Jev in Practice: Browser Use and Agent Workflows

A Codex-written Jev judge covered 34 voice agent metrics, with 27 exceeding 96% on an untuned validation set at about $0.07 of reasoning per metric.

The author applies Jev to pairs of tickets: the model sees both A and B together and outputs a small vector of pair questions (same issue, A blocks B, duplicate), stored in a matrix cell instead of comparing two embeddings with cosine.

A short post notes that @opencomputerHQ serverless agents can now call typesafe/jev-router to make routing decisions.

An Arabic post says OpenRouter has officially integrated Jev to route models in agentic workflows, promising faster responses, lower cost and automatic model selection.

A practical guide that places Jev as the decision layer in agent architecture, covering how the Choice, Score and Noul primitives handle routing, approvals and safety gates.

A roundup of five popular Hermes plugins, including browserclaw, which uses a Jev micro-loop for browser decisions, plus a memory wiki, bot workstations, an SSH workspace, and Google subscription access.

TypeSafe's demo shows a Python script sending state, asking one focused question, and getting a typed result so the code decides the next step, for uses like bug triage, routing, classification, and review decisions.

Max Leiter asks whether most Jev use cases could be handled with similar quality, faster and cheaper, by using embeddings.

Laura Shin relays Austin Griffith's view that not every AI task needs paragraphs of reasoning, and that classifier models like Jev can handle browser and CI work in 200 milliseconds, faster and cheaper than a chatty frontier model.

The author open-sourced JEV/LAYA for local structured decisions, plus two apps built on it: OMNIA News for content filtering and OMNIA Trading for market risk assessment.

The author lists five mistakes people make after installing Omarchy, setting up Hermes Agent and connecting Jev, with the real test being whether the setup can finish one job and flag what needs human review.

VLM Run's gateway added TypeSafe SDK compatibility, letting Gemma make decisions directly from game images to play Snake at under 240ms p99 latency and about $0.00007 per image.