jev-governor
Claude Code mod using Jev to pick model and effort per turn, size subagents, and prune stale context verbatim instead of summarising.
Overview
jev-governor is a Claude Code mod built only on the official mod mechanism (function hooks inside the unmodified client), with no proxy, base-URL override or token handling. It uses Jev via the user's OpenRouter key to cut session costs without lowering quality. Effort (low to max) is picked every turn, and switching to Sonnet 5.5 is priced in dollars, weighing the cost of rewriting context into the other model's cache against output savings; short continuations are decided locally.
Subagents get a model and effort per task at spawn: easy read-only subagents run on Haiku 5.5 and move up to Sonnet when tasks grow or tools keep failing, and generic subagents are matched to specialist roles, with new ones drafted when none exist and near-duplicates merged daily. Compaction removes stale tool calls and results while keeping remaining text verbatim, running at 200k tokens when Jev can remove at least 40 percent and whenever Claude Code compacts. Documentation is in English and Russian.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which model and effort level should this turn use? | low · medium · high · max |
| Choice | Which model and specialist role should this subagent get? | — |
| Yes / no | Is this old tool call and result stale? | — |
How it works
- Hooks intercept each turn and subagent spawn
- Jev judges effort, model and subagent role
- Price model switches against cache costs
- Prune stale tool output verbatim at the context threshold
Strengths
- Uses only official mod hooks
- Verbatim compaction instead of summaries
- Cost-aware model switching
Caveats
- Requires an OpenRouter key
- Depends on the Claude Code mods API introduced in 2.1.287
At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Claude Code mod (function hooks)
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Claude Code mod: Jev-guided model/effort routing, verbatim context compaction…
wh000wh000/awesome-jev-live
Frequently asked questions
What does Jev decide?
Effort and model per turn, subagent model and role, and which old tool output to drop.
Does it proxy my traffic?
No, it uses only official Claude Code mod hooks.
Does it summarise context?
No, it removes stale items and keeps the rest verbatim.
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 and Codex plugin where Jev judges each turn for model routing, delegation, evidence search and compaction.
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