Last updated: August 26, 2026

TL;DR: All new notebooks will now persist only your source files and files under 1GB uploaded or created through the file browser.

Cap on file syncs

All files larger than 1GB are not be saved on teardown. You can still download the files into the notebook and use them in molab, however, they will not be persisted on notebook shutdown. The majority of the files we seen at this size are downloadable artifacts such as HuggingFace models and datasets. If you have a use case that requires storing files outside of this limit please email us.

What persists

To keep your workspaces fast and prevent temporary build artifacts, model caches, and scratch files from cluttering your persistent data, only the following survive a session:

• Your code: notebook.py and other source files, at any depth, so a package layout keeps working
• Files you upload through the file browser: wherever you put them. molab records the exact path and brings the file back next session
• Your `.env` file: see below
• Marimo’s persistent cache files: data cached with mo.persistent_cache, subject to limits that may change at molab’s discretion (including but not limited to limits on file sizes, persistence duration, and so on).

Everything else in the workspace is treated as scratch and is cleared when the notebook shuts down.

Dragging a file into the file browser is now the most direct way to attach data to a notebook.

Secrets stay with your notebook

Your `.env` file will also persist between sessions. Use the Secrets panel in the notebook sidebar. marimo writes the file and loads it automatically. The .env file is not forked so you can share a notebook and its data without sharing your credentials.

Note that the file has to be named exactly `.env`. A file like `secrets.env` looks similar but is treated as ordinary data, forks included.

Bring your own remote storage

If your storage needs exceed our limits, or if you need to access the same data across multiple notebooks, we recommend connecting your notebooks to your own remote cloud storage or Google Drive.

This is also often the better option regardless of limits. Reading a dataset over fsspec fetches only the columns and rows your query touches, so a dataset far larger than any notebook workspace never needs to be copied in at all. For results worth keeping, writing to a repository you own gives you version history that a notebook workspace cannot.

What this means for your notebooks

Any notebook created before August 26 2026 keeps working exactly as it does today. Notebooks created after will be subject to the policy described above.

This policy allows molab to remain fast, stable, and reliable as your workloads grow. If you have edge cases or need higher capacity for specialized projects please reach out to us!