Skip to content

Codex Jev Router

Deterministic evidence selector for noisy Codex investigations; its earlier Jev subagent router is kept for the record, not recommended.

ProjectresearchMIT

Overview

Codex Jev Router's recommended use is now opt-in evidence selection for noisy investigations. A local MCP tool searches a workspace and gives Codex a few exact, cited excerpts instead of a large raw result, while the session keeps its configured model and effort. The default selector is deterministic and needs no Jev account, key or local model; Laya remains an optional experiment, which the author found did not improve cost over deterministic selection in a holdout.

The original subagent-model router, which asked Jev Choice and Noul questions about a task summary to pick a Codex subagent model and reasoning effort with confidence gates and a Sol fallback, remains in the source for reproducibility but is not recommended or installed. Its four-arm benchmark found Jev-selected children 27.2% cheaper than fixed Sol-high children but 69.7% more expensive than a single Sol-xhigh agent, documented in an economics audit. Docs are in English, Russian and Chinese.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich model and reasoning effort should this subagent use? (legacy router)—
Yes / noIs the router confident enough to deviate from the Sol fallback? (legacy)(inferred)yes · no

How it works

  1. Codex calls search_workspace_evidence
  2. Deterministic selector picks exact cited excerpts
  3. Codex reads more with read_selected_evidence if needed
  4. (Legacy) Jev picked subagent model and effort with a Sol fallback

Strengths

  • Honest negative result published for the router
  • No key needed for the recommended mode
  • Workspace-scoped file access

Caveats

  • Jev router not recommended by the author
  • MCP server registered globally

Getting started

Run the installer to add the codex-local-evidence server to ~/.codex/config.toml.

At a glance

Type
Project
Status
research
Runtime
Codex MCP server
Has code
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Coding agents: asks Jev Choice and Noul questions about a short task summary to select a Codex subagent model and reasoning effort, with confidence thresholds and a Sol fallback.

    yibie/awesome-jev
  • Codex subagent routing: Jev chooses a model and reasoning effort from typed Choice and Noul answers; code applies confidence gates and falls back to Sol.

    AnotiaWang/awesome-decision-models
  • Portable JevRouter setup for cost-aware Codex subagent model selection, with English and Russian instructions

    logicrw/awesome-jev-projects
  • Uses Jev Choice and Noul decisions to select a Codex subagent model and reasoning effort; uncertain decisions fall back to Sol.

    cobanov/awesome-jev
  • Use Jev Choice and Noul decisions to select Codex subagent model and reasoning tiers, with local compatible backends and a Sol fallback.

    hellogumbo/awesome-jev

Frequently asked questions

Should I use the Jev router?

The author no longer recommends it; benchmarks showed it cost more than one Sol-xhigh agent.

Does the current tool need Jev?

No, the default selector is deterministic.

Where are the economics?

In AUDIT.md in the repository.

Related projects

  • Claude Code and Codex plugin where Jev judges each turn for model routing, delegation, evidence search and compaction.

    ProjectbetaTypeScript
  • Open-source AI gateway for 100+ LLMs that can use Jev to classify requests for complexity routing and to compact tool results.

    ProxyproductionPython
  • Claude Code mod that picks reasoning effort per prompt with a Haiku judge, optionally Jev, and shows cache and context state.

    ProjectbetaTypeScript
  • Hermes Agent plugin that picks the model each turn with a direct TypeSafe Jev Choice and enforces trim-then-compress context handling.

    PluginbetaPython
  • Claude Code plugin that picks model tier and effort per turn with Jev and weighs each switch against cost, with Jev-based compaction.

    ProjectbetaTypeScript
  • Claude Code mod using Jev to pick model and effort per turn, size subagents, and prune stale context verbatim instead of summarising.

    ProjectbetaTypeScript