Get the weekly radar
One short email with the freshest products, every week.
Clusy plans and runs ML notebook cells from a plain-language description, with real editable code.
Grounded in available product and source data
An agent that writes ML code but hides how it got there isn't much use to a researcher who needs to trust the result — Clusy is agent-native ML notebooks with full code control specifically, so a description in natural language turns into real, editable notebook cells rather than an opaque output.
Once a desired outcome is described, the agent plans the work, sources data, selects a model architecture, and executes an end-to-end pipeline, but every step of that stays visible as an actual notebook a user can inspect, edit, or take over manually at any point. Forking lets a researcher branch a notebook into a separate experiment and compare results side by side, which matters for the kind of iterative testing ML work usually requires rather than committing to one approach upfront.
Data doesn't have to already live in one place either — files can be uploaded directly, public datasets pulled from Hugging Face, or a connection made straight to a Databricks or Snowflake warehouse. Model choice is flexible too: depending on plan, the underlying model can be set to Auto or picked explicitly from DeepSeek, Kimi, Claude, or GPT, rather than being locked to one provider regardless of task.
Compute is where the tiers really differentiate: Free runs on CPU only (8 vCPU, 8 GB RAM), Plus steps up to entry GPUs like the T4 or L4, Pro reaches mid-tier A100s, and Max unlocks H100/H200 GPUs with up to 141 GB of VRAM — a real range from prototyping to serious training runs. All three paid tiers include a usage-based allowance with optional pay-as-you-go beyond that, so a spike in compute needs doesn't necessarily mean jumping to the next tier outright.
All of it — the agent's work lands in a real, editable notebook with full code-level control, so a user can inspect, modify, or take over any step rather than trusting an opaque output.
Yes — forking a notebook creates a separate branch for a new experiment, so two approaches can run and be compared side by side rather than overwriting the original attempt.
No — a live connection to the warehouse is supported directly, so the data can stay exactly where it already is rather than requiring an export or migration step first.
None — Free runs on CPU only, with 8 vCPU and 8 GB RAM; GPU access starts at the Plus tier with entry-level cards and scales up through Pro and Max for heavier training workloads.
Yes, depending on plan — options include an Auto setting or picking explicitly from DeepSeek, Kimi, Claude, or GPT, rather than being locked to a single provider for every task.
These 1 comments were collected from the discovery source. ProductVisit does not host replies in Release A.
Clusy is finally live! Sign up today for free, try our platform, and let us know what breaks. We built Clusy for ourselves with a lot of care, and are now launching to share it with the world. If you are using Jupyter or Google Colab (or any other Python notebook) in your work, we believe that Clusy can help optimize your productivity at almost no friction or migration cost. We started building because the way people work in notebooks has not really caught up with what AI now makes possible. Jupyter and Colab have been amazing, but we think the next generation of notebooks should feel much more goal-driven, collaborative, and agent-native. Therefore, Clusy is built around the end-to-end approach to data science - our agent works alongside you to help across the whole pipeline, starting from your idea and environment setup to actual model training and deployment. Sign up for a free plan or use CLUSYLAUNCH to get 50% off for the first 3 months!
2 source upvotes