a reviewer
A diff read twice before merge.
whyOne model can miss what another catches.
asksHave Codex review this diff, then check what it says.
instibrain / works with / codex
A second AI, same folder, same brain.
OpenAI's coding AI on your ChatGPT plan, beside Claude Code. It reads the same brain through AGENTS.md.
personal/tools/codex/codex.md your plan, your setupasked · an exampleGet a second read on the migration plan before I run it.
the AI, from the pageGPT says the backfill must run before the constraint. I checked the schema: it is right, so I moved the step.
who uses it, and why
Who calls in the second AI.
Two models, one brain, for work that needs a second look.
a founder
A cover image, drawn to your brand.
whyThe launch post needs an image today.
asksDraw a cover for the launch post from our brand page.
a developer
A long task, in a window of its own.
whyA long task would block your only session.
asksOpen a Codex window for the test fixes.
what you can ask
Your AI hands Codex a second read, an image, or a window of its own.
When your Claude limits run out, work goes on: same folder, same pages, another model.
Run a session
In your instibrain folder, reading the brain through AGENTS.md.
Take a task from Claude Code, with no window
codex exec, with the prompt before any image flag.
Draw images to a brief
Claude writes the brief from your pages. Codex draws it.
A second opinion
A GPT read on a plan, page or diff, then checked.
A window of its own
Its own title and background colour, beside the others.
where the credential stays
Your ChatGPT plan is the sign-in.
Codex's limits are its plan's. Its sign-in is yours.
on your computer
Installed with brew or npm, signed in with ChatGPT. Every image is checked before anyone sees it. Logo and copy go on afterwards.
in your brain
A page names your plan and setup. The builder writes AGENTS.md.
what instibrain never touches
Your AI checks every Codex answer before it acts on it.
Codex runs under your sign-in. Our server stores only your setup page.
- Your OpenAI sign-in.
It stays on your computer. No page carries it.
- An unchecked answer.
It changes a plan only after a check against the source.
- Generated text or a logo in an image.
They come out wrong, so both go on afterwards.
the limits
Codex stops where your ChatGPT plan's caps stop it.
Both come with your ChatGPT plan.
| Limit | What it means |
|---|---|
| Its limits are its plan's | GPT models, with your ChatGPT plan's caps. |
| An image costs about a minute and 30,000 tokens | Each one counts toward your plan's caps. |
before you connect it
Codex reads your brain beside Claude Code, on your own plan.
Is Codex part of instibrain?
No. It is OpenAI's AI on your plan. You can also connect it over MCP.
Which AI runs the tools?
Claude Code runs them. Codex adds second reads, images and another model.
Does Codex see a sealed space?
On an admitted computer, as files. Over MCP, open spaces only.
Claude Code and Codex work from one folder and one brain.
A second model starts from the same map on day one.