jev-gate
Claude Code and Codex plugin where Jev judges each turn for model routing, delegation, evidence search and compaction.
Overview
Jev Gate addresses a quiet cost in long coding sessions: one growing main context is re-read on the most capable model for every request, so routine work like searching, checks and reviews pays the same as the hard part. Jev judges each turn, code applies the policy, and the coding host executes it: simple bounded work goes to a fast worker or lighter model, hard reasoning stays, and old context is compacted to what the current task needs. A local dashboard shows every decision and its observed result for both Claude Code and Codex.
The README reports a pilot on two real feature tasks, each a 17-turn Claude Code session on Opus 5.5, with and without Jev Gate defaults: cost fell from $42.81 to $19.65 including Jev requests, total tokens from 111.7M to 41.3M, agent time from 124 to 65 minutes, and peak main context roughly halved, with quality blind-graded against the merged PRs. The author calls this a pilot showing direction, not a guarantee. Setup is installing the plugin and entering a Jev key, with Node 22+.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which model or worker should handle this turn?(inferred) | — |
| Yes / no | Should this work be delegated to a lighter worker?(inferred) | — |
| Yes / no | Is this old context still needed for the current task?(inferred) | — |
How it works
- Install the plugin and enter a Jev key
- Each turn, Jev judges routing, delegation, search and compaction
- Code applies policy; the host executes
- Dashboard shows each decision and its result
Strengths
- Reported 54 percent lower cost in a pilot
- Works with Claude Code and native Codex
- Live dashboard of decisions
Caveats
- Pilot of two task pairs only
- Experimental routing can misjudge difficulty
Getting started
Install the plugin for Claude Code or Codex and enter your Jev API key; Node.js 22+ required.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Claude Code and Codex plugin
- Has code
- yes
Use cases and integrations
How curated lists describe it
Not every coding task needs your best model.
wh000wh000/awesome-jev-liveNot every coding task needs your best model. Experimental Jev-powered model routing for Claude Code — V3 prototype runs today, V4 routes at the task boundary.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
How to route, delegate, search and compact on each turn.
How much did it save?
In a two-task pilot, cost fell 54 percent and tokens 63 percent with blind-graded quality.
What do I need?
Node.js 22+, a Jev API key and your host's normal login.
Related projects
Claude Code mod that picks reasoning effort per prompt with a Haiku judge, optionally Jev, and shows cache and context state.
ProjectbetaTypeScriptClaude Code plugin that picks model tier and effort per turn with Jev and weighs each switch against cost, with Jev-based compaction.
ProjectbetaTypeScriptClaude Code mod using Jev to pick model and effort per turn, size subagents, and prune stale context verbatim instead of summarising.
ProjectbetaTypeScriptOpen-source AI gateway for 100+ LLMs that can use Jev to classify requests for complexity routing and to compact tool results.
ProxyproductionPythonHermes Agent plugin that picks the model each turn with a direct TypeSafe Jev Choice and enforces trim-then-compress context handling.
PluginbetaPython- jev-router★ 593
Per-turn model routing for Claude Code and Codex: wraps the real CLI and lets Jev send simple turns to fast models and hard ones to strong.
ProxybetaJavaScript