Jev & computer-use

Current Work: jev-computer-use

My timeline for the last few days has been taken over by Jev, a system-one model from typesafe.ai. I applied for the waitlist and got in.

To me, an interesting use-case for this kind of model is browser/computer-use. There are lots of decisions to be taken which may not need long reasoning before doing so; the UX improves with the speed we get and automations feel right. Check out jev-ultrafast by browser-use.

General-Purpose Model Lifecycle

Take a general-purpose model (GPT / Claude). The lifecycle of understanding a user request looks like:

1. Plan
2. Generate code / text
3. Produce tool calls
4. Something broke / unexpected? → give the agent new state, ask it to redo stuff

This is useful when the task isn't fully known in advance, involves long reasoning, and takes time. But think of structured inputs like forms or OS accessibility: the decisions of what needs to be done given a goal are more local than you think.

What Is Jev?

This could be very simply put as a classifier. The model accepts a state and questions, and returns a typed decision.

• State: your structured text input
• Questions have three primitives:
  • Choice: picks one of the options you provide + confidence
  • Score: rates something you provide + confidence
  • Noul: yes / no with a probability

General-purpose LLMs are also capable of this, but the main difference is that Jev exposes a decision interface rather than a text-generation interface. You give the context and the decision to make, and receive a typed result. Think of making your agent logic as state machines: give that state, goal, and the choices it can trigger-this works really well.

On Computer-Use

There are multiple ways to do this:

• Screenshot approach: use a model for segmenting and OCR, then convert to structured format and pass it to JEV for decision
• Accessibility APIs: the OS itself provides accessibility APIs to navigate; it's already structured-use it directly

I'm currently experimenting on this: jev-computer-use.

Claims and Things to Look Out For

Although the model is refreshing in the age of LLMs, I think the magic lies in the data. They do claim they are a data research company first, and it's been trained on 100% synthetic data.

There is one more strong claim: almost no hallucination. Being type-safe does not guarantee that.

If you think this runs deterministically: NO. The same prompt gives you different probabilities every run. Confidence score is a derived metric which is tightly coupled with the classification scores, so it may not be the right measure for whatever you are doing. You are open to come up with any definition for the confidence using the probabilities list.