What does jev-cc-codex-router do?

It is a single-file Python proxy that sits between Codex and any OpenAI Responses API-compatible upstream. On every new user turn it asks Jev which of four tiers the task needs, rewrites the request's model field, and keeps that choice for the rest of the turn.

What are the four tiers?

luna, terra, sol and astra, each mapped to a configurable model name and stored per session. If Jev returns an unknown tier or the call fails, the request goes to the fallback model, gpt-5.6-sol by default.

Does it cost or save money?

It costs one small Jev call per new turn plus at most one per failing step, but promises no specific saving: the author says tier prompts are still being tuned and BENCHMARK.md holds one small self-measured estimate with its conditions and limits.

How do I set the router up?

Python 3.9+ (3.14 for built-in zstd), copy .env.example to .env and set TYPESAFE_API_KEY, set JEV_UPSTREAM if it is not http://127.0.0.1:15721, run python3 jev_router.py, then point Codex's model_provider at http://127.0.0.1:8787/v1.

Does the proxy remember sessions across restarts?

No. Sessions are identified by the session-id or thread-id request header, and after a restart the proxy knows no sessions, so requests pass through until the next new turn. The README says the proxy holds no persistent state.

What does it do on failures and retries?

For POST /responses it retries up to 2 times after a 1-second wait on upstream 400/502/503/504 or a connection error, which masks flaky upstreams. It cannot recover a stream that is cut after a 200 response.

What does debug mode log, and is it private?

Off by default. JEV_DEBUG=1 writes decisions.jsonl with tier, probabilities and in/out model per request, req-headers.jsonl, and errors-400/ with full failing requests. These contain prompt excerpts, so keep them private; with debug off the proxy writes no files.

Jev directory

CommunityJev Integrations & Adapters

Jev Cc Codex Router

A single-file Python proxy that asks Jev which of four model tiers each new Codex turn needs — luna, terra, sol or astra — rewrites the model field, and retries flaky upstream errors.

See highlights Open github.com

Stars: 2
Language: Python
License: MIT
Reading time: 3 min
Published by: Community
Author: peterwanghot
Last verified: Oct 2026

peterwanghot/jev-cc-codex-router

User repository on GitHub

View on GitHub

Per-turn model routing proxy for Codex: asks Jev which tier each task needs, rewrites the model, retries flaky upstream errors.

Forks: 1
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity catalog-import logicrw codex routing proxy

Works forJev for Routing

Used inJev for Software & DevTools

Languages: Python
Platforms: API
Capabilities: Open sourceRequires API key

Highlights

Each new turn sends Jev the message plus recent conversation (up to 50 turns, ~4200 characters) and asks for one of four tiers: luna, terra, sol, astra.
Tool-result follow-ups reuse the stored tier, so there is no extra Jev call within a turn; bare continuations reuse the previous tier without asking.
On clear tool failures it asks Jev once more and escalates only to a higher tier, at most once per 20 seconds per session; unknown tiers fall back.
Retries POST /responses up to 2 times after 1 second on upstream 400/502/503/504 or connection errors, and handles zstd-encoded request bodies.
The README labels status early with tier prompts still being tuned; BENCHMARK.md holds one small self-measured cost estimate with its conditions.

Quickstart

bash

cp .env.example .env
python3 jev_router.py

Watch out

Copy reviewed against the truncated repo README on 2026-10-15; the excerpt cuts off inside the notes-and-limits section. The author labels the router early, downgrades are never automatic, and the only cost evidence is a self-measured estimate whose conditions live in BENCHMARK.md.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.

FacebookClinicnotes

X: OpenRouter Decision Model Rankings

X: herdr + jev, an agent router

X: A group chat bot with decision room routing

From the Jev guides

Original write-ups on what this entry is for.

Decision gates in agent frameworks: what each integration ships

Oct 16, 2026

How Pydantic AI, Mastra, TanStack AI, DSPy, Spring AI, and LiteLLM expose Jev: where the typed decision sits in the loop, what each integration returns, and which big frameworks are only adjacent.

Jev Integrations & Adapters#agents#frameworks#integrations

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai