modelrudder
Launchers for Claude Code and Codex that use Jev to pick a model tier for each new user turn while keeping native tools.
Overview
ModelRudder is a free, MIT-licensed local preview that adds model routing to native coding CLIs. Its smart-claude launcher routes Claude Code turns among Haiku, Sonnet and Opus, and smart-codex routes Codex turns among three profiles. Each eligible new user turn gets a model choice that persists through that turn's tool loop, while native tools, approvals and workspace permissions remain unchanged.
Routing uses your own TypeSafe Jev key in observe or auto modes; a pinned mode works without Jev. For Codex, valid Jev choices are applied without a confidence gate, including the profile's reasoning effort, and reported confidence is informational. The installer manages separate native CLI versions and reuses your existing login. It targets developers who want to spend their strongest model only on tasks that need it.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which Claude model should handle this turn? | Haiku · Sonnet · Opus |
| Choice | Which Codex profile should handle this turn? | Luna · Sol · Astra |
How it works
- Launch smart-claude or smart-codex instead of the plain CLI.
- On each new user turn, send the request to Jev.
- Apply the chosen model (and reasoning effort for Codex) for the whole tool loop.
- Use pinned mode to bypass routing.
Strengths
- Keeps native tool approvals and workspace permissions.
- No ModelRudder account required; bring your own keys.
- Pinned mode as a no-Jev fallback.
Caveats
- Experimental preview; Windows support is a preview.
- Codex applies choices without a confidence gate.
- Provider charges and limits still apply.
Getting started
Download the v0.1.0-preview.10 release installer, verify SHA256SUMS, install, then run smart-codex setup (or smart-claude) to add your Jev key.At a glance
- Type
- CLI
- Status
- prototype
- Language
- TypeScript
- Runtime
- Node CLI launchers
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Experimental local LLM router for OpenAI Codex CLI and Claude Code. Automatically selects models per task using TypeSafe Jev. Free, MIT-licensed, bring your own provider account.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Which model tier handles each new user turn in Claude Code or Codex.
Do I need a TypeSafe key?
Yes for observe/auto routing; pinned mode works without one.
Does it change my tool permissions?
No, native tools, approvals and workspace permissions remain in place.
Related projects
- 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 TypeScript/Bun launcher that asks Jev which model and reasoning effort each Codex, Claude Code or Pi turn should use.
ProjectbetaTypeScriptGo CLI model router that uses one Jev analysis to recommend a model, harness and effort level, then selects deterministically.
CLIbetaGoClaude Code and Codex plugin where Jev judges each turn for model routing, delegation, evidence search and compaction.
ProjectbetaTypeScriptClaude Code, Claude Desktop and Codex plugin where Jev reads a task and ranks live models and reasoning effort by quality, speed and cost.
PluginbetaTypeScriptLocal pass-through proxy for Claude Code and Codex where Jev picks a model tier per human turn and only moves up within a session.
ProxybetaTypeScript