Jev in Practice: Browser Use and Agent Workflows

A demo of using TypeSafe Jev for typed choices with confidence scores instead of LLM generation, processing 100 emails in 1.42 seconds with 96 correct at a total cost of seven cents.

An open-source UI tool that classifies inputs with choices, scores, and booleans, wires results into later prompts, and returns custom outputs; the author used it to build a support triage system.

willdepue reflects on Jev's popularity: it is essentially a zero-shot classifier but with near-frontier judgment, and he wonders which other old ML ideas are worth reviving.

The author posts a video claiming a trading bot rebuilt with Jev performs better than gpt-6 astra and Fable, though the post gives no specific data.

For each matchup the author runs five p0 advanced searches (the matchup, injury news, previews, team efficiency, season expectations), deduplicates the results and gives them to Jev as context for its decision.

In a live API demo Jev averaged 213ms latency, completed 107 browser actions in 26 seconds and cost about $0.0004.

The author tested Jev in backtesting and found it cut cost and latency to one percent, and recommends two Jev-plus-trading open-source projects, jev-trader and QuantDinger.

The author previews an upcoming open-source Trading-Jev that lets users plug in their own data CLI, LLM CLI and strategy to run backtests and paper trading.

$NATION has added a Jev integration that lets agents choose actions, score outcomes and make faster autonomous decisions.

The author says tagging research papers looks unglamorous but lets agents use the tags to explore and discover papers, and has started using Jev to search research papers.

The author added a "personal favorite Jev use cases" section to his unofficial TypeSafe Go SDK, embedding scam-shield, Jev Tetris and Jev Pac-Man demos plus screenshots.

A video shows Jev acting as a single decision layer that picks which model to call, which tool to grant access and which result survives, tested at up to 193x faster and 444x cheaper.