Clusy
Drive agents to write and run Jupyter notebooks in the cloud with natural language, while keeping full control of the code.
About Clusy
Clusy is an AI IDE for Jupyter notebooks: describe your goal in natural language, and an agent writes and runs cells on cloud CPU/GPU; results remain inspectable, editable, and rerunnable, with every step kept in a real notebook and code control in your hands. Run experiments in parallel branches and share snapshots; any run can generate a link others can open or fork. Import and export .ipynb and publish in one click to Hugging Face, Kaggle, or GitHub. Upload data or pull public datasets, or connect directly to Databricks and Snowflake for live tables. Compute ranges from a free CPU sandbox to H200-class GPU, with models tiered up to Claude and GPT.
Best for
Best use cases
Describe a goal in natural language and let the agent retrieve data, choose the architecture and compute, and execute end to end
Fork an experiment into parallel runs, compare them side by side, and continue with the best one
Import an existing .ipynb into a cloud GPU sandbox, then publish it in one click to Hugging Face, Kaggle, or GitHub
Connect directly to Databricks or Snowflake to analyze and model live data tables in a notebook
Compare before choosing
The official comparison page says Jupyter and Colab provide a blank editor where you do the work yourself; Clusy is a notebook-centered IDE whose agent can collaborate on data retrieval, architecture and compute selection, and end-to-end execution, while also offering experiment branches and snapshot sharing. The output remains a readable, editable, exportable real notebook.
Features
Natural-language agent
Describe a goal in natural language; the agent plans the task, writes and runs cells on cloud CPU or GPU, and leaves results inspectable, editable, and rerunnable.
Experiment branches and snapshot sharing
Fork an experiment into parallel runs, compare them side by side, and generate a stable link for any run that others can open or fork.
Open .ipynb compatibility
Import and export .ipynb, publish in one click to Hugging Face, Kaggle, or GitHub, or download and open it in Google Colab.
Cloud compute and model tiers
From an 8 vCPU / 8 GB RAM CPU sandbox to an H200-class GPU with up to 141 GB VRAM; models range from automatically routed Auto to Claude and GPT by plan.
Pros
- Describe the goal in natural language and let the agent handle data retrieval, architecture and compute selection, and end-to-end execution, with every line of code visible and editable
- Run experiment branches in parallel, compare them side by side, and create snapshot links that others can open and fork
- Freely import and export .ipynb, and publish in one click to Hugging Face, Kaggle, or GitHub without locking in your work
- The free tier includes the Auto model with no credit card, and compute can scale from a CPU sandbox to H200-class GPU
Cons
- Free-tier sandboxes are released after 3 days of inactivity, although projects and saved states can still be restored
Pricing Plans
Free
- Auto model
- 8 vCPU / 8 GB RAM CPU sandbox
- 5 GB storage
Plus
- DeepSeek V4 Flash and V4 Pro models
- Entry-level T4 GPU (16 GB VRAM, 16 GB sandbox RAM)
- 30 GB storage
Pro
- All open-source models (such as Kimi, Qwen, and GLM)
- L4 and A10 GPUs (up to 24 GB VRAM, 32 GB sandbox RAM)
- 100 GB storage
Max (10x)
- All models, including Claude and GPT
- A100, H100, and H200 GPUs (up to 141 GB VRAM, up to 128 GB sandbox RAM)
- 512 GB storage
Max (30x)
- All models, including Claude and GPT
- A100, H100, and H200 GPUs (up to 141 GB VRAM, up to 128 GB sandbox RAM)
- 2 TB storage
* Pricing information is for reference only. Please refer to the latest announcement on Clusy's official website.