Book a flight in 7 seconds and find code for 30% less: Jev has been turned into five tools

GitTrend summarized five projects built on the TypeSafe Jev decision core in a tweet. They put Jev into different niches: browser operations, context compression, code search, on-chain market making, and SQL condition judgment. Original link: https://x.com/GitTrend0x/status/2113986396062925178

1. browser-use/jev-ultrafast

An ultrafast browser agent from Browser Use and TypeSafe. Jev chooses both the action and target element in a single request, and calls a small model only when typing is needed. The example provided is a Google Flights search from Zurich to London, completed in 7.1 seconds.

Link: https://github.com/browser-use/jev-ultrafast

2. tamaratran/fast-jev-compaction

A Claude Code plugin. It uses Jev to score each tool call and result, deletes or truncates stale content, and leaves retained content untouched. It aims to replace summary-based compression that loses details.

Link: https://github.com/tamaratran/fast-jev-compaction

3. dzhng/jevgrep

Ask 'where is this feature implemented' and it returns relevant files, reading leads, and source code snippets, with Jev judging relevance. The author compared it using 10 SWE-bench tasks: 8/10 successful tasks, the same as the baseline, at about 30% lower cost.

Link: https://github.com/dzhng/jevgrep

4. jarrodwatts/jev-trader

A market-making experiment on Monad. Jev watches Kuru's MON-USDC order book and gives a buy or sell decision roughly every 300 milliseconds per block, placing maker-only limit orders to earn the spread. It defaults to a dry-run that simulates fills first.

Link: https://github.com/jarrodwatts/jev-trader

5. realZachi/pg-jev

A PostgreSQL extension. You write plain-language conditions directly in WHERE, such as 'is this customer very angry?', and each row is passed to Jev for judgment and returns a probability. The material says it requires no indexes or vectors and can be used with JOIN and GROUP BY.

Link: https://github.com/realZachi/pg-jev

Some users asked in replies whether pg-jev can handle the latency of scanning tens of thousands of rows. The material gives no answer. The real cost of such projects is usually hidden in one judgment per row and score-based filtering. The 7 seconds in the browser and the 30% savings in code search are results; full-table semantic judgment in a database still needs real-world testing.

Looking at these projects together, Jev's usage has one commonality: push judgment into the shortest part of the workflow. When booking tickets, it selects actions and elements; when compressing, it scores calls; when finding code, it judges relevance; when market making, it makes a decision every block; in SQL, it gives a probability per row. The direction isn't complicated; the hard part is not letting judgment itself become a new bottleneck.