What Is Jev (and Why I Started Using It)

Hey, everyone!

Anyone who has put an LLM in production knows this ritual:

1. Write a huge prompt.
2. Ask the model to “respond ONLY with JSON”.
3. Pray it doesn’t invent a field.
4. Parse it, handle the error, and hope the result is the same next time.

It works, but it’s using a cannon to kill a mosquito. Most of the time I don’t want the AI to write anything. I want it to decide one small thing.

Jev is a model from TypeSafe built exactly for that. It doesn’t generate text. You send a context and a question with the possible answers, and it gives you back a typed answer with a probability.

Think of it like this

A traditional LLM is like a friend who talks too much. You ask “is the recipe in the caption?” and you get a paragraph back.

Jev is like a straight-to-the-point friend: “Yes, 97% sure.” Done. Then your code decides what to do with that.

The three kinds of question

With Jev you only ask three kinds of question:

• Noul (yes or no): “Does this caption have enough ingredients to build a shopping list?” → 0.97.
• Choice (pick one option): “Where is the recipe?” → comments, with 100% confidence.
• Score (a rating on a scale): “How urgent is this message?” → a position on a scale you define yourself.

To build a question you send three things:

1. The state, meaning the context: the caption, the message, whatever it is.
2. The instruction, which is the question itself.
3. The criteria, which say what each answer means.

Real example: claude_notify

In my Telegram bot that controls Claude Code, when it asks for permission to run a command, it used to only understand y, yes or 1. If I replied “sure, go ahead”, the bot didn’t get it and typed the whole text into the terminal.

Now Jev interprets the reply:

• “sure, go ahead” → approve (1.0). The bot presses Yes.
• “ok, but don’t touch the migrations” → approve with instruction. The bot does NOT press the button, because it would lose the condition I set.
• “why does it need rm -rf?” → question. It sends the text to Claude.

And notice I never configured Portuguese. It just understood.

Real example: MiseSnag

In MiseSnag I used regex to decide whether a video’s caption already had the recipe. If it does, I skip the download and the transcription, which is where the cost is. I tested 6 captions:

• The regex got 2 wrong. It didn’t recognize an ingredient list without quantities, and it thought “Ingredients: see video!” was a recipe.
• Jev got all 6 right. And as a bonus it told me where the recipe was: in the comments, on screen, or spoken.

Every regex mistake costs money: a download, a Whisper run, an LLM call for nothing.

What I like the most

1. The code stays in control. Jev gives its opinion and the code decides. I’m the one who sets “only press the button if you’re 90% sure”. Below that, I fall back to the old behavior, which is safe.
2. It’s fast. Around 0.9s per call, and you can send several questions in the same request.
3. No fragile JSON parsing. The answer already comes in the right shape.
4. Probability becomes a business rule. You can tune the confidence thresholds without touching a single prompt.

Where NOT to use it

If you need to generate text (write a recipe, translate, summarize), that’s an LLM job. Jev is for judgment: classifying, routing, verifying, choosing between options.

The rule I’m following:

LLM to write. Jev to decide. Code to execute.

Fewer giant prompts, less broken parsing, and more predictable decisions.

That’s it, folks! If you try Jev in one of your projects, tell me how it went on Twitter or LinkedIn.