Canaryflux is a Cross-device QA scanner for modern apps. We exist because the gap between "ships fast" and "renders correctly across the device profiles your customers actually use" has never been wider — and the existing tools weren't built for the way teams ship now.
Modern frontends ship in minutes. A Lovable prompt, a Cursor edit, a Vercel deploy — the whole loop can happen between two cups of coffee. But the QA layer hasn't kept up. Lighthouse runs in a desktop Chrome. Cypress is brittle and lives in your repo. Real-device test farms cost a fleet, and a fleet needs a team.
So most teams just… don't. The bug ships. A customer DMs you a broken signup screenshot. You curse, fix, deploy, repeat.
Two things shifted at the same time: vision models got good enough to read a UI screenshot like a human, and headless browsers got cheap enough to run a cross-device matrix on demand.
Put them together and you have a scanner that doesn't need a checklist, doesn't need test scripts, and doesn't care whether your code was written by a person, an LLM, or a no-code drag-and-drop. It looks at what the user sees, asks an AI what's broken, and tells you in plain English.
A QA scanner that runs across device profiles on demand, whenever you paste a URL. It captures, detects, verifies, and ranks — same pipeline whether you're shipping a Next.js app, a Bubble project, or whatever your AI assistant just spat out.
The output is a ranked list of findings you can hand to a non-technical co-founder. The input is a URL.
These aren't aspirations. If a release breaks one, it doesn't ship.
We tune the detector and reproducer against our own seeded test pages before shipping changes that touch model prompts or device profiles. We'd rather hold a release than ship a noisier scan.
Every piece of DOM that touches the LLM passes through mask_pii first: emails, phone numbers and anything that looks like an API key are masked before an AI model reads the page's text, console or network logs. Screenshots go to the models as they are, and our privacy policy says so.
The bug report should make sense to someone who's never opened DevTools. Severity, what broke, where, on which device. That's it.
Scan time scales with device count and page depth — typically 1-5 minutes on Free's 3 devices, longer on Studio's 17. If the median creeps past that for a given tier, we treat it as a regression and fix it.
You don't install anything. You don't write tests. We crawl externally, like a real visitor would. If you change frameworks tomorrow, nothing on our side changes.
Cancel from the dashboard's billing page, through Stripe. No phone call, no exit-intent popup. If it stops being useful, leave clean.
Canaryflux is live, and it works with any site that renders HTML: the scanner doesn't care what built the page.
Made with care in Mexico. The highlights of what we've shipped are in the changelog, and you can read exactly how we handle data.
Farid Islas
First-time solo founder · Mérida, Yucatán
I built Canaryflux because I kept shipping side projects that looked perfect on my MacBook and broken on every other device my friends owned. Every QA tool I tried either wanted me to write tests, install a snippet, or pay for a fleet of real phones I couldn't justify. So I built the version I wanted: paste a URL, get findings.
Canaryflux is a sole-proprietor business registered with Mexico's SAT under the Persona Física con Actividad Empresarial regime. There's no team yet — every line of code, every email reply, and every line of documentation is mine. That's a feature and a constraint: features stay in scope, bugs get personal attention, and decisions move at the speed of one head.
Paste a URL you own or are authorised to test — we'll crawl it across our device profiles and report ranked findings. No card, no SDK, no waiting list.