[Newbie Feedback] First glance at marimo

[dentroai]

Great:

• youtube guides
• single python file
• can change column width easily
• dark mode
• AI integration
• scratchpad!
• detects changes made in VSCode
• lightning fast answers on github

Could be better:

• strange that auto-save isn't on by default
• scratchpad different for each file. Often I work on many files at the same time, and want to have an overall scratchpad for all files. Will keep using an .md file for now
• AI integration doesn't rival the likes of cursor obviously. 2 improvements would be great tho: 1. allow editing cells inline. Right now the LLM output can only be added as a new file, same as with Cursor and working with Jupyter Notebooks. 2. If the knowledge base of marimo isn't used under the hood, that would be a great opportunity. The chatbot seems to now about marimo, so perhaps you already apply a RAG approach with your docs, discord messages and github issues
• when using jupyter Notebooks in VScode forks, I can set breakpoints in juptyer cells and debug them. Something for marimo?

Pretty impressive git history btw, perhaps take a week off c: :

[mscolnick]

strange that auto-save isn't on by default

it is the default. not sure why you didn't see this.

scratchpad different for each file. Often I work on many files at the same time, and want to have an overall scratchpad for all files. Will keep using an .md file for now

do you use a scratchpad in other applications? what do they do? it would be kinda frustrating looking at one notebook; then you write in the scratch page; go to another; write in that scratch pad; go back and now its overwritten.

it's valuable feedback, but we may not change it at the moment.

AI integration doesn't rival the likes of cursor obviously. 2 improvements would be great tho: 1. allow editing cells inline. Right now the LLM output can only be added as a new file, same as with Cursor and working with Jupyter Notebooks. 2. If the knowledge base of marimo isn't used under the hood, that would be a great opportunity. The chatbot seems to now about marimo, so perhaps you already apply a RAG approach with your docs, discord messages and github issues

1. we do have inline cell edits. it is currently under the Labs section of the user config. under experimental. it works, but can be improved before making GA
2. Not yet, but we can improve this. I will create a ticket since this is actionable. Thanks for the feedback. [ai] Include additional marimo knowledge base in the LLM context #4151

when using jupyter Notebooks in VScode forks, I can set breakpoints in juptyer cells and debug them. Something for marimo?

We don't support breakpoints by UI, but you can use pdb:

import pdb; pdb.set_trace()
# or
mo.pdb.set_trace()

Lots more to be desired here, but in the works

[mscolnick]

If the knowledge base of marimo isn't used under the hood, that would be a great opportunity

Also you can specify custom rules for the LLM if you need to beyond just the marimo knowledge base.

[HHammond]

strange that auto-save isn't on by default

This one has bitten me a few times.

If you don't give the notebook a name, when you close or refresh the page all you progress is lost without a confirmation dialogue.

I think adding a confirmation dialogue would make things a lot better.

Here's a screen recording of what the experience looks like:

Screen.Recording.2025-03-18.at.2.33.09.PM.mov

I lost a some work to this one already.

[dentroai]

Actually I do have autosave enabled, but it doesn't seem to save anything.

I need to actively press the disquette icon to save my work.

Also, even thoulgh auto_run_ is enable ("on cell change: autorun"), it does not run the cell when I unfocus it. Maybe I misunderstand something.

[mscolnick]

Cells are not run when un-focused. You can reference the docs for the various runtime configuration: https://docs.marimo.io/guides/configuration/runtime_configuration/

[mscolnick]

@HHammond, thanks for calling that out. We used to have it, but i mightve deleted it by accident at some point: #4154

[mscolnick]

Going to close, feedback should be capture in individual tickets or addressed. @dentroai feel free to followup with individual tickets (ideally actionable). If you have general questions or feedback our discord is a great resource.