How to enable real-time collaboration (RTC) in Marimo?

[palimanohar7-spec]

Hi Marimo team,

I'm exploring how to enable real-time collaboration (RTC) in Marimo notebooks in a multi-user environment.

My use case

I am running Marimo on a shared server (for example behind JupyterHub or a cloud deployment) where multiple users may want to open and edit the same notebook simultaneously.

Currently, when multiple users open the same notebook, Marimo shows the message:

Notebook already connected
another browser tab is already connected to the kernel
Take over session

which suggests that only one active client can control the kernel at a time.

Questions

1. Is there currently a way to enable real-time collaboration for a Marimo notebook?
2. Are there plans to support multiple users editing the same notebook session?
3. Is collaborative editing only available in the hosted platform?
4. If someone wanted to build collaboration support on top of Marimo, what architecture would you recommend?

Goal

Ideally, we would like to support:

• multiple users opening the same notebook
• synchronized edits between users
• shared kernel execution

similar to collaborative notebook environments.

Any guidance or roadmap information would be greatly appreciated.

Thanks for the great work on Marimo!

[mscolnick]

@palimanohar7-spec RTC is still under active development and not recommended to use just yet.

[jenaalsup]

is RTC still under active development? @mscolnick

[Doribelove]

For the September follow-up: current main still treats RTC as an experimental feature, not a documented production multi-user setup. The configuration defines an rtc_v2 experimental flag, and the deployment guide says RTC requires WebSockets and is disabled when the server uses SSE. That is consistent with @mscolnick's earlier advice here not to rely on it yet.

The “Take over session” prompt is session ownership, not synchronized multi-user editing. If you test RTC on a shared server, keep WebSocket support enabled and treat the result as an experiment; I would not promise shared-kernel behavior or a supported JupyterHub deployment based on the flag alone. Only the maintainers can give a release/roadmap commitment.

This is a source-and-docs status check rather than a hands-on multi-user deployment test. AI assistance was used to prepare the reply; I checked the current repository files and this thread before posting.