____ _ _ ____ _____ ____ ____ ___ ____ _______ __ / ___|| | | | _ \| ____| _ \ / ___/ _ \| _ \| ____\ \/ / \___ \| | | | |_) | _| | |_) | | | | | | | | | | _| \ / ___) | |_| | __/| |___| _ < | |__| |_| | |_| | |___ / \ |____/ \___/|_| |_____|_| \_\ \____\___/|____/|_____/_/\_\
Codex, made to keep up with you.
Save every ChatGPT plan you pay for and hop between them mid-task. When one runs out, the next one picks up where it left off. Or move the whole conversation to a model running on your own machine, in one command.
macOS · Linux · Windows·auto-picks the right binary·vv0.130.0
01 · multi-account
every plan you own, one keystroke away.
If you pay for more than one ChatGPT plan, Super Codex remembers all of them. Switch the active one live with /accounts, no restart, no losing the conversation you were in.
When the active account hits its usage cap, the next saved account takes over mid-turn. Your prompt re-fires under a fresh token before you notice you were rate-limited. You get the answer; the rotation happens quietly in the background.
- no more mid-task interruptions. the agent keeps going while you stay in the editor.
- use the plans you already pay for. two seats, two budgets, one tool.
- add, remove, switch from the prompt bar. browser OAuth handles sign-in.
↑ ↓ to move · enter to switch live · /addaccount to add another
installed on this machine
download · curated
auto-loaded with the model's real max context. first turn doesn't overflow.
02 · local models
your machine, your tokens, your rules.
/local walks you through getting LM Studio ready on your computer, then remembers your choice. From the next launch on, the agent runs against a model that lives on your disk. no tokens, no waiting room, no data leaving the box.
Metal on Apple Silicon, CUDA on NVIDIA, ROCm on AMD, CPU everywhere else. Super Codex hands off the heavy lifting to LM Studio and stays out of the way.
- burn local compute, not credits. a 4-bit quant solves most snippets.
- a curated catalog, one ⬇ away. granite 4.1, qwen 3.6, gemma 4, gpt-oss.
- offline-friendly by default. your code stays where you wrote it.
03 · install
one command. every machine you work on.
the npm package is a thin launcher. on first run it picks the right native binary for your platform and pulls it from the matching github release. after that, you just type supercodex.
already have it? update with the same command. the latest tag is always pulled at install time.
04 · trust
open code, honest scope.
Super Codex is an unofficial fork. It does not pretend to be an OpenAI product, and it does not try to replace one. It adds two things to a great open-source agent, then steps back.
fork, not rewrite
Everything you already know about Codex still works. Sign-in, slash commands, IDE extensions, the desktop app, the sandbox. Super Codex only adds the two pieces that were missing.
tracks upstream tightly
Every release maps 1:1 to an upstream Codex stable. When Codex ships, Super Codex catches up the same week. No drift, no surprises.
apache 2.0
Same license as the upstream project. Read the code, run it on your machines, contribute back. No telemetry baked into the fork's own paths.
built by someone who uses it
Maintained by Mattia Beltrami, Computer Engineering at Politecnico di Milano. Super Codex is the tool I wanted to use myself; the README and SUPERCODEX.md cover everything else.