Launched this week
Clusy
AI notebook platform for modern data science
165 followers
AI notebook platform for modern data science
165 followers
Clusy is an agent-native notebook platform for researchers and data teams to build, branch, run, and evaluate ML and data science workflows in the cloud. Describe a goal in natural language, and Clusy plans the workflow, sources datasets, preprocesses data, runs parallel experiments in replicated kernels, compares model architectures, and helps produce optimal models through a human-in-the-loop notebook experience.
Free Options
Launch Team / Built With
Curious how the "agent-native" piece actually plays out in practice, does Clusy suggest next steps on its own or do I still have to drive the branching and comparisons? Also wondering what happens when an experiment fails halfway through a parallel run.
Clusy
@mahirzatizdkuj you can drive it yourself or let Clusy handle it - whichever you prefer. If the experiment fails, Clusy can recover and continue from the same spot in another sandbox
How does Clusy actually decide when to branch a workflow versus just continue in the same notebook, and can I override those branching decisions with my own logic?
Clusy
@kymetrayw usually when you mention multiple approaches or parallel execution, it will decide to do those in separate branches. You can always step in and change that, of course
how does branching actually work under the hood, can i diff notebook states like git commits or is it more of a workflow level fork
Clusy
@sefakrbaycq99, it's kind of similar to git, although not fully. It is more of a work-flow level fork, but we do replicate your whole kernel state pretty much
The replicated-kernel branching is the part I'd want pricing clarity on before going deep. If I fan out into say 5 parallel branches to compare architectures, am I paying for 5x compute running concurrently, or does the platform queue/throttle branches on the free and paid tiers? Trying to figure out if this is something I can use for quick exploration without watching a meter the whole time.
Clusy
@galdayan Hey Gal, thanks for commenting!
Short version: no, fanning out into 5 branches doesn't spin up 5 machines. The way it works is one sandbox runs multiple kernels, and each branch gets its own kernel inside that same box. So five branches is five lightweight Python processes sharing one sandbox, not five separate instances you're paying for in parallel.
Creating a branch is cheap too. It's a copy-on-write fork. We snapshot the parent's state (variables, imports, loaded data and models) and restore it into the child kernel, so you're never re-running all the upstream work five times over. Each branch is its own isolated process, which is what lets you run them independently or stop one without disturbing the others. Branches sitting idle for comparison cost nothing.
For the quick exploration you're describing, comparing a handful of architectures, that all runs on CPU and you can fork away without watching a meter. The real compute cost only shows up on heavy GPU training, and running several of those in parallel is where you'd genuinely spend proportionally more, but only while they're actually running, not for having the branches open.
Let me know if you have any feedback and please let us know!
the branching idea for parallel experiments is genuinely clever, finally feels like a notebook that respects how research actually flows instead of fighting you through a linear script.
Clusy
That's what we spent a lot of time building, and it feels so good to have someone recognize that. Thanks so much @alparslan273712!
EverTutor AI
Love the direction. Curious—what's been the biggest challenge in making multiple AI agents actually share context reliably?
Clusy
@suryansh_tiwari2 the limitations of the context window definitely!
Ran a small branching experiment on a tabular dataset and Clusy handled the parallel runs without me babysitting kernels, which was a nice surprise for a notebook-style tool.
Clusy
@hayriye29240360 thank you for the feedback! Glad to hear you found it helpful :)