2,170 public GitHub projects · 8 application domains · 1,865 new repositories in one week
What are developers actually building with Jev? Our paper, Jev in the Wild: A Data-Driven Analysis of the Jev Model's Functionality, Applications and Ecosystem (arXiv:2609.30216), studies 2,170 verified public GitHub projects. To our knowledge, it is the first data-driven survey and analysis of Jev's application ecosystem.
The short answer: Jev grew rapidly in its first week, spread across many kinds of software, and took on different decision roles in different workflows. GitHub attention grew quickly too, but it was concentrated in a few application areas.
Jev in 20 seconds
TypeSafe AI calls Jev a System One model: a decision model for frequent, small judgments. Its three interfaces answer simple questions:
- Choice: Which option should the software select?
- Noul: Is this statement or condition true?
- Score: How strongly does an input match a criterion?
The application supplies context and acts on Jev's answer. That is why the same model can select an AI agent's next action, filter search results, or route a request to a model or tool.
Six findings from the first Jev application ecosystem survey
1. This is a survey and analysis of real project uses. We included a repository only when its public code or documentation showed Jev used for a concrete task. A passing mention was not enough. The dataset is a snapshot from September 22, 2026.
2. Jev showed rapid early growth. In the week after its September 15 release, the sample reached 1,865 newly created repositories; another 305 older repositories had integrated Jev. New projects and integrations into existing software grew side by side. The new repositories had accumulated 43,750 GitHub Stars by the snapshot, showing substantial early public attention.
Figure 1. First-week growth of newly created Jev repositories. The 305 older repositories are included in the 2,170-project study but not in this project curve.
3. Jev applications span many domains. The eight major categories include interface agents, software engineering, search and memory, safety and governance, routing and automation, simulation and control, content and expert tasks, and infrastructure. Even the largest category holds less than 18% of the projects.
4. Its job changes with the application. In an interface agent, Jev may select an action. In search and memory, it may judge or filter content. In routing and automation, it may choose a model or tool. The surrounding workflow gives Jev its practical role.
Figure 4. Jev's decision purposes differ across domains. Each bar shows the share of observed purpose labels within its category.
5. Developers often combine Jev's interfaces. 61.8% of projects use at least two of Choice, Noul, and Score; 36.8% use all three. A code-review workflow, for example, can judge whether a change needs attention, choose supporting evidence, and score its priority.
Figure 3. Exact combinations of Choice, Noul, and Score in the 2,170-project sample.
6. GitHub attention is concentrated. Routing and Automation plus Interface Agents account for 19.6% of projects but 63.0% of all repository Stars in the sample. Project activity covers many domains, while public attention clusters around agents and routing. Stars measure attention to whole repositories, not how often Jev is used.
Why this matters
Taken together, the findings show an emerging software ecosystem for a reusable decision model. Jev's application, decision purpose, and output interface depend on the workflow around it. Evaluating such a model therefore calls for several kinds of real tasks, including agent actions, search filtering, code triage, and model routing.
The full paper explains the collection method, the eight-domain taxonomy, and seven representative applications. These findings describe public GitHub projects at one point in time; they do not count private or commercial deployments or establish long-term adoption.
Jev in the Wild is by Guoming Ling, Muen Xue, and Zijian Ye.
AI disclosure: This article was drafted with AI assistance. The research claims and figures were checked against the linked paper.
Top comments (0)