XCR

Guides

Codex usage limit: make your Codex quota last longer by handing the heavy work to XCR

Updated

Short answer

Codex runs on your ChatGPT plan's usage limits, and the heavy part of the job, multi-file builds with rounds of test fixes, is what burns through them. Keep Codex for planning, questions and review, and hand the heavy implementation to XCR. XCR builds the change, runs your checks and returns a short report, so your Codex quota goes to the work that needs Codex and lasts longer.

The Codex usage limit, in plain terms

Codex is included with ChatGPT plans, and every plan comes with usage limits. According to OpenAI's Codex help article, usage depends on the model, where the task runs, task complexity, context, reasoning, speed and tools, and a long-running task can use substantially more than a short request. On plans with shared allowances, Codex draws from the same pool as other agent features such as ChatGPT Work. Usage is tracked in a five-hour window and a weekly window.

The exact numbers vary by plan and change over time, so check ChatGPT pricing for your plan's current limits.

What actually burns through your Codex quota

Running out in the middle of a feature is the real pain: the plan is clear, the code is half written, and the window is gone. The usage goes to a few predictable places:

  • Implementation loops. Write code, run the tests, read the failure, fix it, run again. Every round re-reads the context.
  • Long threads. Everything Codex has read and written so far rides along with each new turn.
  • Repository exploration. Opening file after file to understand a codebase before the first line is written.
  • Heavy reasoning on routine work. The strongest models and highest effort settings use the allowance fastest.

Planning a change and reviewing a diff are light by comparison. The build is the heavy part, and that split is the key to making your quota last.

How to check where you stand

Open Settings or your usage dashboard in ChatGPT or the Codex app to see which allowance you've used, your credit balance and the reset time. In an active Codex CLI session, enter /status. When a limit is reached, the limit notice shows the options available to your account: adding credits, using an available reset, upgrading, or waiting for the reset. Starting a new chat does not reset your limits.

Split the work: Codex plans, XCR builds

XCR is a plugin that connects to Codex and takes over the expensive part of the job:

  1. Codex hands the task over. Any request bigger than a small edit (roughly more than 30 lines, or more than one file) goes to XCR.
  2. XCR builds it. It writes the change on a copy of your project, runs your project's own checks and repairs failures.
  3. Codex reviews. It gets a short report with the requirements done, files changed and check results, and checks the change against your request.

XCR routes each task to models from OpenAI, Anthropic and Google, so one account covers all of them. Checks run on your machine. Code is sent to the model only for the task, and XCR doesn't store it. Quick questions and one-line fixes stay in Codex, exactly as before.

Your Codex quota now goes to the parts that need Codex: understanding what you want, deciding the approach and reviewing the result, while XCR takes on the heavy build work.

Want more habits that stretch each window? Read how to get more usage on Codex and 8 ways to make AI coding cheaper.

How to set up XCR in Codex: install and first task

What you need

  • Codex installed and signed in on a computer running macOS or Linux.
  • Node.js 20 or later. If it's missing, Codex installs it for you and asks before it does.
  • A project in a folder on that computer. A git repository works best.

Step 1: Create your XCR account

Open the install page, enter your email and choose Email me a sign-in link. The link arrives in seconds and works for 15 minutes. There is no password, and the same link creates your account the first time.

Step 2: Connect XCR to Codex

  1. Open Install on a computer and choose the Codex tab.
  2. Read What XCR will change on your computer and tick the box to agree.
  3. Click Copy install instruction, paste it into Codex and allow the command when it asks.
  4. The command downloads the plugin from xcr.ai and checks its SHA-256 before it runs anything. Your console switches to Connected.
  5. Restart Codex.

XCR registers itself as an MCP server in ~/.codex/config.toml and adds one short rule to ~/.codex/AGENTS.md, so Codex hands bigger changes to XCR on its own. Your original files are backed up first, and every change sits between clearly marked XCR lines. The instruction works once, for the next 10 minutes; if it expires, copy it again.

Step 3: Hand off your first task

Once your console shows Connected, ask Codex for something bigger than a small edit, as you normally would:

  • "Add CSV export to the orders page, with filters by date and status."
  • "Add unit tests for the payment module and make them pass."

Codex hands the job to XCR, and the task appears under Tasks in your console.

Step 4: Review the result

XCR builds the change on a copy of your project and runs your project's own checks (type check, lint, tests). Codex gets back a short report: requirements done, files changed and check results. When the checks pass, the change lands in your working tree without a commit, ready to review in your usual diff. Anything that needs the internet, such as installing packages, comes back for Codex to finish.

Pause XCR anytime with the switch at the top of Overview, or tell Codex "Uninstall XCR from this computer" to remove exactly what XCR added. More detail: XCR quick start and install XCR in Codex.

Keep Codex. Hand the heavy work to XCR.

Connect it to your AI coding tool in about a minute.

Install XCR →

Questions

What is the Codex usage limit?+

It's the allowance your ChatGPT plan gives Codex, tracked in a five-hour window and a weekly window. The numbers vary by plan; your usage page and /status in the Codex CLI show what's left and when it resets.

Does XCR use my Codex usage limit?+

XCR runs its tasks on its own model access. Codex keeps the conversation, the hand-off and the review, so your Codex quota goes to planning and review and lasts longer.

Which tasks does Codex hand to XCR?+

Changes beyond a small edit: roughly more than 30 lines or more than one file, such as features, tests, bug fixes, refactors and migrations. Quick questions and one-line fixes stay in Codex.

Is XCR an OpenAI product?+

No. XCR is an independent plugin. Codex is named only to describe compatibility.

More guides

All guides