{"slug": "jev-computer-use", "title": "Jev & computer-use", "summary": "A developer has released jev-computer-use, an open-source project that applies typesafe.ai's Jev system-one model to browser and computer-use automation. The project treats agent logic as state machines, feeding structured state and goals to Jev's typed decision interface rather than a text-generation interface, which the developer argues suits local, low-reasoning decisions like form handling and OS accessibility. The developer notes Jev is trained on 100% synthetic data, runs non-deterministically, and that its confidence score is a derived metric tightly coupled to classification scores.", "body_md": "Current work: [jev-computer-use](https://github.com/sarang-pratham/jev-computer-use)\n\nMy TL for the last few days has been taken over by [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev), a system one model from [typesafe.ai](https://typesafe.ai). So i applied for the waitlist and got in.\n\nTo me a interesting use-case for this kind of model is on browser/computer-use, lot of decision to be taken which may not need long reasoning before doing so, the UX improves with the speed we get and automations feel right.\n\nCheckout the [jev-ultrafast](https://github.com/browser-use/jev-ultrafast) by browser-use.\n\nTake a general purpose model (GPT / Claude), the lifecycle of a understand user request looks like:\n\nThis is useful when the task isn't fully known in advance, involves long reasoning and time. But think of structured inputs like forms or OS accessibility, the decisions of what needs be done given a goal is more local than you think.\n\nThis could be very simply put as a classifier. The model accepts a state and questions returns a typed decision. A state is your structured text input and questions have three primitives:\n\nGeneral purpose LLMs are also capable of it but the main difference here is - jev exposes a decision interface rather text generation interface, give the context and decision to make, received a typed result.\n\nThink of making your agent logic as state machines, give that state, goal and the choices it can trigger, this works really well.\n\nThere are multiple ways to do this:\n\nI'm currently experimenting on this: [jev-computer-use](https://github.com/sarang-pratham/jev-computer-use)\n\nA example computer-use flow:\n\nAlthough 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.\n\nThere is one more strong claim on almost no hallucination, being type safe does not guarantee that.\n\nIf you think this run deterministically, NO, same prompt gives you different probabilities every run.\n\nConfidence 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 and you are are open to come up with any definition for the confidence using the probabilities list.", "url": "https://wpnews.pro/news/jev-computer-use", "canonical_source": "https://dev.to/prathamxn/jev-computer-use-51mk", "published_at": "2026-09-21 18:18:22+00:00", "updated_at": "2026-09-21 18:31:21.407460+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "ai-products"], "entities": ["Jev", "typesafe.ai", "browser-use", "jev-computer-use", "jev-ultrafast", "GPT", "Claude"], "alternates": {"html": "https://wpnews.pro/news/jev-computer-use", "markdown": "https://wpnews.pro/news/jev-computer-use.md", "text": "https://wpnews.pro/news/jev-computer-use.txt", "jsonld": "https://wpnews.pro/news/jev-computer-use.jsonld"}}