One AI builds.
The other reviews.
Plyloft runs Claude Code and Codex on the same card. One writes the change; the other, from a different vendor, reads it read-only and tries to break it. You ship when it finds nothing blocking.
Same model,
same blind spots.
Ask a model to check its own work and it tends to agree with itself. Plyloft hands every build to a Challenger from the other vendor, read-only, with one job: find what is wrong.
- Each finding sits under the line it names in the diff
- Every round is kept: what it raised, closed or left open
- Findings go straight to the Builder; a serious issue also reaches your Inbox
Set up in about ten minutes.
Setup checks what is already on your computer, installs what is missing when you say so, and finds Claude Code or Codex signed in with your own subscription. Nine steps, each one skippable. Click through it.
Welcome to Plyloft
Plyloft runs your projects with AI coding agents. You describe a task on a card, an agent works on it in a separate copy of your code, and you review the result before it lands.
How much have you used Git and a terminal?Do you work alone or with a team?
You can change this later in Settings.
Your AI
Plyloft’s agents are Claude Code (from Anthropic) or Codex (from OpenAI), running with your own subscription. One is enough.
Plyloft never uses an API key and never sees your password. Each tool signs in with its own account, and Plyloft starts it for you.
Git and your name
Git keeps the history of your project. Plyloft uses it to give each card its own copy of the code and to merge the result back.
What is Git?GitHub (optional)
GitHub keeps a copy of your projects online, lets you open pull requests and is how Plyloft teams share work. Without it, everything stays on this Mac and Plyloft ships by merging on this Mac.
What is GitHub?Where your projects live
New projects and projects you get from GitHub go in this folder. Folders you already have stay where they are.
Why one folder?Your first project
Pick one to start with. You can add more any time.
How Plyloft builds “acme-web”
These are the settings Plyloft has for “acme-web”. Change them if something is wrong.
Plyloft is ready
Here’s what you have. You can change any of it in Settings.
From a sentence to a shipped change.
Scroll through one card. Every card runs in its own copy of your repo, so several can be in flight without stepping on each other.
Describe it
New card puts a title field at the top of Needs you. Open the card and write what you want in plain words: both agents work from them, and the Challenger checks the result against them.
Claude builds
Start build opens Claude Code in a real terminal, in the card’s own copy of the repo, on your own plan. Your main branch is untouched. When Claude says it has finished, you press Build done.
Codex reviews
The other vendor reads the build read-only. A serious issue is named at the top of the result, and each finding sits under the line it names. The findings go to Claude by themselves; the next Build done starts round 2.
You ship
When the review finds nothing blocking, the card says Ready to ship. Ship it merges here, or opens a pull request from its menu. Several cards can be in flight at once, each in its own worktree.
Not sure of the approach? Build it twice.
Claude and Codex each build the card in their own copy of the repository, each on the model you choose for this card. The Challenger compares the two and recommends one. You pick.
- Two build cards under Now, each with its own session and Build done
- Each result with its files, summary and the Challenger’s findings, judged by the other vendor
- The build you pick becomes the card’s; its Build done starts the full review
Adds calcRoundHalfUp with ties away from zero for both signs, keeps the 2-decimal rounding exact for values like 1.005, and adds five tests and an npm test script.
The Challenger found no problems in it.Adds calcRoundHalfUp with two passing tests. It rounds 1.005 down to 1, because it multiplies before rounding.
The Challenger’s 2 findings- src/billing.js rounds 1.005 to 1: the multiply loses the half before rounding.
- src/billing.test.js does not cover negative ties.
Every rough card can leave the next one easier.
When you correct a card, the Challenger proposes one change to a skill, citing your own words. You approve it, and every later review follows it.
Your plans. Your machine. Your code.
What one AI on its own cannot give you.
Free for one card at a time.
The whole loop is free. Pay when you want more of it at once, or a team. Prices include VAT.- Builder and Challenger, either vendor
- Pull requests, checks and shipping
- Any number of workspaces
- One card building at a time
- Everything in Free
- Cards in parallel
- Build with both on one card
- Merge when checks pass
- Everything in Pro, for each member
- Shared board on GitHub Issues
- Shared skills and review rules
- From 2 seats, the owner’s included
Shipped in the open.
A new release every few days since May, each built with Plyloft itself. Updates install on their own, on Mac and Windows.
Read the changelog →Questions people ask first.
Is it fine to use my Claude or ChatGPT plan this way?
Plyloft starts the official Claude Code and Codex command-line tools in a real terminal on your machine, signed in with your own account, the same as typing claude or codex yourself. It never sees your credentials and uses no API key. Usage counts toward your plan’s limits as usual.
Do I need both Claude and ChatGPT?
No, one is enough: the Challenger can be the same vendor as the Builder, and the card says so. Two vendors give a more independent review, which is the point, so it is what we recommend.
Does my code go through Plyloft?
No. Your code goes to Anthropic or OpenAI when their tool works on it, and to GitHub if you use it. Plyloft’s server knows your GitHub login, your plan and the device using it; it never receives your code, prompts or file names.
What does Free leave out?
Only parallel work and the extras: one card builds at a time, across all your workspaces, and you can write and plan other cards meanwhile. Both roles, reviews, pull requests, checks, shipping and any number of workspaces are included. Pro adds cards in parallel, Build with both, Merge when checks pass, Ship batch and skill improvement proposals.
Does it run on Windows?
Yes. Windows gets the same features as the Mac in every release, with automatic updates.
Can I cancel?
Any time, from the billing portal. You keep your plan until the end of the period you paid for.
Do I need an account?
A GitHub account, free to create, on every plan. Signing in to Plyloft with it is your Plyloft account: identity only, with no access to your code or repositories. Your repositories can stay on your machine.