Cartlift · Ecommerce growth · CRO
Audit a store, test the fixes, ship the winner.
- 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%
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.
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.
How it works
- 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.
- 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.
- 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.
- 04
Let the allocator decide
A scheduled command re-weights traffic with Thompson sampling, and pins the winner once it is confident.
Wake up with
glass skin.
↻ free shipping over $40 · 14-day returns
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
Stack
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
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.
Want Cartlift running for your team, or something like it built?