Jev in Practice: Browser Use and Agent Workflows

The author presents a stack where Jev handles every yes/no, routing and scoring call, waking Opus 5.5 only for low-confidence or prose cases, so 1.2 million decisions in a day led to just 6 paid calls.

A second brain setup that drew 8 million views has open-sourced its full repo, guide and toolchain, including a learning track dedicated to Jev engineering.

A working note based on design notes from Jev's founder stores tool calls, diffs and AGENTS.md as typed, addressable chunks, with Jev deciding each turn which chunks to show and which model and tool to use.

Serverpod's official blog explains how it used Jev to cut its AI agent's response time in half, with a link to the full implementation.

The author open-sourced jevclip, built on Jev, which takes a video plus subtitles and returns a timestamped summary, a highlight clip, and a filler-free cut; processing 200+ GB of video cost only $0.22 in Jev judgments.

The author notes that Jev is not a text-generating LLM but a model for selecting, scoring, and making YES/NO judgments, and says they are currently testing it with the OpenRouter API.

After trying TypeSafe's Jev as a router in a project, the author applied it to the frontend to build a fluid Google Search UI experiment where a matching interface pops up as you type.

In Automata Room, users change robot designs in natural language, and Jev reads real values like joints, reach, mass, and limits to score the design against the brief and decide what to change next.

Jev Brain routes agent decisions locally before the model, auto-executing when confidence is at least 0.80 and escalating only ambiguous cases to Claude, Grok or GPT, with tests showing a 95.2% token reduction.

The author argues for calling expensive models only when needed, and says Jev can bound a model's scope with probability thresholds so calls are targeted.

An in-person buildathon in Bengaluru on 27 September asks participants to bring an agent that does real work, and judges will review every decision it makes.

The author open-sourced the full game on GitHub, added Jev and AI battle modes, and published the entire development process.