Jev in Practice: Browser Use and Agent Workflows

A developer reports seeing a Twitch automod extension that uses Jev to drive its moderation judgments, cited as a real community deployment.

A long post tries to clarify what Jev actually is, when it is worth using over existing setups, and where it falls short.

A discussion of Jev's captcha-solving costs: one Chinese marketplace pays $0.01 per captcha, while Jev solves 100 for $0.0068.

Suggests feeding the Jev use-case collection to your own agent so it can propose new scenarios, and says a full guide is coming soon.

The demo board adds five Jev cases, including seven Jevs playing Warcraft micro, a resume match in 61 seconds, and cheap gates for eToro trading bots.

Using free Jev access on the Vercel AI Gateway, the author built an open-source library and x402 API that returns indexes into the input and slices the text, so no LLM is needed to produce output.

A roundup of eight demos built with Jev and shared on X, including finding flights, trading, and playing Mario.

mac_eth launched jev-decisions.vercel.app so people can try Jev's structured decisions, and said it will be added to surplusintelligence.ai soon.

Trajectory co-founder Ronak open-sourced a Minecraft agent driven by Jev, billed as the fastest AI agent to complete Minecraft, and the code on GitHub is spreading.

Under the same seed and the same four-way decision space, Jev scored 21 points in 60 seconds, beating six models including Haiku and Opus, with a median decision time of 263 ms.

The author open-sourced jeval, which checks what a Jev confidence of 0.8 actually hits on your logs, computes the right threshold from misjudgment cost, and alerts in CI when a model update makes things worse.

The author tested Jev on buttons, forms and delayed UI and found it strong on interactive tasks, while plain search was better for static fact lookup; deterministic rule routing also beat learned routing 60/60.