Marimo Notebooks

Marimo is a modern, reactive Python notebook that replaces Jupyter. Cells re-execute automatically when dependencies change, notebooks are stored as clean .py files (not JSON), and there is no hidden state from out-of-order execution.

Carolina Cloud offers two Marimo flavors:

• Marimo (marimo) — Clean notebook environment. Install what you need.
• Data Science Notebook (datasciencemarimo) — Marimo pre-loaded with polars, pandas, scikit-learn, PyTorch, matplotlib, plotly, and 20+ data science packages. GPU-accelerated libraries (RAPIDS, cuDF) included when you add a GPU.

Access

Marimo notebooks are accessed through your web browser at https://<instance-name>.carolinacloud.io. The access token is appended to the URL automatically when you click Open in the dashboard.

Every Marimo instance also supports SSH for terminal work, package installation, and IDE-based remote development.

S3 integration

Marimo supports S3 pre-fill at creation time. Pass your S3 credentials and a bucket path, and the data will be downloaded into ~/s3-prefill/ when the instance starts. The aws CLI is pre-configured for subsequent aws s3 commands. See S3 Integration for details.

AI integration

Marimo’s AI autocomplete, edit, and chat panels are wired up automatically when you pass an Anthropic API key at creation time. The Claude Code CLI is also pre-installed for use from the Marimo terminal panel. See AI Integration for details.

GPU support

The Data Science Notebook flavor supports GPU attachment. When provisioned with a GPU, PyTorch CUDA and RAPIDS are available directly in notebook cells:

import torch
print(torch.cuda.get_device_name(0))

import cudf
df = cudf.DataFrame({"a": [1, 2, 3]})

Creating a Marimo notebook

# Basic notebook
ccloud new container --cpus 4 --ram 8 --disk 50 --flavor marimo

# Data science notebook with GPU
ccloud new container --cpus 16 --ram 64 --disk 200 \
  --flavor datasciencemarimo --name analysis

For full provisioning details (dashboard, API, CLI), see the Marimo Quickstart.

Auto-stop on idle

Notebook servers are the ideal case for auto-stop — enable it and Marimo will shut itself off after a configurable window of idle CPU. Your notebooks and virtualenv stay on disk; restart any time to pick up where you left off.

Learn more

• Marimo documentation