Agentic coding, on your codebase

Fix bugs. Ship faster.
Your codebase, understood.

Connect your code, describe what you need in plain English, and let an AI agent read, edit, and test it — live. Review the diff, then download.

No credit card. Runs on an isolated copy — your real files are never touched until you say so.

checkout.py cart.py models.py test_checkout.py
1def checkout(cart, user):
2 total = 0
3 for item in cart:
4|
5 return charge(user, total)
6
7# agent verified with tests ✓
Live activity
Fix the crash when an item has no price in checkout.
✓ read checkout.py
+ total += item.get("price", 0)
✓ tests pass

You're already paying for AI every month.

Codex AI does the part that actually matters — reading your code and fixing it — and charges like a tool, not a seat.

It works on the real thing

It opens your repository, reads the files that matter, edits them and runs your tests. You watch every step as it happens, and nothing reaches your real files until you download the change.

₹20/day

Pro is ₹600 a month, billed monthly, cancel whenever. No seats, no per-request billing, no card to start — about what one evening of hunting a bug costs you in time.

In the works

Run out? Earn your way to more

We are building a way to top up your daily allowance without paying again — watch a short video between runs and the credit goes straight back into your compute. Entirely optional, and it will only ever add to what you already have.

Built to actually fix real code

Not autocomplete. A real agent that reads, edits, and verifies across your whole project.

Understands your repo

Searches and reads across files to grasp context before it changes a single line — not blind edits.

Live activity feed

Watch every step as it happens — planning, commands, edits, and test results, streaming in real time.

Safe by design

Runs in an isolated sandbox. Nothing touches your real files, git, or accounts until you approve.

Full diff review

Every change is shown as a clean git diff. You decide what to keep — nothing is hidden.

Verifies its work

Runs your tests after editing, so fixes are checked — not just guessed at and left for you to find.

Budget guardrails

Hard spend caps and loop protection are built in, so a session can never run away with cost.

How it works

Three simple steps — no setup, no configuration.

1

Connect your codebase

Drag in a folder or a .zip. The agent works on an isolated copy, so your real files stay untouched.

2

Describe the task

"Fix this bug", "add validation", "explain this module." Plain English is all you need.

3

Review & download

Watch it work live, see exactly what changed as a diff, then download the updated code.

See Codex AI in action

Watch an AI agent find and fix a real bug from a single prompt.

codexai --agent
$codexai agent "Fix the crash when an item has no price"

Start free. Upgrade when it pays off.

Simple plans. Cancel anytime.

The price you see is the price you pay.

Monthly Yearly 2 months free

Free

See if it understands your code

0/mo
  • Ask anything about your codebase
  • It reads the whole project, not one file
  • 3 questions a day
  • No card. No trial clock.

Lite

For the code you inherited

100/mo
  • 25 questions a day — map a whole project
  • "Where does this break?" "Why is this here?"
  • Answers cite your actual files
  • Read-only — it explains, never edits

Pro Max

For the weeks you ship all day

1000/mo
  • Everything in Pro, twice over
  • Double the machine time — big installs, long test suites
  • Refactors that touch 28 files in one go, not 15
  • Built for client work and deadline weeks

Costs less per month than one hour of a contractor's time.

Start on Free — no card, no countdown. Upgrade the day it saves you an evening, and cancel in one click if it doesn't.

Ready to fix your code?

Connect a codebase and give the agent its first task in under a minute.