI’s own announcement: latency, pricing, and how the published benchmark numbers were produced
- Detecting spam and auto-replies with Jev and the Laravel AI SDK , Freek Van der Herten wiring Jev into a real Laravel app
- Jev: The Language Model That Won’t Talk , Anthony Maio’s closer look at the benchmark, including where Jev falls behind