Pinned Tweet
Check it out at portal.clusy.io
The core mechanism of Portal will also be presented at @AcmMiddleware 2026 in Spain in December. Feel free to say hi to us!
Portability is becoming more and more essential with the abundance of providers.
It is a guarantee of compute independence of the future.
In case you’re curious how Portal work, you need to check this out!
Clusy Inc. retweeted
Every machine on Earth that runs Python is about to become one computer.
Compute has borders. We just removed them.
Portal moves a running Python job between any two machines. Any cloud, any GPU, any CPU. Mid-run.
Everything before hit the wall, so we built a portal.
We are launching Portal.
It makes Python jobs portable across clouds, hardware, and virtually anywhere.
- Jobs survive any preemptions & OOMs
- 40-60% lower compute costs on average
- Works across providers → no lock-in
- Built for your coding agents
Clusy Inc. retweeted
This guy is basically creating free Python jobs.
Genuinely insane.
Kimi K3.1 Pro + Highspeed are live in Clusy ⚡
This interactive 3D pavilion was built with K3.1 Pro in one shot.
Try it → app.clusy.io
Need different compute halfway through training?
Switch the machine. Keep your variables, files, and progress.
Clusy Fluid Compute.
New compute. Resume work.
Clusy Inc. retweeted
As a software engineer and data practitioner, I’ve always felt the gap between how much agents helped me in the former and how much slower progress feels in the latter.
So I looked at the numbers behind that feeling, and the benchmark gap is real.
Jupyter notebooks haven't changed in a decade, Your research has
Describe the experiment, Branch three architectures at once in the cloud with fluid compute
Accelerate your research
i've got some new superpowers
Turns out not having a terminal makes people creative.
In two months, we saw 15+ ways to run shell commands through notebook cells, scripts, and the agent.
Now that we had enough fun, we are adding this simple feature as per many requests.
@clusyio has a terminal now!
Clusy Inc. retweeted
What if data science tools were built around exploration instead of execution?
Interesting problems rarely have one obvious next step.
As agents and compute improve, workflows should explore more of the space instead of forcing everything down one path.
Clusy Inc. retweeted
What if your code could execute anywhere, without its live state belonging to any one machine?
We built SOMA to explore a future where compute is fluid and execution is fully portable.
This helped us cut costs by up to 70% across 10K+ AI/ML workloads.
Toomre & Toomre (1972) showed that galaxy tails depend on disk spin.
We reran their restricted N-body experiment in Clusy with 2 million stars, then forked the same collision two ways.
-Prograde: 28.7% of stars entered tidal tails.
-Retrograde: 0%.
Same encounter. Different spin. Completely different outcome.










