Super Codex is an unofficial fork of OpenAI Codex shaped around how real developers actually work.

---

Why this exists

Codex is a great coding agent. It is also a single-account, single-provider product built around one user model: log in once, use one plan, talk to one set of servers.

Real workflows don't fit that shape.

• You hit a usage cap mid-task and lose your flow while you scramble to switch accounts.
• You'd rather burn local compute than pay for tokens on a snippet a 4-bit quant can solve.
• You bought two ChatGPT plans for a reason and resent juggling logins to use them.

Super Codex keeps everything that's good about Codex — the agent, the TUI, the sandbox model, the IDE integrations — and adds the two things you actually need to push it past one account, one provider, one machine.

---

What's different

Multi-account ChatGPT Save every ChatGPT plan you have and hop between them in one slash command. /accounts lists them; ↑/↓ + Enter switches live, no restart, no lost context. When the active account hits its usage ceiling Super Codex rotates to the next saved account mid-turn, transparently. Your prompt re-fires under the new token before you notice you were rate-limited.

Local models, zero friction /local walks you through getting LM Studio ready on your machine, then remembers your choice. From the next launch on, supercodex runs against the model you picked — Metal/MLX on Apple Silicon, CUDA on NVIDIA, ROCm on AMD Linux, CPU everywhere else. LM Studio owns the heavy lifting; Super Codex owns the glue.

Everything else — sign-in flow, slash commands, IDE extensions, the desktop app, configuration, approvals, sandbox model — works exactly as in upstream Codex.

---

Commands you'll actually use

/accounts         list saved ChatGPT accounts; ↑/↓ + Enter switches live
/addaccount       add a new ChatGPT account (browser OAuth flow)
/removeaccount    delete a saved account (auto-rotates if it was active)

/local            set up or switch to a local model via LM Studio
                  ─ shows installed models first (with size + ctx)
                  ─ then a curated downloadable list, grouped by
                    family release date (Granite 4.1, Qwen 3.6,
                    Gemma 4, Qwen 3.5, gpt-oss, ...) — pick ⬇ to
                    download in the background with `lms get`
                  ─ auto-loads the chosen model with its real max
                    context so the first turn doesn't overflow

Auto-rotation on usage-limit and registry mirror on token refresh are automatic — no command needed, they just work.

---

Install

npm install -g @beltromatti/supercodex
supercodex --help

The npm package is a thin launcher: postinstall downloads the right binary for your platform (macOS arm64, Linux x64, Windows x64) from the matching GitHub Release.

---

How it actually works

→ Read SUPERCODEX.md for the full technical reference

A deep, opinionated, agent-friendly document covering:

• The standalone-crate pattern that keeps the fork's logic out of upstream files (every hook site is grep-discoverable, ≤ 25 sites total)
• The multi-account orchestration: refresh-lock coordination, mirror-on-refresh policy, auto-rotation semantics, plumbing of the shared Arc<AuthManager> from app-server up into the TUI
• The LM Studio integration on top of upstream's codex-lmstudio crate
• The release pipeline (npm package layout, postinstall binary fetch)
• The maintenance workflow an AI agent should follow when rebasing the fork onto a new upstream stable release — including the smoke + behavioural test suite that catches regressions automatically

---

Everything else

Super Codex is a fork, not a rewrite. For installation, sign-in, slash commands, IDE extensions, the desktop app, configuration — refer to upstream:

→ openai/codex README

→ Codex Documentation

The only additions are the ones above. Upstream is the source of truth for everything else.

---

Author

Built by Mattia Beltrami — Computer Engineering student at Politecnico di Milano. Super Codex is the tool I wanted to use myself.

If you find this useful, a star on GitHub helps it reach the next developer who's tired of switching tabs to swap accounts.

---

Contributing

• Contributing guide — scope, where issues belong vs. upstream, merge-window cadence, PR checklist.
• Technical notes — see above; this is also the place to start if you want to send a PR.

---

Disclaimer

Warning

Super Codex is not affiliated with, endorsed by, or supported by OpenAI. It is a personal fork maintained by an independent developer.

The software is provided "as is", without warranty of any kind, express or implied. The author is not responsible for:

• how you use this software, including any interaction with OpenAI's services or terms of use;
• any data loss, account action, billing impact, or damage caused by running it;
• any discrepancy between its behaviour and the official OpenAI Codex;
• anything produced, suggested, or executed by the agent while it is running.

If you need official support, a supported product, or a stable release cadence, use the official OpenAI Codex instead. If you have any doubt about whether a given feature complies with OpenAI's terms of service for your plan, review those terms yourself before enabling it.

---

License

Super Codex inherits the upstream license — Apache-2.0, the same license as OpenAI Codex.

_{Built on top of OpenAI Codex · Not an OpenAI product · Use at your own discretion}