carlosframework.com
The Story of CARLOS
From a ten-second cold start to a global deployment platform.
→ or space to begin
Where it started
lpay.fly.dev
A Rails app on Fly. Ten seconds to wake from hibernation.
But it worked: armed with a Stripe terminal, the kid made a few hundred euro selling 3D prints.
Score.
The next problem was chat.
The same kid ran an iMessage group with his friends — and the Android kids couldn’t join. I threatened to make an app.
— the kid, after lpayI’m starting to think you could do the chat app after all.
> I want to make a chat app. It's kid friendly. It doesn't require any kind of log in: just a secret link to initialise, and when I send someone the link, they get to choose a screen name. I can deploy it easily and run it on a single server. Should it use sqlite? I want all the messages to be encrypted, so that if someone compromises the server, they can't access anything.
Three weeks later
Eleven
elevenmessenger.comKid-friendly, end-to-end encrypted, no accounts. A secret link and a screen name.
Everyone kept looking at the chat app.
The interesting part was a session from July 9.
> How do we make deploying home and the oauth broker apps essentially infinitely scaleable?
Out of that question
A philosophy, then an architecture.
11factor.org
a philosophy for building software responsibly with LLMs.
carlosframework.com
the architecture built on it.
Run it in the open
Operations as an open ledger.
An append-only, verifiable record of what the service is actually doing — not a status page that says “all systems operational”. A direct consequence of 11factor.
And run it anywhere.
When highly-available apps are this cheap and this simple to operate, on-prem stops being a compromise.
Your hardware. Your data. The same availability.
Remember this one.
People were skeptical.
“What’s the point?” Fair.
So: prove it across every shape of app — E2EE, single-process, single-database, multi-region, multi-platform.
Proof, part one
Keymail
keymail.devEmail, but secure.
Suddenly it seemed feasible to replace the SaaS we were paying for.
Secure. Affordable. Ours.
Proof, part two
Slopbox
slopbox.cloudLike OG Dropbox, with no upsells. Simple, fast, encrypted.
Proof, part three
Woodstar
woodstar.appMastodon: too nerdy. Bluesky: URLs where handles should be.
Woodstar: each profile its own database — and it feels like OG Twitter.
Meanwhile
Tito was being rebuilt in this shape.
Vicky, product manager, hit a productivity unlike anything I’d ever seen — UX-first on a resilient, dynamic stack built for extremely fast iteration.
Proof, part four
Kass
kass.trainingMy gym log, shared securely with my personal trainer — not with a US data firm.
Time to extract
Rastrillo
rastrillo.orgWith the examples in hand: the web framework underneath all of them, pulled out and given a name.
And keep proving it
Seapointish
E2EE, self-hostable financial reporting — we know the seapoint.co founders, so it makes a good demo.
Next on the list: porting Vito (vi.to).
The tally
Six weeks in:
Eleven built in three weeks. Keymail, Slopbox, Woodstar, Kass the week after.
A philosophy. An architecture built on the philosophy. A reusable web framework on top of it all.
Apps that cost pennies per month — in the cloud or on-prem. The architecture doesn’t mind.
But there’s one more thing…
What if deployment was the product?
I’d been creating a fresh AWS account for every app.
What if a properly global platform orchestrated it all — deploys in seconds, bring your own box, instances placed anywhere in the world, per tenant rather than per app?
The CARLOS platform, flagship
Carloku
carloku.comInspired by Heroku — and everything Heroku wasn’t: truly multi-region; your boxes, Carloku’s orchestration; or run the whole platform yourself.
One more thing…
Doesn’t GitHub kinda suck these days? Bloated. Slow. AI bolted on.
Amadan
amadan.netBlazing fast. Cloud or self-hosted, first-class. E2EE optional. Familiar concepts, fresh ideas — multiple conversations around a single branch. Shock, horror.
One more thing…
Eleven · Keymail · Slopbox · Woodstar · Amadan