Vibe Coding Discoverjev-codex-router
View on GitHubPer-turn model & reasoning routing for Codex, driven by Jev (TypeSafe System One): picks the model, thinking depth and speed mode for every turn.
Local server that plugs into Codex Router as a generic provider and classifies every turn with Jev to pick model, reasoning effort and speed tier. Reports ~60% savings vs an all-frontier baseline, with fail-open, kill switch, shadow mode and a dry-tandem fallback.
Use Cases
Cut Codex token cost by routing each turn to the cheapest capable modelAdapt reasoning effort per turn instead of always using maxFail over to cheaper models when ChatGPT usage limits are hitEvaluate routing policy savings via 7-day replay backtestsLog per-turn routing decisions locally for calibrationIntercept Codex turns without forking the router via provider extension points
Built With
- Language
- JavaScript
- Frameworks
- LiteLLM · Codex Router · launchd · OpenAI Responses API · Jev (TypeSafe System One)
Tags
model-routing · codex · cost-optimization · reasoning-effort · llm-router · proxy · macos · python · openai-responses · sse-streaming · launchd · fail-open · backtest · shadow-mode · local-first · typesafe