Just-in-time compute
Compute that stops when your code does.
Cirrun finds the cheapest machine on the internet for your work, keeps everything you make safe while it runs, and turns the machine off the second you're done. You never lose a run. You never pay for idle.
03:12 A rented GPU, somewhere
It's late. You go to sleep and trust the machine.
Hours of training. Weeks of ideas. All of it running on a computer you rented from someone you'll never meet.
Hour 6 of 9 progress saved on the machine Hour 7 of 9 progress saved on the machine Hour 8 of 9 Connection lost. The machine is gone. Everything saved on it is gone too.
03:47 The morning after
In the morning, it was gone.
The machine. The training. Every step it had saved. Only one thing was still there:
The bill−$34.74
This happened to our founder, more than once, while competing against some of the best AI teams in the world. Everyone who rents computers has a night like this. Cirrun exists so you never have another one.
04:20 Everywhere at once
And all over the world, people pay for machines doing nothing.
Left on overnight. Waiting for the next job. Bought for a busy week that came once. Rented by the hour, used for minutes. Every amber light is someone's money burning while nothing happens.
05:00 Then Cirrun
So we built the opposite.
- It finds the cheapest machine for you.Across every cloud we work with, every time, in seconds. You see the price and your limit before anything starts.
- It keeps your work safe the whole time.Everything your code makes is copied to your own storage while it runs. If a machine dies, you lose seconds, never hours.
- It turns the machine off the moment you're done.Not in an hour. Not tomorrow morning. The second your work ends and everything is safely copied.
- It shows you exactly what you paid.Every run ends with a receipt: the time used, the real price of the machine and our fee. Nothing hidden.
05:30 Every kind
Not just GPUs.
Any computer power you need.
- GPUsFor AI, from top data-centre chips to gaming cards
- ProcessorsA sliver of one or hundreds at once
- MemoryExactly as much as the job needs
- StoragePay only for the space you use, while you use it
- ServersThat exist only while they're working for you
- Rare hardwareMachines almost nobody rents by the second
Mix and match any of it. One simple bill.
06:00 Everyone
And anyone can earn from it.
Cirrun Host pays you for hardware you already own. A gaming PC in Lagos. Half-empty server racks in Bavaria. Hard drives nobody filled in Dubai. Plug it in, and while you sleep it does real work for people around the world. Cirrun checks every job and pays you for every second.
06:30 Smarter every run
It gets smarter
every time you use it.
- Know the cost before you start how much it will cost and how long it will take, up front
- Never start over if a machine fails, your work carries on from where it was on the next one
- Stops waste for you spots stuck jobs, idle machines and oversized setups, and fixes them
- Free top-tier AI models from Quantavane, our sister lab, ready to use on day one
- Safe for AI agents let your AI tools rent computers for you, with hard spending limits
- Greener and private runs on cleaner power when it can, and a sealed tier for sensitive work
07:04 Dawn
Every machine on Earth,
one command away.
The cheapest machine, ready fast, your work never lost, honest to the cent. That is where Cirrun is going. Built in Dubai, for everyone.
Why we're building this
I started training AI models for competitions on rented GPUs. More than once, I woke up to find the machine gone, and my work gone with it. Hours of training I couldn't get back. The bill was still there.
It wasn't bad luck. It's how renting computers works today: you pay for the machine, not for your work, and nobody is responsible for keeping your work safe.
So we're building the opposite. Cirrun keeps your work safe the whole time, turns the machine off the second you're done, and shows you every cent, for every student with one GPU and every company with thousands.
Nobody should lose their work to a machine they were paying for.
Made for you
Whoever you are, it just works for you.
Pick yourself. Here is what Cirrun changes for you.
Cut your cloud bill without hiring anyone to manage it.
- Pay for work, not waiting. Machines switch off the second the job ends, so idle hours vanish from your bill.
- The cheapest machine every time, chosen for you across clouds, so your runway lasts longer.
- No surprise invoices. Spending limits on every run and a daily summary of what ran and what it cost.
One command. The cheapest machine. Nothing to babysit.
- Works from your terminal, your notebook or your code editor, with the same command everywhere.
- No machines to set up or remember to turn off. Cirrun starts them, watches them and shuts them down.
- Your results land in your own storage, ready to download the moment the run ends.
A dying machine never costs you a training run again.
- Your progress is saved while you train, off the machine, so a crash is a short pause, not a restart.
- Big GPUs for less, from the cheapest provider that has them free right now.
- Free top-tier models from Quantavane, ready to fine-tune and run.
Learn on real GPUs for the price of a coffee.
- Pay by the second, so a quick experiment costs cents, not a whole hour.
- Hard limits you set, so you can never spend more than you meant to.
- Credits for students and courses as we grow, and guides that show you every step.
Let your AI tools rent computers, safely.
- Agents can start a machine, run the work and stop it on their own, through a simple connection.
- Hard spending limits the agent can't go past, and a receipt for everything it did.
- Nothing left running when the agent forgets: machines stop when the work does.
One place for all your team's computing, with the bill under control.
