Thanks for sharing this new tool! Previously I was using Claude Code with Marimo via marimo edit --watch command line option, which mostly works pretty well for a similar use case but I had to take care to avoid overwriting changes either myself or from the Agent. The current design mitigate this nicely.
There are also clear advantages to the agent having direct access to in-memory kernel, but one downside at the moment is that Agents can't figure out how to save changes made to a notebook. I have to do this myself in the UI then tell the Agent to do further processing. This may well be intentional and reflect me not following the pair programming paradigm, but if it's possible then giving access to a save_file API endpoint or similar would be useful.
Thanks for sharing this new tool! Previously I was using Claude Code with Marimo via
marimo edit --watchcommand line option, which mostly works pretty well for a similar use case but I had to take care to avoid overwriting changes either myself or from the Agent. The current design mitigate this nicely.There are also clear advantages to the agent having direct access to in-memory kernel, but one downside at the moment is that Agents can't figure out how to save changes made to a notebook. I have to do this myself in the UI then tell the Agent to do further processing. This may well be intentional and reflect me not following the pair programming paradigm, but if it's possible then giving access to a save_file API endpoint or similar would be useful.