Cartlift · Ecommerce growth · CRO

Audit a store, test the fixes, ship the winner.

Paste a store URL and get an annotated four-lens audit (CRO, SEO, trust, Google Merchant) in about 30 seconds. Cartlift drafts variants, ships them through a 3 KB snippet, and a Thompson-sampling allocator pins the winner, so there is no dashboard to babysit.
Status
Live · v0.1
Licence
MIT
On GitHub
★ 2 · ⑂ 0
four-lens audit of a store page
~30 s
snippet that ships the variants
3 KB
confidence, plus 500 samples, before a winner is pinned
95%
01

Why it exists

Conversion tools for online stores are mostly closed platforms. Cartlift is an open-source one you can read, run and change: paste a store's address and get four reports in about 30 seconds.

After the audit it closes the loop. It drafts page variants from the findings, serves them through a small script on the store, and moves traffic toward whichever one converts better.

02

What it does

  • Conversion audit

    Friction on product, cart and checkout pages, the hero copy, the contrast of the add-to-cart button and social proof, each with a predicted lift.

  • SEO audit

    Titles, meta tags, Product schema, heading order and render-blocking resources, scored from A to F.

  • Trust and policy audit

    Privacy, terms and returns pages, contact details, payment methods and checkout transparency, against a 14-item checklist.

  • Google Merchant audit

    The checks behind Merchant Center suspensions: misrepresentation, prohibited content, and whether shipping and returns match.

  • Variants and A/B tests

    Each finding becomes two or three candidate variants, drafted by Claude. The store loads one script of about 3 KB, which picks a variant for each shopper and reports views and conversions.

  • A winner without a dashboard to watch

    Thompson sampling moves traffic toward the better variant. When the leader has at least 500 samples and 95% confidence, it gets all the traffic.

03

How it works

  1. 01

    Paste a store address

    One request runs all four audits at the same time and returns a report with four tabs that you can share. Reading it is free and needs no login.

  2. 02

    Draft and approve variants

    Sign up to claim the report into a dashboard, generate variants from the findings, and approve the ones you want to test.

  3. 03

    Ship them with one script

    Add the snippet to the store. It gives each shopper a variant and sends a view event and a conversion event. If anything in it fails, the original page renders untouched.

  4. 04

    Let the allocator decide

    A scheduled command re-weights traffic with Thompson sampling, and pins the winner once it is confident.

glowly.kr / products / glass-skin-serum
GLOWLYNEWSKINSHOP
★ bestseller·★★★★★12,403 reviews

Wake up with
glass skin.

$32$45

↻ free shipping over $40 · 14-day returns

cartlift · pdp headline (v3)+14.2%
~/cartlift/experiments● live
pdp · hero headline · v07+18.3%winner
checkout · button copy · v03+11.4%winner
cart · sticky upsell · v02+7.1%winner
homepage · video autoplay−2.1%killed
footer · trust + reviews+0.3%trial
conversion · 30d+38.3%
baseline 2.81%→ 3.89% ↗
cartlift · 1 winner shipped this weekv0.1.0
04

Run it yourself

The API and the database run in Docker; the web app runs on your machine. Without a Claude key every audit returns a fixed sample report, so the whole flow still works.

$ git clone https://github.com/codewithmuh/cartlift.git$ cd cartlift$ cp .env.example .env$ docker compose up -d --build$ cd web && npm install && npm run dev
  • Web app: http://localhost:3050
  • API: http://localhost:8050
05

Stack

Web
Next.js 15React 19TypeScript
API
Django 5.1Django REST FrameworkSimpleJWT
Database
PostgreSQL 16
AI
Claude (your own key)
Local
Docker Compose
06

Where it stands

Built

  • Four audits in one report you can share
  • Variant drafting, the store snippet and the experiment dashboard
  • The Thompson-sampling allocator, and automatic pinning of the winner

Not yet

  • Audits run inside the request today; a job queue is the next step for larger volumes
07

Questions

Which store platforms does it work with?
Any store whose pages can load a script: Shopify, WooCommerce, BigCommerce or a custom storefront.
Do I need an AI key?
Not for the demo: without a key the audits return a fixed sample report. With an Anthropic key the audits and the variants are written for the store you give it.
How does it pick a winner?
With Thompson sampling. The experiment is pinned to the leader once it has at least 500 samples, 95% confidence and a positive uplift.
Is it free?
Yes. The code is MIT licensed.
→Work with me

Want Cartlift running for your team, or something like it built?