- Team budgets and rules, so every project stays inside its limit.
- A sealed, private tier for sensitive work, and the option to use your own machines.
- A full record of every run for finance and security. Available for teams through 2027.
Your idle hardware, paying you by the second.
- Gaming GPUs, spare servers, empty drives: if it's idle, it can earn.
- Cirrun brings the work, checks every job and handles the payments.
- Small data centres fill empty racks with steady demand from all over the world.
Why Cirrun is different
Clouds rent you machines. Cirrun runs your work.
Everyone else sells you time on a machine, and what happens on it is your problem. We sell the finished run.
We're building the power grid for computing. See the whole plan.
See it for yourself
What would your GPU bill be?
Pick a GPU and how many hours your work really runs. We price it everywhere we have a published price, including the hours a rented machine sits doing nothing.
| Provider | How you pay | Per month |
|---|
Cirrun's row is our planned price once our provider agreements are signed: the cheapest published machine for that GPU plus our 8% fee, charged only while your work runs. During the alpha you use your own provider accounts and pay us nothing. When someone else is cheaper, this table says so. Every price and its source: GPU price index.
Honest to the cent
You always know what you're paying.
You see the price before anything starts, it can't change while your work runs, and every run ends with a receipt. If a machine fails, you don't pay for the failure.
- A price and a limit before you startOn by default, on every single run.
- One fee, printed on every receiptThe real machine price plus our fee. No hidden markups.
- No surprise bills, everSpending limits, automatic shut-off and a daily summary.
- Machine
- 1 × H100 GPU
- Provider
- cheapest of 5 when it started
- Time used
- 4 h 50 m
- Machine price
- $13.00
- Cirrun fee (8%)
- $1.04
- Your work
- 128 of 128 files safe
- Machine
- turned off when done
Works with what you already use
Wherever you work, Cirrun is one step away.
Private by design
Your work stays yours.
- Your results go to storage you ownNot ours. Take everything with you any time.
- Your passwords and keys stay with youThey are never written into logs, links or receipts.
- Machines leave nothing behindThey are wiped out as soon as your work is safe.
- A sealed tier is comingFor work that must stay private, with proof on the receipt.
Where we're going
Building the way the whole world rents computers.
We start with the promise that matters most: your work is never lost. Everything else follows, step by step.
- AlphaNever lose a run, on GPUs and processors, across 5 providers.
- Public launchOpen to everyone, with notebooks, VS Code, GitHub and AI agents.
- One bill + Cirrun HostPay Cirrun for everything. People start earning from their hardware.
- Teams and companiesBudgets, rules, the sealed tier and greener scheduling.
- Every machine on EarthStorage, rare hardware and anything idle, rented by the second, anywhere.
Part of the Canli family
Free top-tier AI models, and a family that grows together.
Quantavane builds AI models; Cirrun lets you use them for free. Every Canli company, and every company Quantavane starts, runs on Cirrun. As they grow, the grid grows, and compute gets cheaper for everyone on it. Use more than one and you get shared credits.
- QuantavaneTop-tier AI models, free to use on Cirrun. Every Quantavane company runs on the grid.
- CanliCapitalFinancial research and data.
- TraceAxiomChecks that AI agents did what they claim.
- GIMBIXThe cheapest way to use AI models.
Questions
Straight answers.
Is it really cheaper?
For most work, yes, for two reasons. Cirrun picks the cheapest machine that fits across providers, and you pay only while your work runs, so the hours a machine sits doing nothing disappear from your bill. The calculator compares every provider side by side, including when someone else is cheaper.
What if a machine breaks in the middle?
Your work is already safe. While your code runs, Cirrun keeps copying everything it makes to your own storage, and it double-checks every file before any machine is switched off. You lose seconds, not hours.
Do I need to be an expert?
No. If you can run your code on your own computer, you can run it on Cirrun with one command. We pick the machine, start it, watch it and turn it off.
How do you make money?
During the alpha we don't: you use your own provider accounts and pay us nothing. After launch, you pay the real machine price plus a small fee that's printed on every receipt. Teams can pay for extra features. No hidden markups.
Which machines can I use?
The alpha starts with GPUs and processors on Beam, RunPod, Modal, DigitalOcean and your own computer. More providers, storage and hardware from people and small data centres through Cirrun Host follow after launch.
Is my work private?
Your results go to storage you own, your keys stay with you, and machines are wiped as soon as your work is safe.
When can I start?
The first alpha wave opens at the end of November 2026, and the public launch is planned for January 2027. People who invite others get in earlier.
Who is building this?
Cirrun is built in Dubai by a young founder who lost training runs and data to rented GPUs and decided nobody should again. It's part of the Canli family of companies.
Alpha · first wave end of November 2026
Never lose your work again.
Never pay for idle again.
- Get in earlyThe first wave is small, so we can help every person one by one.
- Jump the line by invitingEvery person who joins with your link moves you up.
- Shape what we buildTalk to the founders directly and vote on what comes next.
- Free during the alphaNo Cirrun fee while we're in alpha.
Get early access
Just your email. We'll tell you the moment your wave opens.