Codex\_ChatGPT\_JEV\_Switch
Fork of codex-with-chatgpt where Jev plus fixed rules decide when to bring in web ChatGPT to plan or review and when Codex just acts.
Overview
Codex x ChatGPT JEV Switch extends codex-with-chatgpt, where web ChatGPT reads a local repository through a read-only MCP connection to plan and review while Codex executes. The original switches only manually and sends every round back to ChatGPT, costing minutes of waiting and Codex tokens even for trivial changes.
The smart switch, off by default, uses Jev plus deterministic rules. For a new task Jev answers eight questions (task type, change scope, whether design trade-offs are needed, whether the goal is clear and four risk categories) and code weights them to decide: Codex does it directly, Codex does it then asks for review, ChatGPT plans first, or ask the user. After repeated command failures, Jev classifies the error and whether a human is needed. When ChatGPT replies DONE with follow-ups, Jev estimates each item's size. Jev can only make the system more cautious, never skip a review; without Jev it falls back to original behaviour, and intake calls are capped at 3 seconds. The README reports an online evaluation on 152 samples with jev-1.13.0. Docs are in Chinese and English.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What type of task is this and how large is the change? | — |
| Yes / no | Does this task carry auth, data, concurrency or external-API risk? | yes · no |
| Choice | What kind of error is this, and does it need the user? | — |
How it works
- New task arrives; Jev answers eight intake questions
- Code weights answers to pick Codex-only, review, ChatGPT-first or ask
- On repeated failures, Jev classifies the error
- On DONE with follow-ups, Jev sizes each item
- Jev can only increase caution
Strengths
- Avoids slow round trips for trivial changes
- Escalates when Codex is stuck
- Never blocks the task; falls back cleanly
Caveats
- Off by default
- Depends on web ChatGPT via a browser
Getting started
See the routing and security docs in the repository; enable smart switching and add a TypeSafe key.At a glance
- Type
- Project
- Status
- beta
- Runtime
- Codex skill / MCP
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
This project integrates Jev to provide structured decisions for its workflow. See the repository for implementation details.
logicrw/awesome-jev-projects
Frequently asked questions
Can Jev skip a ChatGPT review?
No, Jev's answers can only make the system more cautious.
What if Jev is unavailable?
It behaves like the original codex-with-chatgpt.
Is it on by default?
No, smart switching is off by default.
Related projects
Fail-open Codex UserPromptSubmit hook that asks Jev for task type, complexity, risk and execution mode and injects it as advisory context.
HookbetaPythonOne skill and MCP server for Claude Code, Codex and Hermes where Jev gates large-model turns, routes models and guards Git steps.
MCP serverbetaPythonMulti-agent Codex factory: Jev decides task shape, decomposition and routing, and failures are classified, not blindly retried.
ProjectprototypePythonClaude Code skill that splits work into PBIs for Codex, Antigravity or Claude sub-agents, using Jev for model choice and triage.
ProjectprototypePythonSoftware-factory supervisor where Jev independently judges coding agents' completion, test sufficiency, progress and need for a human.
ProjectbetaPythonOpenCode V2 server plugin that uses Jev to route requests between Plan, Build and custom agents and hand off unfinished work.
ProjectbetaTypeScript