NucleusCodex runs OpenAI only. Nucleus runs Codex, Claude Code, and Atlas.
One workspace with Claude Code on your Claude Pro, Codex on your ChatGPT Plus, and Atlas — our own agent — running together. Plus auto review on commit, a whole-repo security audit, and assets that land in the codebase. A self-improving codebase, for $20.
Codex runs OpenAI only. Nucleus runs the same Codex you pay for, plus Claude Code on your Claude Pro and Atlas — our own agent — in one workspace, with auto review on commit, repo-wide audit, and assets in the codebase.
Same bill. Whole stack.
One bill. Four reasons.
The image, the clip, the GLB — in the repo.
ChatGPT generates images in a separate chat thread. You download, you drag into your editor, you wire it up. Nucleus calls Studio inside the agent shell, drops the asset into the right folder, and wires it into the right file. No second tab.
Use your existing ChatGPT subscription. Get started in Nucleus for free.
Codex is $20 — bundled in ChatGPT Plus, but it's locked to OpenAI and ships only what OpenAI builds. You still need a PR reviewer (CodeRabbit), a SAST scanner (Snyk), and an asset generator (Midjourney) to ship a real product. The full kit starts at ~$99/mo and balloons into the $1,000s once real usage kicks in. Nucleus plugs into the ChatGPT sub you already pay; that subsidy carries the inference straight through — no markup, no per-token tax. The IDE is free, and the $20 Plus tier bundles everything else.
AI ships 10× faster. And 10× more vulnerabilities.
AI codegen has us shipping 10× faster — and quietly producing 10× more vulnerabilities. Codex's only review path is ask it again in a thread — no diff surface, no whole-repo audit, no inline guard. Nucleus adds three layers — PR review, multi-stage security review, plus an in-app review as you write.
Self-improving. Any event in. Any output out.
A trigger fires — a PR opens, a Sentry spike, a 9 a.m. cron. A Nucleus agent spins up in a cloud sandbox, works the brief, and an output lands. Codex ships `AGENTS.md` and Cloud tasks — both config-file engineering. Nucleus wires events to outcomes — every result lands in one inbox.
One Subscription. Self Improving Codebase.
Who switches, honestly.
Stay on ChatGPT Codex if OpenAI's models are the only thing you want, and a single-lab agent is the entire workflow you need.
Switch to Nucleus if
- you want Claude Code and Atlas running alongside the Codex you already pay for
- PR review at the diff before the commit
- the whole repo audited for security
- in-repo asset generation
- or you don't want to be locked to OpenAI
Nucleus runs 50+ models across 10+ providers, so every new frontier release stays inside the same subscription instead of forcing a switch.
Having the full stack in one place, compounds.