Awesome Jev usecases, trending daily.
Jev is TypeSafe's System One decision model: typed choices, scores, and calibrated probabilities in 70 to 500 milliseconds, with output tokens free. The ecosystem index passed a thousand projects in under two weeks, and the stars keep compounding. This wiki shows you the ones worth your time, what is moving, and what the independent tests actually found.
Trending this week movers, 24 to 30 Sep 2026
The top 60 repositories alone gained 11,097 stars in six days, a 16% jump. Here is where it went, and the week the press caught up.
jev-ultrafast
★ 21.4kThe browser agent that books a Zurich to London flight in 7.1 seconds for $0.0039. Jev picks each operation and DOM element in one request.
jev-chat-jarvis
★ 7.1kMobile chat copilot for WeChat, QQ, X, and Feishu. Reads the conversation, proposes replies, fills the box. You decide to send. Non-invasive, screen only.
fast-jev-compaction
★ 7.2kClaude Code compaction where Jev judges every tool call and result: keep verbatim, truncate, or drop. Nothing kept is rewritten by a model.
AnyJev
★ 949Nokia's open decision model, Jev-API compatible, running locally in about 15ms. The clearest sign yet that the format, not the vendor, is the asset.
von
★ 610+Sub-15ms non-autoregressive System One model, self-hosted, Jev-shaped. The open-implementation wave now spans Nokia, von, jeff, JevK5, and a dozen more.
The week it went mainstream
40M viewsBusiness Insider sent someone to a Jev hackathon at CodeRabbit's office. Developers described wiring it into 80% of an agent's code infrastructure. Signups reopened on the 28th after being suspended, and the $5 signup credit was withdrawn over abuse.
New this week first seen between 24 and 30 September
The freshest entries that already prove something. Each links to its repository.
new building-with-typesafe-jev
Agent skill for writing Jev programs: typed questions, confidence gates, and when not to reach for it. 121 stars in four days.
new quicksilver
Claude Code skill that hands bulk judgment calls to Jev. Claims 86% fewer Claude tokens across a 12-task benchmark, self-reported.
new OneJev
A multimodal System One decision model, answering typed questions over image input rather than text alone.
new jevpipe
A Unix filter. Pipe anything in, get typed decisions out. The shell-pipeline mental model applied to a decision model.
new awesome-jev-security
Curated list of papers on the security, robustness, and safety of Jev and System One models. The failure-mode literature, gathered in one place.
new zh-decision-bench
First Chinese-language calibration benchmark for Jev-class decision models. Most calibration work is English-only, so the gap matters.
new jev-mail-filter
Gmail filters written in plain English and judged by Jev. The most literal reading of the whole idea, and it works.
new stefafafan/jev
Provider-neutral Unix client for the System One API, written in Go. Another entrant in the crowded-but-tiny Go SDK category.
new graphify-jev
Codebase knowledge graph with Jev-driven semantic navigation and no vector database, per the description.
new anofox-decide
DuckDB extension that evaluates natural-language predicates and answer sets with Jev. The SQL-with-judgment genre, still growing.
Browse by usecase
The full curated list lives on one page, sorted by what you are building.
What Jev is, in one table
One request carries any number of typed questions, evaluated in parallel against the same state. Your code owns the thresholds, the fallbacks, and every side effect.
| Primitive | You declare | It returns | Typical use |
|---|---|---|---|
| Noul | a yes/no statement | a probability from 0 to 1 | eligibility, membership, health checks |
| Choice | a list of options, up to 255 | selected option + probabilities + confidence | routing, classification, matching |
| Score | an ordered rubric, 2 to 10 levels | weighted score + distribution + confidence | severity, relevance, quality grading |
Last updated 30 September 2026.