Jev in Practice: Browser Use and Agent Workflows

A list of 20 open-source projects shows where Jev handles fast decisions in agents, coding, trading, and automation, from a browser agent and context compaction to code review, market making, and Mario.

The author built open-jev, an open in-browser version of Jev that takes text in and returns typed answers in one forward pass, fully on-device, and since it generates nothing it cannot hallucinate.

A walkthrough of a pipeline from request to structured state, Jev router, cheapest capable model, worker, approval check, and tool, moving yes/no decisions off expensive models; the author reports it is 193x faster and 444x cheaper.

A Claude Code mod uses Jev, via the Typesafe AI API or Vercel AI Gateway, to pick the subagent model and effort level for each request, sending simple tasks to Haiku at low effort while the interface stays the same.

The author lists eight practical Jev use cases, from an agent decision layer and context compaction to Android phone control and Minecraft action selection, with GitHub repos for each.

A video shows a BTC trading bot built with Jev that goes from real-time price data to rapid decisions and autonomous execution, running 100x leverage on a $10K balance and showing $286 profit after 3 minutes.

While hospitalized, the author listened to a talk by Jev CEO Diogo Almeida on designing a model that doesn't flatter users and outputs calibrated confidence, and judges the company may be acquired or absorbed into larger models.

A tool that takes a sentence and pulls matching clips from long videos, with BaoCut handling subtitle recognition and export while Jev judges from the subtitles whether a clip matches the search intent.

An open-source local Jev built on KaLM-Reranker that offers Choice, Score and Noul judgments in Nano, Small and Large sizes, callable via HTTP or Python and suited to support triage and agent routing.

Companies without accumulated domain data can start labeling data with Jev, have business staff check the results, then run fusion prediction, cutting a process that used to take 3-6 months down to hours or a week.

A roundup of eight practical cases of Jev within a week of its release, including finding flights, trading and playing Mario.

The author says they built a Jev layer with agenkit in their trading system to generate alpha, is openly sharing a cheatsheet, and questions using Jev plus Polymarket as a trading router.