Jev in production › Model and agent routing

ego-decision-layer

Replaces the per-step LLM turn in ego lite browser automation with one Jev decision to pick the next action, cutting median latency to 434ms (author).

Open on GitHub ↗

434msmeasured against a baseline, as published by the source
Use
Model and agent routing
Industry
Developer tools
Form
Open-source tool
Stage
Beta
Plugs into
ego lite
Listed
2026-09-26
Found via
discord
Repository
jiangkoumo/ego-decision-layer
Stars
5
Forks
0
Last push
2026-09-27
Language
JavaScript
License
MIT
ego-decision-layer screenshot
docs/demo.gif in the jiangkoumo/ego-decision-layer README, MIT; shown from GitHub.

The README opens with

给 ego lite 的可插拔决策层：默认 System One（Jev），可换本地 / 其他 OpenAI 兼容后端；执行层带 fail-closed 护栏。

为什么是这一个：差异只有一条 —— 它是被测量的。本机 2026-09-26 全量复跑 16 套测试 / 429 项检查 （除 native-select 是成功率测量外全部 exit 0）；原始数据在 bench/raw/，每个数字都能用仓库里的 脚本复算；CHANGELOG.md 里记着被实测推翻的旧结论（包括我们自己先前公布的口径错误）； 执行层是 fail-closed（陈旧 ref、被遮挡、跨 frame 命中失败、危险动作一律 0 次派发）； 代码级溯源登记在 THIRD-PARTY.md。

English summary: ego-decision-layer replaces the per-step LLM round trip in browser automation with TypeSafe's System One model Jev. Jev reads one indexed element table and answers, in a single request, both the operation (click / typetext / select / scrollup / scrolldown / wait / done / blocked) and the target element. Code owns observation, execution, verification and exit conditions. Measured after the engine...

Badge

For the project's own README, linking back here:

Listed in Jev in production

Also used for model and agent routing