Jev in Practice: Browser Use and Agent Workflows

A video traces Jev's path from a $40M seed round to a $7.5B valuation in 24 days, and OpenAI's fast follow with its Decisions API.

The author shares the agent setup they use every day: Opus 5.5 as the main driver and Jev for fast yes/no judgments at specific points in the workflow.

The author shows how to let Opus plan, Haiku read files, Sonnet fix issues and a verifier agent re-check, while handing deterministic branches like tool choice to Jev, with a prompt to paste in.

The desk screens with free hard rules first, then uses Jev for soft filters like exit risk, momentum and liquidity, as the account grew from $12,155 to $14,853.

The open-source jev-seo project crawls a whole site, checks it against Google's 52 rules, and uses the Jev model to judge each page's purpose, content strength and keyword conflicts, outputting PDF, Excel and Markdown reports in one run.

Six Grok agents and a crawler trade BTC futures, with a Vet agent validating signals through Jev before each position opens; the post shows four consecutive long trades, the last filling in 179 milliseconds.

Using jev-recall to filter Obsidian notes before Claude answers, so only relevant memories are fed to the model, with setup steps included.

A hands-on test shows that replacing a conclusion-style state with separate evidence fields raised Jev's confidence on which worker runs next from 0.61 to 0.93.

The author of the open basal decision models reports that basal-omni-ultra scores 0.879 macro accuracy on his 5,000-item Werdykt v1 test versus 0.822 for TypeSafe's Jev 1.14, leading in 9 of 10 areas, with the leaderboard and samples public pending independent retesting.

The author walks through the purpose of each Termix campaign, noting that buying the Jev FC Buddy AI football companion earns 500 TP plus another 500 TP for sharing on X.

The author shows routing judgments like whether it is safe, whether the task is done, and which tool comes next to Jev, which returns Choice, Noul or Score in 250ms, with cost and accuracy figures from a 100-email fraud pipeline.

A developer who tested Jev on classifying emails and notifications says it could not do so reliably and was not accurate four out of five times, calling it too dumb for decision-making.